1 // -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
2 // vim: ts=8 sw=2 smarttab
4 * Ceph - scalable distributed file system
6 * Copyright (C) 2004-2006 Sage Weil <sage@newdream.net>
7 * Copyright (C) 2013,2014 Cloudwatt <libre.licensing@cloudwatt.com>
8 * Copyright (C) 2014 Red Hat <contact@redhat.com>
10 * Author: Loic Dachary <loic@dachary.org>
12 * This is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU Lesser General Public
14 * License version 2.1, as published by the Free Software
15 * Foundation. See file COPYING.
20 #include <boost/algorithm/string.hpp>
24 #include "mon/OSDMonitor.h"
25 #include "mon/Monitor.h"
26 #include "mon/MDSMonitor.h"
27 #include "mon/PGMonitor.h"
28 #include "mon/MgrStatMonitor.h"
29 #include "mon/AuthMonitor.h"
30 #include "mon/ConfigKeyService.h"
32 #include "mon/MonitorDBStore.h"
33 #include "mon/Session.h"
35 #include "crush/CrushWrapper.h"
36 #include "crush/CrushTester.h"
37 #include "crush/CrushTreeDumper.h"
39 #include "messages/MOSDBeacon.h"
40 #include "messages/MOSDFailure.h"
41 #include "messages/MOSDMarkMeDown.h"
42 #include "messages/MOSDFull.h"
43 #include "messages/MOSDMap.h"
44 #include "messages/MMonGetOSDMap.h"
45 #include "messages/MOSDBoot.h"
46 #include "messages/MOSDAlive.h"
47 #include "messages/MPoolOp.h"
48 #include "messages/MPoolOpReply.h"
49 #include "messages/MOSDPGCreate.h"
50 #include "messages/MOSDPGCreated.h"
51 #include "messages/MOSDPGTemp.h"
52 #include "messages/MMonCommand.h"
53 #include "messages/MRemoveSnaps.h"
54 #include "messages/MOSDScrub.h"
55 #include "messages/MRoute.h"
57 #include "common/TextTable.h"
58 #include "common/Timer.h"
59 #include "common/ceph_argparse.h"
60 #include "common/perf_counters.h"
61 #include "common/strtol.h"
63 #include "common/config.h"
64 #include "common/errno.h"
66 #include "erasure-code/ErasureCodePlugin.h"
67 #include "compressor/Compressor.h"
68 #include "common/Checksummer.h"
70 #include "include/compat.h"
71 #include "include/assert.h"
72 #include "include/stringify.h"
73 #include "include/util.h"
74 #include "common/cmdparse.h"
75 #include "include/str_list.h"
76 #include "include/str_map.h"
77 #include "include/scope_guard.h"
79 #include "json_spirit/json_spirit_reader.h"
81 #include <boost/algorithm/string/predicate.hpp>
83 #define dout_subsys ceph_subsys_mon
84 static const string OSD_PG_CREATING_PREFIX("osd_pg_creating");
85 static const string OSD_METADATA_PREFIX("osd_metadata");
89 const uint32_t MAX_POOL_APPLICATIONS = 4;
90 const uint32_t MAX_POOL_APPLICATION_KEYS = 64;
91 const uint32_t MAX_POOL_APPLICATION_LENGTH = 128;
93 } // anonymous namespace
95 void LastEpochClean::Lec::report(ps_t ps, epoch_t last_epoch_clean)
97 if (epoch_by_pg.size() <= ps) {
98 epoch_by_pg.resize(ps + 1, 0);
100 const auto old_lec = epoch_by_pg[ps];
101 if (old_lec >= last_epoch_clean) {
105 epoch_by_pg[ps] = last_epoch_clean;
106 if (last_epoch_clean < floor) {
107 floor = last_epoch_clean;
108 } else if (last_epoch_clean > floor) {
109 if (old_lec == floor) {
110 // probably should increase floor?
111 auto new_floor = std::min_element(std::begin(epoch_by_pg),
112 std::end(epoch_by_pg));
116 if (ps != next_missing) {
119 for (; next_missing < epoch_by_pg.size(); next_missing++) {
120 if (epoch_by_pg[next_missing] == 0) {
126 void LastEpochClean::remove_pool(uint64_t pool)
128 report_by_pool.erase(pool);
131 void LastEpochClean::report(const pg_t& pg, epoch_t last_epoch_clean)
133 auto& lec = report_by_pool[pg.pool()];
134 return lec.report(pg.ps(), last_epoch_clean);
137 epoch_t LastEpochClean::get_lower_bound(const OSDMap& latest) const
139 auto floor = latest.get_epoch();
140 for (auto& pool : latest.get_pools()) {
141 auto reported = report_by_pool.find(pool.first);
142 if (reported == report_by_pool.end()) {
145 if (reported->second.next_missing < pool.second.get_pg_num()) {
148 if (reported->second.floor < floor) {
149 floor = reported->second.floor;
156 struct C_UpdateCreatingPGs : public Context {
160 C_UpdateCreatingPGs(OSDMonitor *osdmon, epoch_t e) :
161 osdmon(osdmon), start(ceph_clock_now()), epoch(e) {}
162 void finish(int r) override {
164 utime_t end = ceph_clock_now();
165 dout(10) << "osdmap epoch " << epoch << " mapping took "
166 << (end - start) << " seconds" << dendl;
167 osdmon->update_creating_pgs();
168 osdmon->check_pg_creates_subs();
174 #define dout_prefix _prefix(_dout, mon, osdmap)
175 static ostream& _prefix(std::ostream *_dout, Monitor *mon, const OSDMap& osdmap) {
176 return *_dout << "mon." << mon->name << "@" << mon->rank
177 << "(" << mon->get_state_name()
178 << ").osd e" << osdmap.get_epoch() << " ";
181 OSDMonitor::OSDMonitor(
185 const string& service_name)
186 : PaxosService(mn, p, service_name),
188 inc_osd_cache(g_conf->mon_osd_cache_size),
189 full_osd_cache(g_conf->mon_osd_cache_size),
190 last_attempted_minwait_time(utime_t()),
191 mapper(mn->cct, &mn->cpu_tp),
192 op_tracker(cct, true, 1)
195 bool OSDMonitor::_have_pending_crush()
197 return pending_inc.crush.length() > 0;
200 CrushWrapper &OSDMonitor::_get_stable_crush()
202 return *osdmap.crush;
205 void OSDMonitor::_get_pending_crush(CrushWrapper& newcrush)
208 if (pending_inc.crush.length())
209 bl = pending_inc.crush;
211 osdmap.crush->encode(bl, CEPH_FEATURES_SUPPORTED_DEFAULT);
213 bufferlist::iterator p = bl.begin();
217 void OSDMonitor::create_initial()
219 dout(10) << "create_initial for " << mon->monmap->fsid << dendl;
224 mon->store->get("mkfs", "osdmap", bl);
228 newmap.set_fsid(mon->monmap->fsid);
230 newmap.build_simple(g_ceph_context, 0, mon->monmap->fsid, 0);
233 newmap.created = newmap.modified = ceph_clock_now();
235 // new clusters should sort bitwise by default.
236 newmap.set_flag(CEPH_OSDMAP_SORTBITWISE);
238 // new cluster should require latest by default
239 if (g_conf->mon_debug_no_require_luminous) {
240 newmap.require_osd_release = CEPH_RELEASE_KRAKEN;
241 derr << __func__ << " mon_debug_no_require_luminous=true" << dendl;
243 newmap.require_osd_release = CEPH_RELEASE_LUMINOUS;
245 CEPH_OSDMAP_RECOVERY_DELETES |
246 CEPH_OSDMAP_PURGED_SNAPDIRS;
247 newmap.full_ratio = g_conf->mon_osd_full_ratio;
248 if (newmap.full_ratio > 1.0) newmap.full_ratio /= 100;
249 newmap.backfillfull_ratio = g_conf->mon_osd_backfillfull_ratio;
250 if (newmap.backfillfull_ratio > 1.0) newmap.backfillfull_ratio /= 100;
251 newmap.nearfull_ratio = g_conf->mon_osd_nearfull_ratio;
252 if (newmap.nearfull_ratio > 1.0) newmap.nearfull_ratio /= 100;
253 int r = ceph_release_from_name(
254 g_conf->mon_osd_initial_require_min_compat_client.c_str());
256 assert(0 == "mon_osd_initial_require_min_compat_client is not valid");
258 newmap.require_min_compat_client = r;
261 // encode into pending incremental
262 newmap.encode(pending_inc.fullmap,
263 mon->get_quorum_con_features() | CEPH_FEATURE_RESERVED);
264 pending_inc.full_crc = newmap.get_crc();
265 dout(20) << " full crc " << pending_inc.full_crc << dendl;
268 void OSDMonitor::get_store_prefixes(std::set<string>& s)
270 s.insert(service_name);
271 s.insert(OSD_PG_CREATING_PREFIX);
272 s.insert(OSD_METADATA_PREFIX);
275 void OSDMonitor::update_from_paxos(bool *need_bootstrap)
277 version_t version = get_last_committed();
278 if (version == osdmap.epoch)
280 assert(version > osdmap.epoch);
282 dout(15) << "update_from_paxos paxos e " << version
283 << ", my e " << osdmap.epoch << dendl;
286 if (!mapping_job->is_done()) {
287 dout(1) << __func__ << " mapping job "
288 << mapping_job.get() << " did not complete, "
289 << mapping_job->shards << " left, canceling" << dendl;
290 mapping_job->abort();
298 * We will possibly have a stashed latest that *we* wrote, and we will
299 * always be sure to have the oldest full map in the first..last range
300 * due to encode_trim_extra(), which includes the oldest full map in the trim
303 * encode_trim_extra() does not however write the full map's
304 * version to 'full_latest'. This is only done when we are building the
305 * full maps from the incremental versions. But don't panic! We make sure
306 * that the following conditions find whichever full map version is newer.
308 version_t latest_full = get_version_latest_full();
309 if (latest_full == 0 && get_first_committed() > 1)
310 latest_full = get_first_committed();
312 if (get_first_committed() > 1 &&
313 latest_full < get_first_committed()) {
314 // the monitor could be just sync'ed with its peer, and the latest_full key
315 // is not encoded in the paxos commits in encode_pending(), so we need to
316 // make sure we get it pointing to a proper version.
317 version_t lc = get_last_committed();
318 version_t fc = get_first_committed();
320 dout(10) << __func__ << " looking for valid full map in interval"
321 << " [" << fc << ", " << lc << "]" << dendl;
324 for (version_t v = lc; v >= fc; v--) {
325 string full_key = "full_" + stringify(v);
326 if (mon->store->exists(get_service_name(), full_key)) {
327 dout(10) << __func__ << " found latest full map v " << v << dendl;
333 assert(latest_full > 0);
334 auto t(std::make_shared<MonitorDBStore::Transaction>());
335 put_version_latest_full(t, latest_full);
336 mon->store->apply_transaction(t);
337 dout(10) << __func__ << " updated the on-disk full map version to "
338 << latest_full << dendl;
341 if ((latest_full > 0) && (latest_full > osdmap.epoch)) {
342 bufferlist latest_bl;
343 get_version_full(latest_full, latest_bl);
344 assert(latest_bl.length() != 0);
345 dout(7) << __func__ << " loading latest full map e" << latest_full << dendl;
346 osdmap.decode(latest_bl);
349 if (mon->monmap->get_required_features().contains_all(
350 ceph::features::mon::FEATURE_LUMINOUS)) {
352 if (!mon->store->get(OSD_PG_CREATING_PREFIX, "creating", bl)) {
354 std::lock_guard<std::mutex> l(creating_pgs_lock);
355 creating_pgs.decode(p);
356 dout(7) << __func__ << " loading creating_pgs last_scan_epoch "
357 << creating_pgs.last_scan_epoch
358 << " with " << creating_pgs.pgs.size() << " pgs" << dendl;
360 dout(1) << __func__ << " missing creating pgs; upgrade from post-kraken?"
365 // make sure we're using the right pg service.. remove me post-luminous!
366 if (osdmap.require_osd_release >= CEPH_RELEASE_LUMINOUS) {
367 dout(10) << __func__ << " pgservice is mgrstat" << dendl;
368 mon->pgservice = mon->mgrstatmon()->get_pg_stat_service();
370 dout(10) << __func__ << " pgservice is pg" << dendl;
371 mon->pgservice = mon->pgmon()->get_pg_stat_service();
374 // walk through incrementals
375 MonitorDBStore::TransactionRef t;
377 while (version > osdmap.epoch) {
379 int err = get_version(osdmap.epoch+1, inc_bl);
381 assert(inc_bl.length());
383 dout(7) << "update_from_paxos applying incremental " << osdmap.epoch+1
385 OSDMap::Incremental inc(inc_bl);
386 err = osdmap.apply_incremental(inc);
390 t.reset(new MonitorDBStore::Transaction);
392 // Write out the full map for all past epochs. Encode the full
393 // map with the same features as the incremental. If we don't
394 // know, use the quorum features. If we don't know those either,
395 // encode with all features.
396 uint64_t f = inc.encode_features;
398 f = mon->get_quorum_con_features();
402 osdmap.encode(full_bl, f | CEPH_FEATURE_RESERVED);
403 tx_size += full_bl.length();
405 bufferlist orig_full_bl;
406 get_version_full(osdmap.epoch, orig_full_bl);
407 if (orig_full_bl.length()) {
408 // the primary provided the full map
409 assert(inc.have_crc);
410 if (inc.full_crc != osdmap.crc) {
411 // This will happen if the mons were running mixed versions in
412 // the past or some other circumstance made the full encoded
413 // maps divergent. Reloading here will bring us back into
414 // sync with the primary for this and all future maps. OSDs
415 // will also be brought back into sync when they discover the
416 // crc mismatch and request a full map from a mon.
417 derr << __func__ << " full map CRC mismatch, resetting to canonical"
420 osdmap.decode(orig_full_bl);
423 assert(!inc.have_crc);
424 put_version_full(t, osdmap.epoch, full_bl);
426 put_version_latest_full(t, osdmap.epoch);
429 dout(1) << osdmap << dendl;
431 if (osdmap.epoch == 1) {
432 t->erase("mkfs", "osdmap");
435 // make sure we're using the right pg service.. remove me post-luminous!
436 if (osdmap.require_osd_release >= CEPH_RELEASE_LUMINOUS) {
437 dout(10) << __func__ << " pgservice is mgrstat" << dendl;
438 mon->pgservice = mon->mgrstatmon()->get_pg_stat_service();
440 dout(10) << __func__ << " pgservice is pg" << dendl;
441 mon->pgservice = mon->pgmon()->get_pg_stat_service();
444 if (tx_size > g_conf->mon_sync_max_payload_size*2) {
445 mon->store->apply_transaction(t);
446 t = MonitorDBStore::TransactionRef();
449 if (mon->monmap->get_required_features().contains_all(
450 ceph::features::mon::FEATURE_LUMINOUS)) {
451 for (const auto &osd_state : inc.new_state) {
452 if (osd_state.second & CEPH_OSD_UP) {
453 // could be marked up *or* down, but we're too lazy to check which
454 last_osd_report.erase(osd_state.first);
456 if (osd_state.second & CEPH_OSD_EXISTS) {
457 // could be created *or* destroyed, but we can safely drop it
458 osd_epochs.erase(osd_state.first);
465 mon->store->apply_transaction(t);
468 for (int o = 0; o < osdmap.get_max_osd(); o++) {
469 if (osdmap.is_out(o))
471 auto found = down_pending_out.find(o);
472 if (osdmap.is_down(o)) {
473 // populate down -> out map
474 if (found == down_pending_out.end()) {
475 dout(10) << " adding osd." << o << " to down_pending_out map" << dendl;
476 down_pending_out[o] = ceph_clock_now();
479 if (found != down_pending_out.end()) {
480 dout(10) << " removing osd." << o << " from down_pending_out map" << dendl;
481 down_pending_out.erase(found);
485 // XXX: need to trim MonSession connected with a osd whose id > max_osd?
487 if (mon->is_leader()) {
488 // kick pgmon, make sure it's seen the latest map
489 mon->pgmon()->check_osd_map(osdmap.epoch);
493 check_pg_creates_subs();
495 share_map_with_random_osd();
500 // make sure our feature bits reflect the latest map
501 update_msgr_features();
503 if (!mon->is_leader()) {
504 // will be called by on_active() on the leader, avoid doing so twice
509 void OSDMonitor::start_mapping()
511 // initiate mapping job
513 dout(10) << __func__ << " canceling previous mapping_job " << mapping_job.get()
515 mapping_job->abort();
517 if (!osdmap.get_pools().empty()) {
518 auto fin = new C_UpdateCreatingPGs(this, osdmap.get_epoch());
519 mapping_job = mapping.start_update(osdmap, mapper,
520 g_conf->mon_osd_mapping_pgs_per_chunk);
521 dout(10) << __func__ << " started mapping job " << mapping_job.get()
522 << " at " << fin->start << dendl;
523 mapping_job->set_finish_event(fin);
525 dout(10) << __func__ << " no pools, no mapping job" << dendl;
526 mapping_job = nullptr;
530 void OSDMonitor::update_msgr_features()
533 types.insert((int)entity_name_t::TYPE_OSD);
534 types.insert((int)entity_name_t::TYPE_CLIENT);
535 types.insert((int)entity_name_t::TYPE_MDS);
536 types.insert((int)entity_name_t::TYPE_MON);
537 for (set<int>::iterator q = types.begin(); q != types.end(); ++q) {
539 uint64_t features = osdmap.get_features(*q, &mask);
540 if ((mon->messenger->get_policy(*q).features_required & mask) != features) {
541 dout(0) << "crush map has features " << features << ", adjusting msgr requires" << dendl;
542 Messenger::Policy p = mon->messenger->get_policy(*q);
543 p.features_required = (p.features_required & ~mask) | features;
544 mon->messenger->set_policy(*q, p);
549 void OSDMonitor::on_active()
553 if (mon->is_leader()) {
554 mon->clog->debug() << "osdmap " << osdmap;
556 list<MonOpRequestRef> ls;
557 take_all_failures(ls);
558 while (!ls.empty()) {
559 MonOpRequestRef op = ls.front();
560 op->mark_osdmon_event(__func__);
568 void OSDMonitor::on_restart()
570 last_osd_report.clear();
573 void OSDMonitor::on_shutdown()
575 dout(10) << __func__ << dendl;
577 dout(10) << __func__ << " canceling previous mapping_job " << mapping_job.get()
579 mapping_job->abort();
582 // discard failure info, waiters
583 list<MonOpRequestRef> ls;
584 take_all_failures(ls);
588 void OSDMonitor::update_logger()
590 dout(10) << "update_logger" << dendl;
592 mon->cluster_logger->set(l_cluster_num_osd, osdmap.get_num_osds());
593 mon->cluster_logger->set(l_cluster_num_osd_up, osdmap.get_num_up_osds());
594 mon->cluster_logger->set(l_cluster_num_osd_in, osdmap.get_num_in_osds());
595 mon->cluster_logger->set(l_cluster_osd_epoch, osdmap.get_epoch());
598 void OSDMonitor::create_pending()
600 pending_inc = OSDMap::Incremental(osdmap.epoch+1);
601 pending_inc.fsid = mon->monmap->fsid;
603 dout(10) << "create_pending e " << pending_inc.epoch << dendl;
605 // clean up pg_temp, primary_temp
606 OSDMap::clean_temps(g_ceph_context, osdmap, &pending_inc);
607 dout(10) << "create_pending did clean_temps" << dendl;
609 // On upgrade OSDMap has new field set by mon_osd_backfillfull_ratio config
610 // instead of osd_backfill_full_ratio config
611 if (osdmap.backfillfull_ratio <= 0) {
612 pending_inc.new_backfillfull_ratio = g_conf->mon_osd_backfillfull_ratio;
613 if (pending_inc.new_backfillfull_ratio > 1.0)
614 pending_inc.new_backfillfull_ratio /= 100;
615 dout(1) << __func__ << " setting backfillfull_ratio = "
616 << pending_inc.new_backfillfull_ratio << dendl;
618 if (osdmap.get_epoch() > 0 &&
619 osdmap.require_osd_release < CEPH_RELEASE_LUMINOUS) {
620 // transition full ratios from PGMap to OSDMap (on upgrade)
621 float full_ratio = mon->pgservice->get_full_ratio();
622 float nearfull_ratio = mon->pgservice->get_nearfull_ratio();
623 if (osdmap.full_ratio != full_ratio) {
624 dout(10) << __func__ << " full_ratio " << osdmap.full_ratio
625 << " -> " << full_ratio << " (from pgmap)" << dendl;
626 pending_inc.new_full_ratio = full_ratio;
628 if (osdmap.nearfull_ratio != nearfull_ratio) {
629 dout(10) << __func__ << " nearfull_ratio " << osdmap.nearfull_ratio
630 << " -> " << nearfull_ratio << " (from pgmap)" << dendl;
631 pending_inc.new_nearfull_ratio = nearfull_ratio;
634 // safety check (this shouldn't really happen)
635 if (osdmap.full_ratio <= 0) {
636 pending_inc.new_full_ratio = g_conf->mon_osd_full_ratio;
637 if (pending_inc.new_full_ratio > 1.0)
638 pending_inc.new_full_ratio /= 100;
639 dout(1) << __func__ << " setting full_ratio = "
640 << pending_inc.new_full_ratio << dendl;
642 if (osdmap.nearfull_ratio <= 0) {
643 pending_inc.new_nearfull_ratio = g_conf->mon_osd_nearfull_ratio;
644 if (pending_inc.new_nearfull_ratio > 1.0)
645 pending_inc.new_nearfull_ratio /= 100;
646 dout(1) << __func__ << " setting nearfull_ratio = "
647 << pending_inc.new_nearfull_ratio << dendl;
651 // Rewrite CRUSH rule IDs if they are using legacy "ruleset"
653 if (osdmap.crush->has_legacy_rule_ids()) {
654 CrushWrapper newcrush;
655 _get_pending_crush(newcrush);
657 // First, for all pools, work out which rule they really used
658 // by resolving ruleset to rule.
659 for (const auto &i : osdmap.get_pools()) {
660 const auto pool_id = i.first;
661 const auto &pool = i.second;
662 int new_rule_id = newcrush.find_rule(pool.crush_rule,
663 pool.type, pool.size);
665 dout(1) << __func__ << " rewriting pool "
666 << osdmap.get_pool_name(pool_id) << " crush ruleset "
667 << pool.crush_rule << " -> rule id " << new_rule_id << dendl;
668 if (pending_inc.new_pools.count(pool_id) == 0) {
669 pending_inc.new_pools[pool_id] = pool;
671 pending_inc.new_pools[pool_id].crush_rule = new_rule_id;
674 // Now, go ahead and renumber all the rules so that their
675 // rule_id field corresponds to their position in the array
676 auto old_to_new = newcrush.renumber_rules();
677 dout(1) << __func__ << " Rewrote " << old_to_new << " crush IDs:" << dendl;
678 for (const auto &i : old_to_new) {
679 dout(1) << __func__ << " " << i.first << " -> " << i.second << dendl;
681 pending_inc.crush.clear();
682 newcrush.encode(pending_inc.crush, mon->get_quorum_con_features());
687 OSDMonitor::update_pending_pgs(const OSDMap::Incremental& inc)
689 dout(10) << __func__ << dendl;
690 creating_pgs_t pending_creatings;
692 std::lock_guard<std::mutex> l(creating_pgs_lock);
693 pending_creatings = creating_pgs;
695 // check for new or old pools
696 if (pending_creatings.last_scan_epoch < inc.epoch) {
697 if (osdmap.get_epoch() &&
698 osdmap.require_osd_release < CEPH_RELEASE_LUMINOUS) {
700 mon->pgservice->maybe_add_creating_pgs(creating_pgs.last_scan_epoch,
703 dout(7) << __func__ << " " << added << " pgs added from pgmap" << dendl;
706 queued += scan_for_creating_pgs(osdmap.get_pools(),
710 queued += scan_for_creating_pgs(inc.new_pools,
714 dout(10) << __func__ << " " << queued << " pools queued" << dendl;
715 for (auto deleted_pool : inc.old_pools) {
716 auto removed = pending_creatings.remove_pool(deleted_pool);
717 dout(10) << __func__ << " " << removed
718 << " pg removed because containing pool deleted: "
719 << deleted_pool << dendl;
720 last_epoch_clean.remove_pool(deleted_pool);
722 // pgmon updates its creating_pgs in check_osd_map() which is called by
723 // on_active() and check_osd_map() could be delayed if lease expires, so its
724 // creating_pgs could be stale in comparison with the one of osdmon. let's
725 // trim them here. otherwise, they will be added back after being erased.
726 unsigned removed = 0;
727 for (auto& pg : pending_created_pgs) {
728 dout(20) << __func__ << " noting created pg " << pg << dendl;
729 pending_creatings.created_pools.insert(pg.pool());
730 removed += pending_creatings.pgs.erase(pg);
732 pending_created_pgs.clear();
733 dout(10) << __func__ << " " << removed
734 << " pgs removed because they're created" << dendl;
735 pending_creatings.last_scan_epoch = osdmap.get_epoch();
739 unsigned max = MAX(1, g_conf->mon_osd_max_creating_pgs);
740 const auto total = pending_creatings.pgs.size();
741 while (pending_creatings.pgs.size() < max &&
742 !pending_creatings.queue.empty()) {
743 auto p = pending_creatings.queue.begin();
744 int64_t poolid = p->first;
745 dout(10) << __func__ << " pool " << poolid
746 << " created " << p->second.created
747 << " modified " << p->second.modified
748 << " [" << p->second.start << "-" << p->second.end << ")"
750 int n = MIN(max - pending_creatings.pgs.size(),
751 p->second.end - p->second.start);
752 ps_t first = p->second.start;
753 ps_t end = first + n;
754 for (ps_t ps = first; ps < end; ++ps) {
755 const pg_t pgid{ps, static_cast<uint64_t>(poolid)};
756 // NOTE: use the *current* epoch as the PG creation epoch so that the
757 // OSD does not have to generate a long set of PastIntervals.
758 pending_creatings.pgs.emplace(pgid, make_pair(inc.epoch,
759 p->second.modified));
760 dout(10) << __func__ << " adding " << pgid << dendl;
762 p->second.start = end;
763 if (p->second.done()) {
764 dout(10) << __func__ << " done with queue for " << poolid << dendl;
765 pending_creatings.queue.erase(p);
767 dout(10) << __func__ << " pool " << poolid
768 << " now [" << p->second.start << "-" << p->second.end << ")"
772 dout(10) << __func__ << " queue remaining: " << pending_creatings.queue.size()
773 << " pools" << dendl;
775 << " " << (pending_creatings.pgs.size() - total)
776 << "/" << pending_creatings.pgs.size()
777 << " pgs added from queued pools" << dendl;
778 return pending_creatings;
781 void OSDMonitor::maybe_prime_pg_temp()
784 if (pending_inc.crush.length()) {
785 dout(10) << __func__ << " new crush map, all" << dendl;
789 if (!pending_inc.new_up_client.empty()) {
790 dout(10) << __func__ << " new up osds, all" << dendl;
794 // check for interesting OSDs
796 for (auto p = pending_inc.new_state.begin();
797 !all && p != pending_inc.new_state.end();
799 if ((p->second & CEPH_OSD_UP) &&
800 osdmap.is_up(p->first)) {
801 osds.insert(p->first);
804 for (map<int32_t,uint32_t>::iterator p = pending_inc.new_weight.begin();
805 !all && p != pending_inc.new_weight.end();
807 if (p->second < osdmap.get_weight(p->first)) {
809 osds.insert(p->first);
811 dout(10) << __func__ << " osd." << p->first << " weight increase, all"
817 if (!all && osds.empty())
822 mapping.get_osd_acting_pgs(*osds.begin()).size() * osds.size();
823 if (estimate > mapping.get_num_pgs() *
824 g_conf->mon_osd_prime_pg_temp_max_estimate) {
825 dout(10) << __func__ << " estimate " << estimate << " pgs on "
826 << osds.size() << " osds >= "
827 << g_conf->mon_osd_prime_pg_temp_max_estimate << " of total "
828 << mapping.get_num_pgs() << " pgs, all"
832 dout(10) << __func__ << " estimate " << estimate << " pgs on "
833 << osds.size() << " osds" << dendl;
838 next.deepish_copy_from(osdmap);
839 next.apply_incremental(pending_inc);
841 if (next.get_pools().empty()) {
842 dout(10) << __func__ << " no pools, no pg_temp priming" << dendl;
844 PrimeTempJob job(next, this);
845 mapper.queue(&job, g_conf->mon_osd_mapping_pgs_per_chunk);
846 if (job.wait_for(g_conf->mon_osd_prime_pg_temp_max_time)) {
847 dout(10) << __func__ << " done in " << job.get_duration() << dendl;
849 dout(10) << __func__ << " did not finish in "
850 << g_conf->mon_osd_prime_pg_temp_max_time
851 << ", stopping" << dendl;
855 dout(10) << __func__ << " " << osds.size() << " interesting osds" << dendl;
856 utime_t stop = ceph_clock_now();
857 stop += g_conf->mon_osd_prime_pg_temp_max_time;
858 const int chunk = 1000;
860 std::unordered_set<pg_t> did_pgs;
861 for (auto osd : osds) {
862 auto& pgs = mapping.get_osd_acting_pgs(osd);
863 dout(20) << __func__ << " osd." << osd << " " << pgs << dendl;
864 for (auto pgid : pgs) {
865 if (!did_pgs.insert(pgid).second) {
868 prime_pg_temp(next, pgid);
871 if (ceph_clock_now() > stop) {
872 dout(10) << __func__ << " consumed more than "
873 << g_conf->mon_osd_prime_pg_temp_max_time
874 << " seconds, stopping"
884 void OSDMonitor::prime_pg_temp(
888 if (mon->monmap->get_required_features().contains_all(
889 ceph::features::mon::FEATURE_LUMINOUS)) {
890 // TODO: remove this creating_pgs direct access?
891 if (creating_pgs.pgs.count(pgid)) {
895 if (mon->pgservice->is_creating_pg(pgid)) {
899 if (!osdmap.pg_exists(pgid)) {
903 vector<int> up, acting;
904 mapping.get(pgid, &up, nullptr, &acting, nullptr);
906 vector<int> next_up, next_acting;
907 int next_up_primary, next_acting_primary;
908 next.pg_to_up_acting_osds(pgid, &next_up, &next_up_primary,
909 &next_acting, &next_acting_primary);
910 if (acting == next_acting && next_up != next_acting)
911 return; // no change since last epoch
914 return; // if previously empty now we can be no worse off
915 const pg_pool_t *pool = next.get_pg_pool(pgid.pool());
916 if (pool && acting.size() < pool->min_size)
917 return; // can be no worse off than before
919 if (next_up == next_acting) {
921 dout(20) << __func__ << "next_up === next_acting now, clear pg_temp"
925 dout(20) << __func__ << " " << pgid << " " << up << "/" << acting
926 << " -> " << next_up << "/" << next_acting
927 << ", priming " << acting
930 Mutex::Locker l(prime_pg_temp_lock);
931 // do not touch a mapping if a change is pending
932 pending_inc.new_pg_temp.emplace(
934 mempool::osdmap::vector<int>(acting.begin(), acting.end()));
939 * @note receiving a transaction in this function gives a fair amount of
940 * freedom to the service implementation if it does need it. It shouldn't.
942 void OSDMonitor::encode_pending(MonitorDBStore::TransactionRef t)
944 dout(10) << "encode_pending e " << pending_inc.epoch
947 // finalize up pending_inc
948 pending_inc.modified = ceph_clock_now();
950 int r = pending_inc.propagate_snaps_to_tiers(g_ceph_context, osdmap);
954 if (!mapping_job->is_done()) {
955 dout(1) << __func__ << " skipping prime_pg_temp; mapping job "
956 << mapping_job.get() << " did not complete, "
957 << mapping_job->shards << " left" << dendl;
958 mapping_job->abort();
959 } else if (mapping.get_epoch() < osdmap.get_epoch()) {
960 dout(1) << __func__ << " skipping prime_pg_temp; mapping job "
961 << mapping_job.get() << " is prior epoch "
962 << mapping.get_epoch() << dendl;
964 if (g_conf->mon_osd_prime_pg_temp) {
965 maybe_prime_pg_temp();
968 } else if (g_conf->mon_osd_prime_pg_temp) {
969 dout(1) << __func__ << " skipping prime_pg_temp; mapping job did not start"
974 // ensure we don't have blank new_state updates. these are interrpeted as
975 // CEPH_OSD_UP (and almost certainly not what we want!).
976 auto p = pending_inc.new_state.begin();
977 while (p != pending_inc.new_state.end()) {
978 if (p->second == 0) {
979 dout(10) << "new_state for osd." << p->first << " is 0, removing" << dendl;
980 p = pending_inc.new_state.erase(p);
990 tmp.deepish_copy_from(osdmap);
991 tmp.apply_incremental(pending_inc);
993 if (tmp.require_osd_release >= CEPH_RELEASE_LUMINOUS) {
994 // remove any legacy osdmap nearfull/full flags
996 if (tmp.test_flag(CEPH_OSDMAP_FULL | CEPH_OSDMAP_NEARFULL)) {
997 dout(10) << __func__ << " clearing legacy osdmap nearfull/full flag"
999 remove_flag(CEPH_OSDMAP_NEARFULL);
1000 remove_flag(CEPH_OSDMAP_FULL);
1003 // collect which pools are currently affected by
1004 // the near/backfill/full osd(s),
1005 // and set per-pool near/backfill/full flag instead
1006 set<int64_t> full_pool_ids;
1007 set<int64_t> backfillfull_pool_ids;
1008 set<int64_t> nearfull_pool_ids;
1009 tmp.get_full_pools(g_ceph_context,
1011 &backfillfull_pool_ids,
1012 &nearfull_pool_ids);
1013 if (full_pool_ids.empty() ||
1014 backfillfull_pool_ids.empty() ||
1015 nearfull_pool_ids.empty()) {
1016 // normal case - no nearfull, backfillfull or full osds
1017 // try cancel any improper nearfull/backfillfull/full pool
1019 for (auto &pool: tmp.get_pools()) {
1020 auto p = pool.first;
1021 if (tmp.get_pg_pool(p)->has_flag(pg_pool_t::FLAG_NEARFULL) &&
1022 nearfull_pool_ids.empty()) {
1023 dout(10) << __func__ << " clearing pool '" << tmp.pool_name[p]
1024 << "'s nearfull flag" << dendl;
1025 if (pending_inc.new_pools.count(p) == 0) {
1026 // load original pool info first!
1027 pending_inc.new_pools[p] = pool.second;
1029 pending_inc.new_pools[p].flags &= ~pg_pool_t::FLAG_NEARFULL;
1031 if (tmp.get_pg_pool(p)->has_flag(pg_pool_t::FLAG_BACKFILLFULL) &&
1032 backfillfull_pool_ids.empty()) {
1033 dout(10) << __func__ << " clearing pool '" << tmp.pool_name[p]
1034 << "'s backfillfull flag" << dendl;
1035 if (pending_inc.new_pools.count(p) == 0) {
1036 pending_inc.new_pools[p] = pool.second;
1038 pending_inc.new_pools[p].flags &= ~pg_pool_t::FLAG_BACKFILLFULL;
1040 if (tmp.get_pg_pool(p)->has_flag(pg_pool_t::FLAG_FULL) &&
1041 full_pool_ids.empty()) {
1042 if (tmp.get_pg_pool(p)->has_flag(pg_pool_t::FLAG_FULL_NO_QUOTA)) {
1043 // set by EQUOTA, skipping
1046 dout(10) << __func__ << " clearing pool '" << tmp.pool_name[p]
1047 << "'s full flag" << dendl;
1048 if (pending_inc.new_pools.count(p) == 0) {
1049 pending_inc.new_pools[p] = pool.second;
1051 pending_inc.new_pools[p].flags &= ~pg_pool_t::FLAG_FULL;
1055 if (!full_pool_ids.empty()) {
1056 dout(10) << __func__ << " marking pool(s) " << full_pool_ids
1057 << " as full" << dendl;
1058 for (auto &p: full_pool_ids) {
1059 if (tmp.get_pg_pool(p)->has_flag(pg_pool_t::FLAG_FULL)) {
1062 if (pending_inc.new_pools.count(p) == 0) {
1063 pending_inc.new_pools[p] = tmp.pools[p];
1065 pending_inc.new_pools[p].flags |= pg_pool_t::FLAG_FULL;
1066 pending_inc.new_pools[p].flags &= ~pg_pool_t::FLAG_BACKFILLFULL;
1067 pending_inc.new_pools[p].flags &= ~pg_pool_t::FLAG_NEARFULL;
1069 // cancel FLAG_FULL for pools which are no longer full too
1070 for (auto &pool: tmp.get_pools()) {
1071 auto p = pool.first;
1072 if (full_pool_ids.count(p)) {
1073 // skip pools we have just marked as full above
1076 if (!tmp.get_pg_pool(p)->has_flag(pg_pool_t::FLAG_FULL) ||
1077 tmp.get_pg_pool(p)->has_flag(pg_pool_t::FLAG_FULL_NO_QUOTA)) {
1078 // don't touch if currently is not full
1079 // or is running out of quota (and hence considered as full)
1082 dout(10) << __func__ << " clearing pool '" << tmp.pool_name[p]
1083 << "'s full flag" << dendl;
1084 if (pending_inc.new_pools.count(p) == 0) {
1085 pending_inc.new_pools[p] = pool.second;
1087 pending_inc.new_pools[p].flags &= ~pg_pool_t::FLAG_FULL;
1090 if (!backfillfull_pool_ids.empty()) {
1091 for (auto &p: backfillfull_pool_ids) {
1092 if (full_pool_ids.count(p)) {
1093 // skip pools we have already considered as full above
1096 if (tmp.get_pg_pool(p)->has_flag(pg_pool_t::FLAG_FULL_NO_QUOTA)) {
1097 // make sure FLAG_FULL is truly set, so we are safe not
1098 // to set a extra (redundant) FLAG_BACKFILLFULL flag
1099 assert(tmp.get_pg_pool(p)->has_flag(pg_pool_t::FLAG_FULL));
1102 if (tmp.get_pg_pool(p)->has_flag(pg_pool_t::FLAG_BACKFILLFULL)) {
1103 // don't bother if pool is already marked as backfillfull
1106 dout(10) << __func__ << " marking pool '" << tmp.pool_name[p]
1107 << "'s as backfillfull" << dendl;
1108 if (pending_inc.new_pools.count(p) == 0) {
1109 pending_inc.new_pools[p] = tmp.pools[p];
1111 pending_inc.new_pools[p].flags |= pg_pool_t::FLAG_BACKFILLFULL;
1112 pending_inc.new_pools[p].flags &= ~pg_pool_t::FLAG_NEARFULL;
1114 // cancel FLAG_BACKFILLFULL for pools
1115 // which are no longer backfillfull too
1116 for (auto &pool: tmp.get_pools()) {
1117 auto p = pool.first;
1118 if (full_pool_ids.count(p) || backfillfull_pool_ids.count(p)) {
1119 // skip pools we have just marked as backfillfull/full above
1122 if (!tmp.get_pg_pool(p)->has_flag(pg_pool_t::FLAG_BACKFILLFULL)) {
1123 // and don't touch if currently is not backfillfull
1126 dout(10) << __func__ << " clearing pool '" << tmp.pool_name[p]
1127 << "'s backfillfull flag" << dendl;
1128 if (pending_inc.new_pools.count(p) == 0) {
1129 pending_inc.new_pools[p] = pool.second;
1131 pending_inc.new_pools[p].flags &= ~pg_pool_t::FLAG_BACKFILLFULL;
1134 if (!nearfull_pool_ids.empty()) {
1135 for (auto &p: nearfull_pool_ids) {
1136 if (full_pool_ids.count(p) || backfillfull_pool_ids.count(p)) {
1139 if (tmp.get_pg_pool(p)->has_flag(pg_pool_t::FLAG_FULL_NO_QUOTA)) {
1140 // make sure FLAG_FULL is truly set, so we are safe not
1141 // to set a extra (redundant) FLAG_NEARFULL flag
1142 assert(tmp.get_pg_pool(p)->has_flag(pg_pool_t::FLAG_FULL));
1145 if (tmp.get_pg_pool(p)->has_flag(pg_pool_t::FLAG_NEARFULL)) {
1146 // don't bother if pool is already marked as nearfull
1149 dout(10) << __func__ << " marking pool '" << tmp.pool_name[p]
1150 << "'s as nearfull" << dendl;
1151 if (pending_inc.new_pools.count(p) == 0) {
1152 pending_inc.new_pools[p] = tmp.pools[p];
1154 pending_inc.new_pools[p].flags |= pg_pool_t::FLAG_NEARFULL;
1156 // cancel FLAG_NEARFULL for pools
1157 // which are no longer nearfull too
1158 for (auto &pool: tmp.get_pools()) {
1159 auto p = pool.first;
1160 if (full_pool_ids.count(p) ||
1161 backfillfull_pool_ids.count(p) ||
1162 nearfull_pool_ids.count(p)) {
1163 // skip pools we have just marked as
1164 // nearfull/backfillfull/full above
1167 if (!tmp.get_pg_pool(p)->has_flag(pg_pool_t::FLAG_NEARFULL)) {
1168 // and don't touch if currently is not nearfull
1171 dout(10) << __func__ << " clearing pool '" << tmp.pool_name[p]
1172 << "'s nearfull flag" << dendl;
1173 if (pending_inc.new_pools.count(p) == 0) {
1174 pending_inc.new_pools[p] = pool.second;
1176 pending_inc.new_pools[p].flags &= ~pg_pool_t::FLAG_NEARFULL;
1180 // min_compat_client?
1181 if (tmp.require_min_compat_client == 0) {
1182 auto mv = tmp.get_min_compat_client();
1183 dout(1) << __func__ << " setting require_min_compat_client to currently "
1184 << "required " << ceph_release_name(mv) << dendl;
1185 mon->clog->info() << "setting require_min_compat_client to currently "
1186 << "required " << ceph_release_name(mv);
1187 pending_inc.new_require_min_compat_client = mv;
1190 if (osdmap.require_osd_release < CEPH_RELEASE_LUMINOUS) {
1191 // convert ec profile ruleset-* -> crush-*
1192 for (auto& p : tmp.erasure_code_profiles) {
1193 bool changed = false;
1194 map<string,string> newprofile;
1195 for (auto& q : p.second) {
1196 if (q.first.find("ruleset-") == 0) {
1197 string key = "crush-";
1198 key += q.first.substr(8);
1199 newprofile[key] = q.second;
1201 dout(20) << " updating ec profile " << p.first
1202 << " key " << q.first << " -> " << key << dendl;
1204 newprofile[q.first] = q.second;
1208 dout(10) << " updated ec profile " << p.first << ": "
1209 << newprofile << dendl;
1210 pending_inc.new_erasure_code_profiles[p.first] = newprofile;
1214 // auto-enable pool applications upon upgrade
1215 // NOTE: this can be removed post-Luminous assuming upgrades need to
1216 // proceed through Luminous
1217 for (auto &pool_pair : tmp.pools) {
1218 int64_t pool_id = pool_pair.first;
1219 pg_pool_t pg_pool = pool_pair.second;
1220 if (pg_pool.is_tier()) {
1224 std::string pool_name = tmp.get_pool_name(pool_id);
1225 uint32_t match_count = 0;
1228 FSMap const &pending_fsmap = mon->mdsmon()->get_pending();
1229 if (pending_fsmap.pool_in_use(pool_id)) {
1230 dout(10) << __func__ << " auto-enabling CephFS on pool '"
1231 << pool_name << "'" << dendl;
1232 pg_pool.application_metadata.insert(
1233 {pg_pool_t::APPLICATION_NAME_CEPHFS, {}});
1237 // RBD heuristics (default OpenStack pool names from docs and
1239 if (boost::algorithm::contains(pool_name, "rbd") ||
1240 pool_name == "images" || pool_name == "volumes" ||
1241 pool_name == "backups" || pool_name == "vms") {
1242 dout(10) << __func__ << " auto-enabling RBD on pool '"
1243 << pool_name << "'" << dendl;
1244 pg_pool.application_metadata.insert(
1245 {pg_pool_t::APPLICATION_NAME_RBD, {}});
1250 if (boost::algorithm::contains(pool_name, ".rgw") ||
1251 boost::algorithm::contains(pool_name, ".log") ||
1252 boost::algorithm::contains(pool_name, ".intent-log") ||
1253 boost::algorithm::contains(pool_name, ".usage") ||
1254 boost::algorithm::contains(pool_name, ".users")) {
1255 dout(10) << __func__ << " auto-enabling RGW on pool '"
1256 << pool_name << "'" << dendl;
1257 pg_pool.application_metadata.insert(
1258 {pg_pool_t::APPLICATION_NAME_RGW, {}});
1262 // OpenStack gnocchi (from ceph-ansible)
1263 if (pool_name == "metrics" && match_count == 0) {
1264 dout(10) << __func__ << " auto-enabling OpenStack Gnocchi on pool '"
1265 << pool_name << "'" << dendl;
1266 pg_pool.application_metadata.insert({"openstack_gnocchi", {}});
1270 if (match_count == 1) {
1271 pg_pool.last_change = pending_inc.epoch;
1272 pending_inc.new_pools[pool_id] = pg_pool;
1273 } else if (match_count > 1) {
1274 auto pstat = mon->pgservice->get_pool_stat(pool_id);
1275 if (pstat != nullptr && pstat->stats.sum.num_objects > 0) {
1276 mon->clog->info() << "unable to auto-enable application for pool "
1277 << "'" << pool_name << "'";
1286 for (auto i = pending_inc.new_state.begin();
1287 i != pending_inc.new_state.end();
1289 int s = i->second ? i->second : CEPH_OSD_UP;
1290 if (s & CEPH_OSD_UP)
1291 dout(2) << " osd." << i->first << " DOWN" << dendl;
1292 if (s & CEPH_OSD_EXISTS)
1293 dout(2) << " osd." << i->first << " DNE" << dendl;
1295 for (map<int32_t,entity_addr_t>::iterator i = pending_inc.new_up_client.begin();
1296 i != pending_inc.new_up_client.end();
1298 //FIXME: insert cluster addresses too
1299 dout(2) << " osd." << i->first << " UP " << i->second << dendl;
1301 for (map<int32_t,uint32_t>::iterator i = pending_inc.new_weight.begin();
1302 i != pending_inc.new_weight.end();
1304 if (i->second == CEPH_OSD_OUT) {
1305 dout(2) << " osd." << i->first << " OUT" << dendl;
1306 } else if (i->second == CEPH_OSD_IN) {
1307 dout(2) << " osd." << i->first << " IN" << dendl;
1309 dout(2) << " osd." << i->first << " WEIGHT " << hex << i->second << dec << dendl;
1313 // features for osdmap and its incremental
1314 uint64_t features = mon->get_quorum_con_features();
1316 // encode full map and determine its crc
1319 tmp.deepish_copy_from(osdmap);
1320 tmp.apply_incremental(pending_inc);
1322 // determine appropriate features
1323 if (tmp.require_osd_release < CEPH_RELEASE_LUMINOUS) {
1324 dout(10) << __func__ << " encoding without feature SERVER_LUMINOUS"
1326 features &= ~CEPH_FEATURE_SERVER_LUMINOUS;
1328 if (tmp.require_osd_release < CEPH_RELEASE_KRAKEN) {
1329 dout(10) << __func__ << " encoding without feature SERVER_KRAKEN | "
1330 << "MSG_ADDR2" << dendl;
1331 features &= ~(CEPH_FEATURE_SERVER_KRAKEN |
1332 CEPH_FEATURE_MSG_ADDR2);
1334 if (tmp.require_osd_release < CEPH_RELEASE_JEWEL) {
1335 dout(10) << __func__ << " encoding without feature SERVER_JEWEL" << dendl;
1336 features &= ~CEPH_FEATURE_SERVER_JEWEL;
1338 dout(10) << __func__ << " encoding full map with " << features << dendl;
1341 ::encode(tmp, fullbl, features | CEPH_FEATURE_RESERVED);
1342 pending_inc.full_crc = tmp.get_crc();
1344 // include full map in the txn. note that old monitors will
1345 // overwrite this. new ones will now skip the local full map
1346 // encode and reload from this.
1347 put_version_full(t, pending_inc.epoch, fullbl);
1351 assert(get_last_committed() + 1 == pending_inc.epoch);
1352 ::encode(pending_inc, bl, features | CEPH_FEATURE_RESERVED);
1354 dout(20) << " full_crc " << tmp.get_crc()
1355 << " inc_crc " << pending_inc.inc_crc << dendl;
1357 /* put everything in the transaction */
1358 put_version(t, pending_inc.epoch, bl);
1359 put_last_committed(t, pending_inc.epoch);
1362 for (map<int,bufferlist>::iterator p = pending_metadata.begin();
1363 p != pending_metadata.end();
1365 t->put(OSD_METADATA_PREFIX, stringify(p->first), p->second);
1366 for (set<int>::iterator p = pending_metadata_rm.begin();
1367 p != pending_metadata_rm.end();
1369 t->erase(OSD_METADATA_PREFIX, stringify(*p));
1370 pending_metadata.clear();
1371 pending_metadata_rm.clear();
1373 // and pg creating, also!
1374 if (mon->monmap->get_required_features().contains_all(
1375 ceph::features::mon::FEATURE_LUMINOUS)) {
1376 auto pending_creatings = update_pending_pgs(pending_inc);
1377 if (osdmap.get_epoch() &&
1378 osdmap.require_osd_release < CEPH_RELEASE_LUMINOUS) {
1379 dout(7) << __func__ << " in the middle of upgrading, "
1380 << " trimming pending creating_pgs using pgmap" << dendl;
1381 mon->pgservice->maybe_trim_creating_pgs(&pending_creatings);
1383 bufferlist creatings_bl;
1384 ::encode(pending_creatings, creatings_bl);
1385 t->put(OSD_PG_CREATING_PREFIX, "creating", creatings_bl);
1389 health_check_map_t next;
1390 tmp.check_health(&next);
1391 encode_health(next, t);
1394 void OSDMonitor::trim_creating_pgs(creating_pgs_t* creating_pgs,
1395 const ceph::unordered_map<pg_t,pg_stat_t>& pg_stat)
1397 auto p = creating_pgs->pgs.begin();
1398 while (p != creating_pgs->pgs.end()) {
1399 auto q = pg_stat.find(p->first);
1400 if (q != pg_stat.end() &&
1401 !(q->second.state & PG_STATE_CREATING)) {
1402 dout(20) << __func__ << " pgmap shows " << p->first << " is created"
1404 p = creating_pgs->pgs.erase(p);
1411 int OSDMonitor::load_metadata(int osd, map<string, string>& m, ostream *err)
1414 int r = mon->store->get(OSD_METADATA_PREFIX, stringify(osd), bl);
1418 bufferlist::iterator p = bl.begin();
1421 catch (buffer::error& e) {
1423 *err << "osd." << osd << " metadata is corrupt";
1429 void OSDMonitor::count_metadata(const string& field, map<string,int> *out)
1431 for (int osd = 0; osd < osdmap.get_max_osd(); ++osd) {
1432 if (osdmap.is_up(osd)) {
1433 map<string,string> meta;
1434 load_metadata(osd, meta, nullptr);
1435 auto p = meta.find(field);
1436 if (p == meta.end()) {
1437 (*out)["unknown"]++;
1439 (*out)[p->second]++;
1445 void OSDMonitor::count_metadata(const string& field, Formatter *f)
1447 map<string,int> by_val;
1448 count_metadata(field, &by_val);
1449 f->open_object_section(field.c_str());
1450 for (auto& p : by_val) {
1451 f->dump_int(p.first.c_str(), p.second);
1456 int OSDMonitor::get_osd_objectstore_type(int osd, string *type)
1458 map<string, string> metadata;
1459 int r = load_metadata(osd, metadata, nullptr);
1463 auto it = metadata.find("osd_objectstore");
1464 if (it == metadata.end())
1470 bool OSDMonitor::is_pool_currently_all_bluestore(int64_t pool_id,
1471 const pg_pool_t &pool,
1474 // just check a few pgs for efficiency - this can't give a guarantee anyway,
1475 // since filestore osds could always join the pool later
1476 set<int> checked_osds;
1477 for (unsigned ps = 0; ps < MIN(8, pool.get_pg_num()); ++ps) {
1478 vector<int> up, acting;
1479 pg_t pgid(ps, pool_id, -1);
1480 osdmap.pg_to_up_acting_osds(pgid, up, acting);
1481 for (int osd : up) {
1482 if (checked_osds.find(osd) != checked_osds.end())
1484 string objectstore_type;
1485 int r = get_osd_objectstore_type(osd, &objectstore_type);
1486 // allow with missing metadata, e.g. due to an osd never booting yet
1487 if (r < 0 || objectstore_type == "bluestore") {
1488 checked_osds.insert(osd);
1491 *err << "osd." << osd << " uses " << objectstore_type;
1498 int OSDMonitor::dump_osd_metadata(int osd, Formatter *f, ostream *err)
1500 map<string,string> m;
1501 if (int r = load_metadata(osd, m, err))
1503 for (map<string,string>::iterator p = m.begin(); p != m.end(); ++p)
1504 f->dump_string(p->first.c_str(), p->second);
1508 void OSDMonitor::print_nodes(Formatter *f)
1510 // group OSDs by their hosts
1511 map<string, list<int> > osds; // hostname => osd
1512 for (int osd = 0; osd < osdmap.get_max_osd(); osd++) {
1513 map<string, string> m;
1514 if (load_metadata(osd, m, NULL)) {
1517 map<string, string>::iterator hostname = m.find("hostname");
1518 if (hostname == m.end()) {
1519 // not likely though
1522 osds[hostname->second].push_back(osd);
1525 dump_services(f, osds, "osd");
1528 void OSDMonitor::share_map_with_random_osd()
1530 if (osdmap.get_num_up_osds() == 0) {
1531 dout(10) << __func__ << " no up osds, don't share with anyone" << dendl;
1535 MonSession *s = mon->session_map.get_random_osd_session(&osdmap);
1537 dout(10) << __func__ << " no up osd on our session map" << dendl;
1541 dout(10) << "committed, telling random " << s->inst << " all about it" << dendl;
1542 // whatev, they'll request more if they need it
1543 MOSDMap *m = build_incremental(osdmap.get_epoch() - 1, osdmap.get_epoch());
1544 s->con->send_message(m);
1545 // NOTE: do *not* record osd has up to this epoch (as we do
1546 // elsewhere) as they may still need to request older values.
1549 version_t OSDMonitor::get_trim_to()
1551 if (mon->get_quorum().empty()) {
1552 dout(10) << __func__ << ": quorum not formed" << dendl;
1557 if (mon->monmap->get_required_features().contains_all(
1558 ceph::features::mon::FEATURE_LUMINOUS)) {
1560 // TODO: Get this hidden in PGStatService
1561 std::lock_guard<std::mutex> l(creating_pgs_lock);
1562 if (!creating_pgs.pgs.empty()) {
1566 floor = get_min_last_epoch_clean();
1568 if (!mon->pgservice->is_readable())
1570 if (mon->pgservice->have_creating_pgs()) {
1573 floor = mon->pgservice->get_min_last_epoch_clean();
1576 dout(10) << " min_last_epoch_clean " << floor << dendl;
1577 if (g_conf->mon_osd_force_trim_to > 0 &&
1578 g_conf->mon_osd_force_trim_to < (int)get_last_committed()) {
1579 floor = g_conf->mon_osd_force_trim_to;
1580 dout(10) << " explicit mon_osd_force_trim_to = " << floor << dendl;
1582 unsigned min = g_conf->mon_min_osdmap_epochs;
1583 if (floor + min > get_last_committed()) {
1584 if (min < get_last_committed())
1585 floor = get_last_committed() - min;
1589 if (floor > get_first_committed())
1595 epoch_t OSDMonitor::get_min_last_epoch_clean() const
1597 auto floor = last_epoch_clean.get_lower_bound(osdmap);
1598 // also scan osd epochs
1599 // don't trim past the oldest reported osd epoch
1600 for (auto& osd_epoch : osd_epochs) {
1601 if (osd_epoch.second < floor) {
1602 floor = osd_epoch.second;
1608 void OSDMonitor::encode_trim_extra(MonitorDBStore::TransactionRef tx,
1611 dout(10) << __func__ << " including full map for e " << first << dendl;
1613 get_version_full(first, bl);
1614 put_version_full(tx, first, bl);
1619 bool OSDMonitor::preprocess_query(MonOpRequestRef op)
1621 op->mark_osdmon_event(__func__);
1622 Message *m = op->get_req();
1623 dout(10) << "preprocess_query " << *m << " from " << m->get_orig_source_inst() << dendl;
1625 switch (m->get_type()) {
1627 case MSG_MON_COMMAND:
1628 return preprocess_command(op);
1629 case CEPH_MSG_MON_GET_OSDMAP:
1630 return preprocess_get_osdmap(op);
1633 case MSG_OSD_MARK_ME_DOWN:
1634 return preprocess_mark_me_down(op);
1636 return preprocess_full(op);
1637 case MSG_OSD_FAILURE:
1638 return preprocess_failure(op);
1640 return preprocess_boot(op);
1642 return preprocess_alive(op);
1643 case MSG_OSD_PG_CREATED:
1644 return preprocess_pg_created(op);
1645 case MSG_OSD_PGTEMP:
1646 return preprocess_pgtemp(op);
1647 case MSG_OSD_BEACON:
1648 return preprocess_beacon(op);
1650 case CEPH_MSG_POOLOP:
1651 return preprocess_pool_op(op);
1653 case MSG_REMOVE_SNAPS:
1654 return preprocess_remove_snaps(op);
1662 bool OSDMonitor::prepare_update(MonOpRequestRef op)
1664 op->mark_osdmon_event(__func__);
1665 Message *m = op->get_req();
1666 dout(7) << "prepare_update " << *m << " from " << m->get_orig_source_inst() << dendl;
1668 switch (m->get_type()) {
1670 case MSG_OSD_MARK_ME_DOWN:
1671 return prepare_mark_me_down(op);
1673 return prepare_full(op);
1674 case MSG_OSD_FAILURE:
1675 return prepare_failure(op);
1677 return prepare_boot(op);
1679 return prepare_alive(op);
1680 case MSG_OSD_PG_CREATED:
1681 return prepare_pg_created(op);
1682 case MSG_OSD_PGTEMP:
1683 return prepare_pgtemp(op);
1684 case MSG_OSD_BEACON:
1685 return prepare_beacon(op);
1687 case MSG_MON_COMMAND:
1688 return prepare_command(op);
1690 case CEPH_MSG_POOLOP:
1691 return prepare_pool_op(op);
1693 case MSG_REMOVE_SNAPS:
1694 return prepare_remove_snaps(op);
1704 bool OSDMonitor::should_propose(double& delay)
1706 dout(10) << "should_propose" << dendl;
1708 // if full map, propose immediately! any subsequent changes will be clobbered.
1709 if (pending_inc.fullmap.length())
1712 // adjust osd weights?
1713 if (!osd_weight.empty() &&
1714 osd_weight.size() == (unsigned)osdmap.get_max_osd()) {
1715 dout(0) << " adjusting osd weights based on " << osd_weight << dendl;
1716 osdmap.adjust_osd_weights(osd_weight, pending_inc);
1722 // propose as fast as possible if updating up_thru or pg_temp
1723 // want to merge OSDMap changes as much as possible
1724 if ((pending_inc.new_primary_temp.size() == 1
1725 || pending_inc.new_up_thru.size() == 1)
1726 && pending_inc.new_state.size() < 2) {
1727 dout(15) << " propose as fast as possible for up_thru/pg_temp" << dendl;
1729 utime_t now = ceph_clock_now();
1730 if (now - last_attempted_minwait_time > g_conf->paxos_propose_interval
1731 && now - paxos->get_last_commit_time() > g_conf->paxos_min_wait) {
1732 delay = g_conf->paxos_min_wait;
1733 last_attempted_minwait_time = now;
1738 return PaxosService::should_propose(delay);
1743 // ---------------------------
1746 bool OSDMonitor::preprocess_get_osdmap(MonOpRequestRef op)
1748 op->mark_osdmon_event(__func__);
1749 MMonGetOSDMap *m = static_cast<MMonGetOSDMap*>(op->get_req());
1750 dout(10) << __func__ << " " << *m << dendl;
1751 MOSDMap *reply = new MOSDMap(mon->monmap->fsid);
1752 epoch_t first = get_first_committed();
1753 epoch_t last = osdmap.get_epoch();
1754 int max = g_conf->osd_map_message_max;
1755 for (epoch_t e = MAX(first, m->get_full_first());
1756 e <= MIN(last, m->get_full_last()) && max > 0;
1758 int r = get_version_full(e, reply->maps[e]);
1761 for (epoch_t e = MAX(first, m->get_inc_first());
1762 e <= MIN(last, m->get_inc_last()) && max > 0;
1764 int r = get_version(e, reply->incremental_maps[e]);
1767 reply->oldest_map = first;
1768 reply->newest_map = last;
1769 mon->send_reply(op, reply);
1774 // ---------------------------
1779 bool OSDMonitor::check_source(PaxosServiceMessage *m, uuid_d fsid) {
1780 // check permissions
1781 MonSession *session = m->get_session();
1784 if (!session->is_capable("osd", MON_CAP_X)) {
1785 dout(0) << "got MOSDFailure from entity with insufficient caps "
1786 << session->caps << dendl;
1789 if (fsid != mon->monmap->fsid) {
1790 dout(0) << "check_source: on fsid " << fsid
1791 << " != " << mon->monmap->fsid << dendl;
1798 bool OSDMonitor::preprocess_failure(MonOpRequestRef op)
1800 op->mark_osdmon_event(__func__);
1801 MOSDFailure *m = static_cast<MOSDFailure*>(op->get_req());
1802 // who is target_osd
1803 int badboy = m->get_target().name.num();
1805 // check permissions
1806 if (check_source(m, m->fsid))
1809 // first, verify the reporting host is valid
1810 if (m->get_orig_source().is_osd()) {
1811 int from = m->get_orig_source().num();
1812 if (!osdmap.exists(from) ||
1813 osdmap.get_addr(from) != m->get_orig_source_inst().addr ||
1814 (osdmap.is_down(from) && m->if_osd_failed())) {
1815 dout(5) << "preprocess_failure from dead osd." << from << ", ignoring" << dendl;
1816 send_incremental(op, m->get_epoch()+1);
1823 if (osdmap.is_down(badboy)) {
1824 dout(5) << "preprocess_failure dne(/dup?): " << m->get_target() << ", from " << m->get_orig_source_inst() << dendl;
1825 if (m->get_epoch() < osdmap.get_epoch())
1826 send_incremental(op, m->get_epoch()+1);
1829 if (osdmap.get_inst(badboy) != m->get_target()) {
1830 dout(5) << "preprocess_failure wrong osd: report " << m->get_target() << " != map's " << osdmap.get_inst(badboy)
1831 << ", from " << m->get_orig_source_inst() << dendl;
1832 if (m->get_epoch() < osdmap.get_epoch())
1833 send_incremental(op, m->get_epoch()+1);
1837 // already reported?
1838 if (osdmap.is_down(badboy) ||
1839 osdmap.get_up_from(badboy) > m->get_epoch()) {
1840 dout(5) << "preprocess_failure dup/old: " << m->get_target() << ", from " << m->get_orig_source_inst() << dendl;
1841 if (m->get_epoch() < osdmap.get_epoch())
1842 send_incremental(op, m->get_epoch()+1);
1846 if (!can_mark_down(badboy)) {
1847 dout(5) << "preprocess_failure ignoring report of " << m->get_target() << " from " << m->get_orig_source_inst() << dendl;
1851 dout(10) << "preprocess_failure new: " << m->get_target() << ", from " << m->get_orig_source_inst() << dendl;
1858 class C_AckMarkedDown : public C_MonOp {
1864 : C_MonOp(op), osdmon(osdmon) {}
1866 void _finish(int) override {
1867 MOSDMarkMeDown *m = static_cast<MOSDMarkMeDown*>(op->get_req());
1868 osdmon->mon->send_reply(
1874 false)); // ACK itself does not request an ack
1876 ~C_AckMarkedDown() override {
1880 bool OSDMonitor::preprocess_mark_me_down(MonOpRequestRef op)
1882 op->mark_osdmon_event(__func__);
1883 MOSDMarkMeDown *m = static_cast<MOSDMarkMeDown*>(op->get_req());
1884 int requesting_down = m->get_target().name.num();
1885 int from = m->get_orig_source().num();
1887 // check permissions
1888 if (check_source(m, m->fsid))
1891 // first, verify the reporting host is valid
1892 if (!m->get_orig_source().is_osd())
1895 if (!osdmap.exists(from) ||
1896 osdmap.is_down(from) ||
1897 osdmap.get_addr(from) != m->get_target().addr) {
1898 dout(5) << "preprocess_mark_me_down from dead osd."
1899 << from << ", ignoring" << dendl;
1900 send_incremental(op, m->get_epoch()+1);
1904 // no down might be set
1905 if (!can_mark_down(requesting_down))
1908 dout(10) << "MOSDMarkMeDown for: " << m->get_target() << dendl;
1912 if (m->request_ack) {
1913 Context *c(new C_AckMarkedDown(this, op));
1919 bool OSDMonitor::prepare_mark_me_down(MonOpRequestRef op)
1921 op->mark_osdmon_event(__func__);
1922 MOSDMarkMeDown *m = static_cast<MOSDMarkMeDown*>(op->get_req());
1923 int target_osd = m->get_target().name.num();
1925 assert(osdmap.is_up(target_osd));
1926 assert(osdmap.get_addr(target_osd) == m->get_target().addr);
1928 mon->clog->info() << "osd." << target_osd << " marked itself down";
1929 pending_inc.new_state[target_osd] = CEPH_OSD_UP;
1931 wait_for_finished_proposal(op, new C_AckMarkedDown(this, op));
1935 bool OSDMonitor::can_mark_down(int i)
1937 if (osdmap.test_flag(CEPH_OSDMAP_NODOWN)) {
1938 dout(5) << __func__ << " NODOWN flag set, will not mark osd." << i
1939 << " down" << dendl;
1943 if (osdmap.is_nodown(i)) {
1944 dout(5) << __func__ << " osd." << i << " is marked as nodown, "
1945 << "will not mark it down" << dendl;
1949 int num_osds = osdmap.get_num_osds();
1950 if (num_osds == 0) {
1951 dout(5) << __func__ << " no osds" << dendl;
1954 int up = osdmap.get_num_up_osds() - pending_inc.get_net_marked_down(&osdmap);
1955 float up_ratio = (float)up / (float)num_osds;
1956 if (up_ratio < g_conf->mon_osd_min_up_ratio) {
1957 dout(2) << __func__ << " current up_ratio " << up_ratio << " < min "
1958 << g_conf->mon_osd_min_up_ratio
1959 << ", will not mark osd." << i << " down" << dendl;
1965 bool OSDMonitor::can_mark_up(int i)
1967 if (osdmap.test_flag(CEPH_OSDMAP_NOUP)) {
1968 dout(5) << __func__ << " NOUP flag set, will not mark osd." << i
1973 if (osdmap.is_noup(i)) {
1974 dout(5) << __func__ << " osd." << i << " is marked as noup, "
1975 << "will not mark it up" << dendl;
1983 * @note the parameter @p i apparently only exists here so we can output the
1984 * osd's id on messages.
1986 bool OSDMonitor::can_mark_out(int i)
1988 if (osdmap.test_flag(CEPH_OSDMAP_NOOUT)) {
1989 dout(5) << __func__ << " NOOUT flag set, will not mark osds out" << dendl;
1993 if (osdmap.is_noout(i)) {
1994 dout(5) << __func__ << " osd." << i << " is marked as noout, "
1995 << "will not mark it out" << dendl;
1999 int num_osds = osdmap.get_num_osds();
2000 if (num_osds == 0) {
2001 dout(5) << __func__ << " no osds" << dendl;
2004 int in = osdmap.get_num_in_osds() - pending_inc.get_net_marked_out(&osdmap);
2005 float in_ratio = (float)in / (float)num_osds;
2006 if (in_ratio < g_conf->mon_osd_min_in_ratio) {
2008 dout(5) << __func__ << " current in_ratio " << in_ratio << " < min "
2009 << g_conf->mon_osd_min_in_ratio
2010 << ", will not mark osd." << i << " out" << dendl;
2012 dout(5) << __func__ << " current in_ratio " << in_ratio << " < min "
2013 << g_conf->mon_osd_min_in_ratio
2014 << ", will not mark osds out" << dendl;
2021 bool OSDMonitor::can_mark_in(int i)
2023 if (osdmap.test_flag(CEPH_OSDMAP_NOIN)) {
2024 dout(5) << __func__ << " NOIN flag set, will not mark osd." << i
2029 if (osdmap.is_noin(i)) {
2030 dout(5) << __func__ << " osd." << i << " is marked as noin, "
2031 << "will not mark it in" << dendl;
2038 bool OSDMonitor::check_failures(utime_t now)
2040 bool found_failure = false;
2041 for (map<int,failure_info_t>::iterator p = failure_info.begin();
2042 p != failure_info.end();
2044 if (can_mark_down(p->first)) {
2045 found_failure |= check_failure(now, p->first, p->second);
2048 return found_failure;
2051 bool OSDMonitor::check_failure(utime_t now, int target_osd, failure_info_t& fi)
2053 // already pending failure?
2054 if (pending_inc.new_state.count(target_osd) &&
2055 pending_inc.new_state[target_osd] & CEPH_OSD_UP) {
2056 dout(10) << " already pending failure" << dendl;
2060 set<string> reporters_by_subtree;
2061 string reporter_subtree_level = g_conf->mon_osd_reporter_subtree_level;
2062 utime_t orig_grace(g_conf->osd_heartbeat_grace, 0);
2063 utime_t max_failed_since = fi.get_failed_since();
2064 utime_t failed_for = now - max_failed_since;
2066 utime_t grace = orig_grace;
2067 double my_grace = 0, peer_grace = 0;
2069 if (g_conf->mon_osd_adjust_heartbeat_grace) {
2070 double halflife = (double)g_conf->mon_osd_laggy_halflife;
2071 decay_k = ::log(.5) / halflife;
2073 // scale grace period based on historical probability of 'lagginess'
2074 // (false positive failures due to slowness).
2075 const osd_xinfo_t& xi = osdmap.get_xinfo(target_osd);
2076 double decay = exp((double)failed_for * decay_k);
2077 dout(20) << " halflife " << halflife << " decay_k " << decay_k
2078 << " failed_for " << failed_for << " decay " << decay << dendl;
2079 my_grace = decay * (double)xi.laggy_interval * xi.laggy_probability;
2083 // consider the peers reporting a failure a proxy for a potential
2084 // 'subcluster' over the overall cluster that is similarly
2085 // laggy. this is clearly not true in all cases, but will sometimes
2086 // help us localize the grace correction to a subset of the system
2087 // (say, a rack with a bad switch) that is unhappy.
2088 assert(fi.reporters.size());
2089 for (map<int,failure_reporter_t>::iterator p = fi.reporters.begin();
2090 p != fi.reporters.end();
2092 // get the parent bucket whose type matches with "reporter_subtree_level".
2093 // fall back to OSD if the level doesn't exist.
2094 map<string, string> reporter_loc = osdmap.crush->get_full_location(p->first);
2095 map<string, string>::iterator iter = reporter_loc.find(reporter_subtree_level);
2096 if (iter == reporter_loc.end()) {
2097 reporters_by_subtree.insert("osd." + to_string(p->first));
2099 reporters_by_subtree.insert(iter->second);
2101 if (g_conf->mon_osd_adjust_heartbeat_grace) {
2102 const osd_xinfo_t& xi = osdmap.get_xinfo(p->first);
2103 utime_t elapsed = now - xi.down_stamp;
2104 double decay = exp((double)elapsed * decay_k);
2105 peer_grace += decay * (double)xi.laggy_interval * xi.laggy_probability;
2109 if (g_conf->mon_osd_adjust_heartbeat_grace) {
2110 peer_grace /= (double)fi.reporters.size();
2111 grace += peer_grace;
2114 dout(10) << " osd." << target_osd << " has "
2115 << fi.reporters.size() << " reporters, "
2116 << grace << " grace (" << orig_grace << " + " << my_grace
2117 << " + " << peer_grace << "), max_failed_since " << max_failed_since
2120 if (failed_for >= grace &&
2121 (int)reporters_by_subtree.size() >= g_conf->mon_osd_min_down_reporters) {
2122 dout(1) << " we have enough reporters to mark osd." << target_osd
2123 << " down" << dendl;
2124 pending_inc.new_state[target_osd] = CEPH_OSD_UP;
2126 mon->clog->info() << "osd." << target_osd << " failed ("
2127 << osdmap.crush->get_full_location_ordered_string(
2130 << (int)reporters_by_subtree.size()
2131 << " reporters from different "
2132 << reporter_subtree_level << " after "
2133 << failed_for << " >= grace " << grace << ")";
2139 void OSDMonitor::force_failure(int target_osd, int by)
2141 // already pending failure?
2142 if (pending_inc.new_state.count(target_osd) &&
2143 pending_inc.new_state[target_osd] & CEPH_OSD_UP) {
2144 dout(10) << " already pending failure" << dendl;
2148 dout(1) << " we're forcing failure of osd." << target_osd << dendl;
2149 pending_inc.new_state[target_osd] = CEPH_OSD_UP;
2151 mon->clog->info() << "osd." << target_osd << " failed ("
2152 << osdmap.crush->get_full_location_ordered_string(target_osd)
2153 << ") (connection refused reported by osd." << by << ")";
2157 bool OSDMonitor::prepare_failure(MonOpRequestRef op)
2159 op->mark_osdmon_event(__func__);
2160 MOSDFailure *m = static_cast<MOSDFailure*>(op->get_req());
2161 dout(1) << "prepare_failure " << m->get_target()
2162 << " from " << m->get_orig_source_inst()
2163 << " is reporting failure:" << m->if_osd_failed() << dendl;
2165 int target_osd = m->get_target().name.num();
2166 int reporter = m->get_orig_source().num();
2167 assert(osdmap.is_up(target_osd));
2168 assert(osdmap.get_addr(target_osd) == m->get_target().addr);
2170 if (m->if_osd_failed()) {
2171 // calculate failure time
2172 utime_t now = ceph_clock_now();
2173 utime_t failed_since =
2174 m->get_recv_stamp() - utime_t(m->failed_for, 0);
2177 if (m->is_immediate()) {
2178 mon->clog->debug() << m->get_target() << " reported immediately failed by "
2179 << m->get_orig_source_inst();
2180 force_failure(target_osd, reporter);
2183 mon->clog->debug() << m->get_target() << " reported failed by "
2184 << m->get_orig_source_inst();
2186 failure_info_t& fi = failure_info[target_osd];
2187 MonOpRequestRef old_op = fi.add_report(reporter, failed_since, op);
2189 mon->no_reply(old_op);
2192 return check_failure(now, target_osd, fi);
2194 // remove the report
2195 mon->clog->debug() << m->get_target() << " failure report canceled by "
2196 << m->get_orig_source_inst();
2197 if (failure_info.count(target_osd)) {
2198 failure_info_t& fi = failure_info[target_osd];
2199 MonOpRequestRef report_op = fi.cancel_report(reporter);
2201 mon->no_reply(report_op);
2203 if (fi.reporters.empty()) {
2204 dout(10) << " removing last failure_info for osd." << target_osd
2206 failure_info.erase(target_osd);
2208 dout(10) << " failure_info for osd." << target_osd << " now "
2209 << fi.reporters.size() << " reporters" << dendl;
2212 dout(10) << " no failure_info for osd." << target_osd << dendl;
2220 void OSDMonitor::process_failures()
2222 map<int,failure_info_t>::iterator p = failure_info.begin();
2223 while (p != failure_info.end()) {
2224 if (osdmap.is_up(p->first)) {
2227 dout(10) << "process_failures osd." << p->first << dendl;
2228 list<MonOpRequestRef> ls;
2229 p->second.take_report_messages(ls);
2230 failure_info.erase(p++);
2232 while (!ls.empty()) {
2233 MonOpRequestRef o = ls.front();
2235 o->mark_event(__func__);
2236 MOSDFailure *m = o->get_req<MOSDFailure>();
2237 send_latest(o, m->get_epoch());
2245 void OSDMonitor::take_all_failures(list<MonOpRequestRef>& ls)
2247 dout(10) << __func__ << " on " << failure_info.size() << " osds" << dendl;
2249 for (map<int,failure_info_t>::iterator p = failure_info.begin();
2250 p != failure_info.end();
2252 p->second.take_report_messages(ls);
2254 failure_info.clear();
2260 bool OSDMonitor::preprocess_boot(MonOpRequestRef op)
2262 op->mark_osdmon_event(__func__);
2263 MOSDBoot *m = static_cast<MOSDBoot*>(op->get_req());
2264 int from = m->get_orig_source_inst().name.num();
2266 // check permissions, ignore if failed (no response expected)
2267 MonSession *session = m->get_session();
2270 if (!session->is_capable("osd", MON_CAP_X)) {
2271 dout(0) << "got preprocess_boot message from entity with insufficient caps"
2272 << session->caps << dendl;
2276 if (m->sb.cluster_fsid != mon->monmap->fsid) {
2277 dout(0) << "preprocess_boot on fsid " << m->sb.cluster_fsid
2278 << " != " << mon->monmap->fsid << dendl;
2282 if (m->get_orig_source_inst().addr.is_blank_ip()) {
2283 dout(0) << "preprocess_boot got blank addr for " << m->get_orig_source_inst() << dendl;
2287 assert(m->get_orig_source_inst().name.is_osd());
2289 // check if osd has required features to boot
2290 if ((osdmap.get_features(CEPH_ENTITY_TYPE_OSD, NULL) &
2291 CEPH_FEATURE_OSD_ERASURE_CODES) &&
2292 !(m->get_connection()->get_features() & CEPH_FEATURE_OSD_ERASURE_CODES)) {
2293 dout(0) << __func__ << " osdmap requires erasure code but osd at "
2294 << m->get_orig_source_inst()
2295 << " doesn't announce support -- ignore" << dendl;
2299 if ((osdmap.get_features(CEPH_ENTITY_TYPE_OSD, NULL) &
2300 CEPH_FEATURE_ERASURE_CODE_PLUGINS_V2) &&
2301 !(m->get_connection()->get_features() & CEPH_FEATURE_ERASURE_CODE_PLUGINS_V2)) {
2302 dout(0) << __func__ << " osdmap requires erasure code plugins v2 but osd at "
2303 << m->get_orig_source_inst()
2304 << " doesn't announce support -- ignore" << dendl;
2308 if ((osdmap.get_features(CEPH_ENTITY_TYPE_OSD, NULL) &
2309 CEPH_FEATURE_ERASURE_CODE_PLUGINS_V3) &&
2310 !(m->get_connection()->get_features() & CEPH_FEATURE_ERASURE_CODE_PLUGINS_V3)) {
2311 dout(0) << __func__ << " osdmap requires erasure code plugins v3 but osd at "
2312 << m->get_orig_source_inst()
2313 << " doesn't announce support -- ignore" << dendl;
2317 if (osdmap.require_osd_release >= CEPH_RELEASE_LUMINOUS &&
2318 !HAVE_FEATURE(m->osd_features, SERVER_LUMINOUS)) {
2319 mon->clog->info() << "disallowing boot of OSD "
2320 << m->get_orig_source_inst()
2321 << " because the osdmap requires"
2322 << " CEPH_FEATURE_SERVER_LUMINOUS"
2323 << " but the osd lacks CEPH_FEATURE_SERVER_LUMINOUS";
2327 if (osdmap.require_osd_release >= CEPH_RELEASE_JEWEL &&
2328 !(m->osd_features & CEPH_FEATURE_SERVER_JEWEL)) {
2329 mon->clog->info() << "disallowing boot of OSD "
2330 << m->get_orig_source_inst()
2331 << " because the osdmap requires"
2332 << " CEPH_FEATURE_SERVER_JEWEL"
2333 << " but the osd lacks CEPH_FEATURE_SERVER_JEWEL";
2337 if (osdmap.require_osd_release >= CEPH_RELEASE_KRAKEN &&
2338 !HAVE_FEATURE(m->osd_features, SERVER_KRAKEN)) {
2339 mon->clog->info() << "disallowing boot of OSD "
2340 << m->get_orig_source_inst()
2341 << " because the osdmap requires"
2342 << " CEPH_FEATURE_SERVER_KRAKEN"
2343 << " but the osd lacks CEPH_FEATURE_SERVER_KRAKEN";
2347 if (osdmap.test_flag(CEPH_OSDMAP_SORTBITWISE) &&
2348 !(m->osd_features & CEPH_FEATURE_OSD_BITWISE_HOBJ_SORT)) {
2349 mon->clog->info() << "disallowing boot of OSD "
2350 << m->get_orig_source_inst()
2351 << " because 'sortbitwise' osdmap flag is set and OSD lacks the OSD_BITWISE_HOBJ_SORT feature";
2355 if (osdmap.test_flag(CEPH_OSDMAP_RECOVERY_DELETES) &&
2356 !(m->osd_features & CEPH_FEATURE_OSD_RECOVERY_DELETES)) {
2357 mon->clog->info() << "disallowing boot of OSD "
2358 << m->get_orig_source_inst()
2359 << " because 'recovery_deletes' osdmap flag is set and OSD lacks the OSD_RECOVERY_DELETES feature";
2363 if (any_of(osdmap.get_pools().begin(),
2364 osdmap.get_pools().end(),
2365 [](const std::pair<int64_t,pg_pool_t>& pool)
2366 { return pool.second.use_gmt_hitset; })) {
2367 assert(osdmap.get_num_up_osds() == 0 ||
2368 osdmap.get_up_osd_features() & CEPH_FEATURE_OSD_HITSET_GMT);
2369 if (!(m->osd_features & CEPH_FEATURE_OSD_HITSET_GMT)) {
2370 dout(0) << __func__ << " one or more pools uses GMT hitsets but osd at "
2371 << m->get_orig_source_inst()
2372 << " doesn't announce support -- ignore" << dendl;
2377 // make sure upgrades stop at luminous
2378 if (HAVE_FEATURE(m->osd_features, SERVER_M) &&
2379 osdmap.require_osd_release < CEPH_RELEASE_LUMINOUS) {
2380 mon->clog->info() << "disallowing boot of post-luminous OSD "
2381 << m->get_orig_source_inst()
2382 << " because require_osd_release < luminous";
2386 // make sure upgrades stop at jewel
2387 if (HAVE_FEATURE(m->osd_features, SERVER_KRAKEN) &&
2388 osdmap.require_osd_release < CEPH_RELEASE_JEWEL) {
2389 mon->clog->info() << "disallowing boot of post-jewel OSD "
2390 << m->get_orig_source_inst()
2391 << " because require_osd_release < jewel";
2395 // make sure upgrades stop at hammer
2396 // * HAMMER_0_94_4 is the required hammer feature
2397 // * MON_METADATA is the first post-hammer feature
2398 if (osdmap.get_num_up_osds() > 0) {
2399 if ((m->osd_features & CEPH_FEATURE_MON_METADATA) &&
2400 !(osdmap.get_up_osd_features() & CEPH_FEATURE_HAMMER_0_94_4)) {
2401 mon->clog->info() << "disallowing boot of post-hammer OSD "
2402 << m->get_orig_source_inst()
2403 << " because one or more up OSDs is pre-hammer v0.94.4";
2406 if (!(m->osd_features & CEPH_FEATURE_HAMMER_0_94_4) &&
2407 (osdmap.get_up_osd_features() & CEPH_FEATURE_MON_METADATA)) {
2408 mon->clog->info() << "disallowing boot of pre-hammer v0.94.4 OSD "
2409 << m->get_orig_source_inst()
2410 << " because all up OSDs are post-hammer";
2416 if (osdmap.is_up(from) &&
2417 osdmap.get_inst(from) == m->get_orig_source_inst() &&
2418 osdmap.get_cluster_addr(from) == m->cluster_addr) {
2420 dout(7) << "preprocess_boot dup from " << m->get_orig_source_inst()
2421 << " == " << osdmap.get_inst(from) << dendl;
2426 if (osdmap.exists(from) &&
2427 !osdmap.get_uuid(from).is_zero() &&
2428 osdmap.get_uuid(from) != m->sb.osd_fsid) {
2429 dout(7) << __func__ << " from " << m->get_orig_source_inst()
2430 << " clashes with existing osd: different fsid"
2431 << " (ours: " << osdmap.get_uuid(from)
2432 << " ; theirs: " << m->sb.osd_fsid << ")" << dendl;
2436 if (osdmap.exists(from) &&
2437 osdmap.get_info(from).up_from > m->version &&
2438 osdmap.get_most_recent_inst(from) == m->get_orig_source_inst()) {
2439 dout(7) << "prepare_boot msg from before last up_from, ignoring" << dendl;
2440 send_latest(op, m->sb.current_epoch+1);
2445 if (!can_mark_up(from)) {
2446 dout(7) << "preprocess_boot ignoring boot from " << m->get_orig_source_inst() << dendl;
2447 send_latest(op, m->sb.current_epoch+1);
2451 dout(10) << "preprocess_boot from " << m->get_orig_source_inst() << dendl;
2458 bool OSDMonitor::prepare_boot(MonOpRequestRef op)
2460 op->mark_osdmon_event(__func__);
2461 MOSDBoot *m = static_cast<MOSDBoot*>(op->get_req());
2462 dout(7) << __func__ << " from " << m->get_orig_source_inst() << " sb " << m->sb
2463 << " cluster_addr " << m->cluster_addr
2464 << " hb_back_addr " << m->hb_back_addr
2465 << " hb_front_addr " << m->hb_front_addr
2468 assert(m->get_orig_source().is_osd());
2469 int from = m->get_orig_source().num();
2471 // does this osd exist?
2472 if (from >= osdmap.get_max_osd()) {
2473 dout(1) << "boot from osd." << from << " >= max_osd "
2474 << osdmap.get_max_osd() << dendl;
2478 int oldstate = osdmap.exists(from) ? osdmap.get_state(from) : CEPH_OSD_NEW;
2479 if (pending_inc.new_state.count(from))
2480 oldstate ^= pending_inc.new_state[from];
2482 // already up? mark down first?
2483 if (osdmap.is_up(from)) {
2484 dout(7) << __func__ << " was up, first marking down "
2485 << osdmap.get_inst(from) << dendl;
2486 // preprocess should have caught these; if not, assert.
2487 assert(osdmap.get_inst(from) != m->get_orig_source_inst() ||
2488 osdmap.get_cluster_addr(from) != m->cluster_addr);
2489 assert(osdmap.get_uuid(from) == m->sb.osd_fsid);
2491 if (pending_inc.new_state.count(from) == 0 ||
2492 (pending_inc.new_state[from] & CEPH_OSD_UP) == 0) {
2493 // mark previous guy down
2494 pending_inc.new_state[from] = CEPH_OSD_UP;
2496 wait_for_finished_proposal(op, new C_RetryMessage(this, op));
2497 } else if (pending_inc.new_up_client.count(from)) {
2498 // already prepared, just wait
2499 dout(7) << __func__ << " already prepared, waiting on "
2500 << m->get_orig_source_addr() << dendl;
2501 wait_for_finished_proposal(op, new C_RetryMessage(this, op));
2504 pending_inc.new_up_client[from] = m->get_orig_source_addr();
2505 if (!m->cluster_addr.is_blank_ip())
2506 pending_inc.new_up_cluster[from] = m->cluster_addr;
2507 pending_inc.new_hb_back_up[from] = m->hb_back_addr;
2508 if (!m->hb_front_addr.is_blank_ip())
2509 pending_inc.new_hb_front_up[from] = m->hb_front_addr;
2511 down_pending_out.erase(from); // if any
2514 osd_weight[from] = m->sb.weight;
2517 dout(10) << " setting osd." << from << " uuid to " << m->sb.osd_fsid
2519 if (!osdmap.exists(from) || osdmap.get_uuid(from) != m->sb.osd_fsid) {
2520 // preprocess should have caught this; if not, assert.
2521 assert(!osdmap.exists(from) || osdmap.get_uuid(from).is_zero());
2522 pending_inc.new_uuid[from] = m->sb.osd_fsid;
2526 if (m->sb.newest_map == 0 && osdmap.exists(from)) {
2527 const osd_info_t& i = osdmap.get_info(from);
2528 if (i.up_from > i.lost_at) {
2529 dout(10) << " fresh osd; marking lost_at too" << dendl;
2530 pending_inc.new_lost[from] = osdmap.get_epoch();
2535 bufferlist osd_metadata;
2536 ::encode(m->metadata, osd_metadata);
2537 pending_metadata[from] = osd_metadata;
2538 pending_metadata_rm.erase(from);
2540 // adjust last clean unmount epoch?
2541 const osd_info_t& info = osdmap.get_info(from);
2542 dout(10) << " old osd_info: " << info << dendl;
2543 if (m->sb.mounted > info.last_clean_begin ||
2544 (m->sb.mounted == info.last_clean_begin &&
2545 m->sb.clean_thru > info.last_clean_end)) {
2546 epoch_t begin = m->sb.mounted;
2547 epoch_t end = m->sb.clean_thru;
2549 dout(10) << __func__ << " osd." << from << " last_clean_interval "
2550 << "[" << info.last_clean_begin << "," << info.last_clean_end
2551 << ") -> [" << begin << "-" << end << ")"
2553 pending_inc.new_last_clean_interval[from] =
2554 pair<epoch_t,epoch_t>(begin, end);
2557 osd_xinfo_t xi = osdmap.get_xinfo(from);
2558 if (m->boot_epoch == 0) {
2559 xi.laggy_probability *= (1.0 - g_conf->mon_osd_laggy_weight);
2560 xi.laggy_interval *= (1.0 - g_conf->mon_osd_laggy_weight);
2561 dout(10) << " not laggy, new xi " << xi << dendl;
2563 if (xi.down_stamp.sec()) {
2564 int interval = ceph_clock_now().sec() -
2565 xi.down_stamp.sec();
2566 if (g_conf->mon_osd_laggy_max_interval &&
2567 (interval > g_conf->mon_osd_laggy_max_interval)) {
2568 interval = g_conf->mon_osd_laggy_max_interval;
2571 interval * g_conf->mon_osd_laggy_weight +
2572 xi.laggy_interval * (1.0 - g_conf->mon_osd_laggy_weight);
2574 xi.laggy_probability =
2575 g_conf->mon_osd_laggy_weight +
2576 xi.laggy_probability * (1.0 - g_conf->mon_osd_laggy_weight);
2577 dout(10) << " laggy, now xi " << xi << dendl;
2580 // set features shared by the osd
2581 if (m->osd_features)
2582 xi.features = m->osd_features;
2584 xi.features = m->get_connection()->get_features();
2587 if ((g_conf->mon_osd_auto_mark_auto_out_in &&
2588 (oldstate & CEPH_OSD_AUTOOUT)) ||
2589 (g_conf->mon_osd_auto_mark_new_in && (oldstate & CEPH_OSD_NEW)) ||
2590 (g_conf->mon_osd_auto_mark_in)) {
2591 if (can_mark_in(from)) {
2592 if (osdmap.osd_xinfo[from].old_weight > 0) {
2593 pending_inc.new_weight[from] = osdmap.osd_xinfo[from].old_weight;
2596 pending_inc.new_weight[from] = CEPH_OSD_IN;
2599 dout(7) << __func__ << " NOIN set, will not mark in "
2600 << m->get_orig_source_addr() << dendl;
2604 pending_inc.new_xinfo[from] = xi;
2607 wait_for_finished_proposal(op, new C_Booted(this, op));
2612 void OSDMonitor::_booted(MonOpRequestRef op, bool logit)
2614 op->mark_osdmon_event(__func__);
2615 MOSDBoot *m = static_cast<MOSDBoot*>(op->get_req());
2616 dout(7) << "_booted " << m->get_orig_source_inst()
2617 << " w " << m->sb.weight << " from " << m->sb.current_epoch << dendl;
2620 mon->clog->info() << m->get_orig_source_inst() << " boot";
2623 send_latest(op, m->sb.current_epoch+1);
2630 bool OSDMonitor::preprocess_full(MonOpRequestRef op)
2632 op->mark_osdmon_event(__func__);
2633 MOSDFull *m = static_cast<MOSDFull*>(op->get_req());
2634 int from = m->get_orig_source().num();
2636 unsigned mask = CEPH_OSD_NEARFULL | CEPH_OSD_BACKFILLFULL | CEPH_OSD_FULL;
2638 // check permissions, ignore if failed
2639 MonSession *session = m->get_session();
2642 if (!session->is_capable("osd", MON_CAP_X)) {
2643 dout(0) << "MOSDFull from entity with insufficient privileges:"
2644 << session->caps << dendl;
2648 // ignore a full message from the osd instance that already went down
2649 if (!osdmap.exists(from)) {
2650 dout(7) << __func__ << " ignoring full message from nonexistent "
2651 << m->get_orig_source_inst() << dendl;
2654 if ((!osdmap.is_up(from) &&
2655 osdmap.get_most_recent_inst(from) == m->get_orig_source_inst()) ||
2656 (osdmap.is_up(from) &&
2657 osdmap.get_inst(from) != m->get_orig_source_inst())) {
2658 dout(7) << __func__ << " ignoring full message from down "
2659 << m->get_orig_source_inst() << dendl;
2663 OSDMap::calc_state_set(osdmap.get_state(from), state);
2665 if ((osdmap.get_state(from) & mask) == m->state) {
2666 dout(7) << __func__ << " state already " << state << " for osd." << from
2667 << " " << m->get_orig_source_inst() << dendl;
2668 _reply_map(op, m->version);
2672 dout(10) << __func__ << " want state " << state << " for osd." << from
2673 << " " << m->get_orig_source_inst() << dendl;
2680 bool OSDMonitor::prepare_full(MonOpRequestRef op)
2682 op->mark_osdmon_event(__func__);
2683 const MOSDFull *m = static_cast<MOSDFull*>(op->get_req());
2684 const int from = m->get_orig_source().num();
2686 const unsigned mask = CEPH_OSD_NEARFULL | CEPH_OSD_BACKFILLFULL | CEPH_OSD_FULL;
2687 const unsigned want_state = m->state & mask; // safety first
2689 unsigned cur_state = osdmap.get_state(from);
2690 auto p = pending_inc.new_state.find(from);
2691 if (p != pending_inc.new_state.end()) {
2692 cur_state ^= p->second;
2696 set<string> want_state_set, cur_state_set;
2697 OSDMap::calc_state_set(want_state, want_state_set);
2698 OSDMap::calc_state_set(cur_state, cur_state_set);
2700 if (cur_state != want_state) {
2701 if (p != pending_inc.new_state.end()) {
2704 pending_inc.new_state[from] = 0;
2706 pending_inc.new_state[from] |= (osdmap.get_state(from) & mask) ^ want_state;
2707 dout(7) << __func__ << " osd." << from << " " << cur_state_set
2708 << " -> " << want_state_set << dendl;
2710 dout(7) << __func__ << " osd." << from << " " << cur_state_set
2711 << " = wanted " << want_state_set << ", just waiting" << dendl;
2714 wait_for_finished_proposal(op, new C_ReplyMap(this, op, m->version));
2721 bool OSDMonitor::preprocess_alive(MonOpRequestRef op)
2723 op->mark_osdmon_event(__func__);
2724 MOSDAlive *m = static_cast<MOSDAlive*>(op->get_req());
2725 int from = m->get_orig_source().num();
2727 // check permissions, ignore if failed
2728 MonSession *session = m->get_session();
2731 if (!session->is_capable("osd", MON_CAP_X)) {
2732 dout(0) << "attempt to send MOSDAlive from entity with insufficient privileges:"
2733 << session->caps << dendl;
2737 if (!osdmap.is_up(from) ||
2738 osdmap.get_inst(from) != m->get_orig_source_inst()) {
2739 dout(7) << "preprocess_alive ignoring alive message from down " << m->get_orig_source_inst() << dendl;
2743 if (osdmap.get_up_thru(from) >= m->want) {
2745 dout(7) << "preprocess_alive want up_thru " << m->want << " dup from " << m->get_orig_source_inst() << dendl;
2746 _reply_map(op, m->version);
2750 dout(10) << "preprocess_alive want up_thru " << m->want
2751 << " from " << m->get_orig_source_inst() << dendl;
2758 bool OSDMonitor::prepare_alive(MonOpRequestRef op)
2760 op->mark_osdmon_event(__func__);
2761 MOSDAlive *m = static_cast<MOSDAlive*>(op->get_req());
2762 int from = m->get_orig_source().num();
2764 if (0) { // we probably don't care much about these
2765 mon->clog->debug() << m->get_orig_source_inst() << " alive";
2768 dout(7) << "prepare_alive want up_thru " << m->want << " have " << m->version
2769 << " from " << m->get_orig_source_inst() << dendl;
2771 update_up_thru(from, m->version); // set to the latest map the OSD has
2772 wait_for_finished_proposal(op, new C_ReplyMap(this, op, m->version));
2776 void OSDMonitor::_reply_map(MonOpRequestRef op, epoch_t e)
2778 op->mark_osdmon_event(__func__);
2779 dout(7) << "_reply_map " << e
2780 << " from " << op->get_req()->get_orig_source_inst()
2786 bool OSDMonitor::preprocess_pg_created(MonOpRequestRef op)
2788 op->mark_osdmon_event(__func__);
2789 auto m = static_cast<MOSDPGCreated*>(op->get_req());
2790 dout(10) << __func__ << " " << *m << dendl;
2791 auto session = m->get_session();
2793 dout(10) << __func__ << ": no monitor session!" << dendl;
2796 if (!session->is_capable("osd", MON_CAP_X)) {
2797 derr << __func__ << " received from entity "
2798 << "with insufficient privileges " << session->caps << dendl;
2801 // always forward the "created!" to the leader
2805 bool OSDMonitor::prepare_pg_created(MonOpRequestRef op)
2807 op->mark_osdmon_event(__func__);
2808 auto m = static_cast<MOSDPGCreated*>(op->get_req());
2809 dout(10) << __func__ << " " << *m << dendl;
2810 auto src = m->get_orig_source();
2811 auto from = src.num();
2812 if (!src.is_osd() ||
2813 !mon->osdmon()->osdmap.is_up(from) ||
2814 m->get_orig_source_inst() != mon->osdmon()->osdmap.get_inst(from)) {
2815 dout(1) << __func__ << " ignoring stats from non-active osd." << dendl;
2818 pending_created_pgs.push_back(m->pgid);
2825 bool OSDMonitor::preprocess_pgtemp(MonOpRequestRef op)
2827 MOSDPGTemp *m = static_cast<MOSDPGTemp*>(op->get_req());
2828 dout(10) << "preprocess_pgtemp " << *m << dendl;
2829 mempool::osdmap::vector<int> empty;
2830 int from = m->get_orig_source().num();
2831 size_t ignore_cnt = 0;
2834 MonSession *session = m->get_session();
2837 if (!session->is_capable("osd", MON_CAP_X)) {
2838 dout(0) << "attempt to send MOSDPGTemp from entity with insufficient caps "
2839 << session->caps << dendl;
2843 if (!osdmap.is_up(from) ||
2844 osdmap.get_inst(from) != m->get_orig_source_inst()) {
2845 dout(7) << "ignoring pgtemp message from down " << m->get_orig_source_inst() << dendl;
2853 for (auto p = m->pg_temp.begin(); p != m->pg_temp.end(); ++p) {
2854 dout(20) << " " << p->first
2855 << (osdmap.pg_temp->count(p->first) ? osdmap.pg_temp->get(p->first) : empty)
2856 << " -> " << p->second << dendl;
2858 // does the pool exist?
2859 if (!osdmap.have_pg_pool(p->first.pool())) {
2861 * 1. If the osdmap does not have the pool, it means the pool has been
2862 * removed in-between the osd sending this message and us handling it.
2863 * 2. If osdmap doesn't have the pool, it is safe to assume the pool does
2864 * not exist in the pending either, as the osds would not send a
2865 * message about a pool they know nothing about (yet).
2866 * 3. However, if the pool does exist in the pending, then it must be a
2867 * new pool, and not relevant to this message (see 1).
2869 dout(10) << __func__ << " ignore " << p->first << " -> " << p->second
2870 << ": pool has been removed" << dendl;
2875 int acting_primary = -1;
2876 osdmap.pg_to_up_acting_osds(
2877 p->first, nullptr, nullptr, nullptr, &acting_primary);
2878 if (acting_primary != from) {
2879 /* If the source isn't the primary based on the current osdmap, we know
2880 * that the interval changed and that we can discard this message.
2881 * Indeed, we must do so to avoid 16127 since we can't otherwise determine
2882 * which of two pg temp mappings on the same pg is more recent.
2884 dout(10) << __func__ << " ignore " << p->first << " -> " << p->second
2885 << ": primary has changed" << dendl;
2891 if (p->second.empty() && (osdmap.pg_temp->count(p->first) ||
2892 osdmap.primary_temp->count(p->first)))
2895 // NOTE: we assume that this will clear pg_primary, so consider
2896 // an existing pg_primary field to imply a change
2897 if (p->second.size() &&
2898 (osdmap.pg_temp->count(p->first) == 0 ||
2899 !vectors_equal(osdmap.pg_temp->get(p->first), p->second) ||
2900 osdmap.primary_temp->count(p->first)))
2904 // should we ignore all the pgs?
2905 if (ignore_cnt == m->pg_temp.size())
2908 dout(7) << "preprocess_pgtemp e" << m->map_epoch << " no changes from " << m->get_orig_source_inst() << dendl;
2909 _reply_map(op, m->map_epoch);
2916 void OSDMonitor::update_up_thru(int from, epoch_t up_thru)
2918 epoch_t old_up_thru = osdmap.get_up_thru(from);
2919 auto ut = pending_inc.new_up_thru.find(from);
2920 if (ut != pending_inc.new_up_thru.end()) {
2921 old_up_thru = ut->second;
2923 if (up_thru > old_up_thru) {
2924 // set up_thru too, so the osd doesn't have to ask again
2925 pending_inc.new_up_thru[from] = up_thru;
2929 bool OSDMonitor::prepare_pgtemp(MonOpRequestRef op)
2931 op->mark_osdmon_event(__func__);
2932 MOSDPGTemp *m = static_cast<MOSDPGTemp*>(op->get_req());
2933 int from = m->get_orig_source().num();
2934 dout(7) << "prepare_pgtemp e" << m->map_epoch << " from " << m->get_orig_source_inst() << dendl;
2935 for (map<pg_t,vector<int32_t> >::iterator p = m->pg_temp.begin(); p != m->pg_temp.end(); ++p) {
2936 uint64_t pool = p->first.pool();
2937 if (pending_inc.old_pools.count(pool)) {
2938 dout(10) << __func__ << " ignore " << p->first << " -> " << p->second
2939 << ": pool pending removal" << dendl;
2942 if (!osdmap.have_pg_pool(pool)) {
2943 dout(10) << __func__ << " ignore " << p->first << " -> " << p->second
2944 << ": pool has been removed" << dendl;
2947 pending_inc.new_pg_temp[p->first] =
2948 mempool::osdmap::vector<int>(p->second.begin(), p->second.end());
2950 // unconditionally clear pg_primary (until this message can encode
2951 // a change for that, too.. at which point we need to also fix
2952 // preprocess_pg_temp)
2953 if (osdmap.primary_temp->count(p->first) ||
2954 pending_inc.new_primary_temp.count(p->first))
2955 pending_inc.new_primary_temp[p->first] = -1;
2958 // set up_thru too, so the osd doesn't have to ask again
2959 update_up_thru(from, m->map_epoch);
2961 wait_for_finished_proposal(op, new C_ReplyMap(this, op, m->map_epoch));
2968 bool OSDMonitor::preprocess_remove_snaps(MonOpRequestRef op)
2970 op->mark_osdmon_event(__func__);
2971 MRemoveSnaps *m = static_cast<MRemoveSnaps*>(op->get_req());
2972 dout(7) << "preprocess_remove_snaps " << *m << dendl;
2974 // check privilege, ignore if failed
2975 MonSession *session = m->get_session();
2978 if (!session->caps.is_capable(
2980 CEPH_ENTITY_TYPE_MON,
2981 session->entity_name,
2982 "osd", "osd pool rmsnap", {}, true, true, false)) {
2983 dout(0) << "got preprocess_remove_snaps from entity with insufficient caps "
2984 << session->caps << dendl;
2988 for (map<int, vector<snapid_t> >::iterator q = m->snaps.begin();
2989 q != m->snaps.end();
2991 if (!osdmap.have_pg_pool(q->first)) {
2992 dout(10) << " ignoring removed_snaps " << q->second << " on non-existent pool " << q->first << dendl;
2995 const pg_pool_t *pi = osdmap.get_pg_pool(q->first);
2996 for (vector<snapid_t>::iterator p = q->second.begin();
2997 p != q->second.end();
2999 if (*p > pi->get_snap_seq() ||
3000 !pi->removed_snaps.contains(*p))
3009 bool OSDMonitor::prepare_remove_snaps(MonOpRequestRef op)
3011 op->mark_osdmon_event(__func__);
3012 MRemoveSnaps *m = static_cast<MRemoveSnaps*>(op->get_req());
3013 dout(7) << "prepare_remove_snaps " << *m << dendl;
3015 for (map<int, vector<snapid_t> >::iterator p = m->snaps.begin();
3016 p != m->snaps.end();
3019 if (!osdmap.have_pg_pool(p->first)) {
3020 dout(10) << " ignoring removed_snaps " << p->second << " on non-existent pool " << p->first << dendl;
3024 pg_pool_t& pi = osdmap.pools[p->first];
3025 for (vector<snapid_t>::iterator q = p->second.begin();
3026 q != p->second.end();
3028 if (!pi.removed_snaps.contains(*q) &&
3029 (!pending_inc.new_pools.count(p->first) ||
3030 !pending_inc.new_pools[p->first].removed_snaps.contains(*q))) {
3031 pg_pool_t *newpi = pending_inc.get_new_pool(p->first, &pi);
3032 newpi->removed_snaps.insert(*q);
3033 dout(10) << " pool " << p->first << " removed_snaps added " << *q
3034 << " (now " << newpi->removed_snaps << ")" << dendl;
3035 if (*q > newpi->get_snap_seq()) {
3036 dout(10) << " pool " << p->first << " snap_seq " << newpi->get_snap_seq() << " -> " << *q << dendl;
3037 newpi->set_snap_seq(*q);
3039 newpi->set_snap_epoch(pending_inc.epoch);
3047 bool OSDMonitor::preprocess_beacon(MonOpRequestRef op)
3049 op->mark_osdmon_event(__func__);
3050 auto beacon = static_cast<MOSDBeacon*>(op->get_req());
3052 auto session = beacon->get_session();
3054 dout(10) << __func__ << " no monitor session!" << dendl;
3057 if (!session->is_capable("osd", MON_CAP_X)) {
3058 derr << __func__ << " received from entity "
3059 << "with insufficient privileges " << session->caps << dendl;
3062 // Always forward the beacon to the leader, even if they are the same as
3063 // the old one. The leader will mark as down osds that haven't sent
3064 // beacon for a few minutes.
3068 bool OSDMonitor::prepare_beacon(MonOpRequestRef op)
3070 op->mark_osdmon_event(__func__);
3071 const auto beacon = static_cast<MOSDBeacon*>(op->get_req());
3072 const auto src = beacon->get_orig_source();
3073 dout(10) << __func__ << " " << *beacon
3074 << " from " << src << dendl;
3075 int from = src.num();
3077 if (!src.is_osd() ||
3078 !osdmap.is_up(from) ||
3079 beacon->get_orig_source_inst() != osdmap.get_inst(from)) {
3080 dout(1) << " ignoring beacon from non-active osd." << dendl;
3084 last_osd_report[from] = ceph_clock_now();
3085 osd_epochs[from] = beacon->version;
3087 for (const auto& pg : beacon->pgs) {
3088 last_epoch_clean.report(pg, beacon->min_last_epoch_clean);
3096 void OSDMonitor::send_latest(MonOpRequestRef op, epoch_t start)
3098 op->mark_osdmon_event(__func__);
3099 dout(5) << "send_latest to " << op->get_req()->get_orig_source_inst()
3100 << " start " << start << dendl;
3104 send_incremental(op, start);
3108 MOSDMap *OSDMonitor::build_latest_full()
3110 MOSDMap *r = new MOSDMap(mon->monmap->fsid);
3111 get_version_full(osdmap.get_epoch(), r->maps[osdmap.get_epoch()]);
3112 r->oldest_map = get_first_committed();
3113 r->newest_map = osdmap.get_epoch();
3117 MOSDMap *OSDMonitor::build_incremental(epoch_t from, epoch_t to)
3119 dout(10) << "build_incremental [" << from << ".." << to << "]" << dendl;
3120 MOSDMap *m = new MOSDMap(mon->monmap->fsid);
3121 m->oldest_map = get_first_committed();
3122 m->newest_map = osdmap.get_epoch();
3124 for (epoch_t e = to; e >= from && e > 0; e--) {
3126 int err = get_version(e, bl);
3128 assert(bl.length());
3129 // if (get_version(e, bl) > 0) {
3130 dout(20) << "build_incremental inc " << e << " "
3131 << bl.length() << " bytes" << dendl;
3132 m->incremental_maps[e] = bl;
3134 assert(err == -ENOENT);
3135 assert(!bl.length());
3136 get_version_full(e, bl);
3137 if (bl.length() > 0) {
3138 //else if (get_version("full", e, bl) > 0) {
3139 dout(20) << "build_incremental full " << e << " "
3140 << bl.length() << " bytes" << dendl;
3143 ceph_abort(); // we should have all maps.
3150 void OSDMonitor::send_full(MonOpRequestRef op)
3152 op->mark_osdmon_event(__func__);
3153 dout(5) << "send_full to " << op->get_req()->get_orig_source_inst() << dendl;
3154 mon->send_reply(op, build_latest_full());
3157 void OSDMonitor::send_incremental(MonOpRequestRef op, epoch_t first)
3159 op->mark_osdmon_event(__func__);
3161 MonSession *s = op->get_session();
3165 s->proxy_con->has_feature(CEPH_FEATURE_MON_ROUTE_OSDMAP)) {
3166 // oh, we can tell the other mon to do it
3167 dout(10) << __func__ << " asking proxying mon to send_incremental from "
3169 MRoute *r = new MRoute(s->proxy_tid, NULL);
3170 r->send_osdmap_first = first;
3171 s->proxy_con->send_message(r);
3172 op->mark_event("reply: send routed send_osdmap_first reply");
3175 send_incremental(first, s, false, op);
3179 void OSDMonitor::send_incremental(epoch_t first,
3180 MonSession *session,
3182 MonOpRequestRef req)
3184 dout(5) << "send_incremental [" << first << ".." << osdmap.get_epoch() << "]"
3185 << " to " << session->inst << dendl;
3187 if (first <= session->osd_epoch) {
3188 dout(10) << __func__ << " " << session->inst << " should already have epoch "
3189 << session->osd_epoch << dendl;
3190 first = session->osd_epoch + 1;
3193 if (first < get_first_committed()) {
3194 first = get_first_committed();
3196 int err = get_version_full(first, bl);
3198 assert(bl.length());
3200 dout(20) << "send_incremental starting with base full "
3201 << first << " " << bl.length() << " bytes" << dendl;
3203 MOSDMap *m = new MOSDMap(osdmap.get_fsid());
3204 m->oldest_map = get_first_committed();
3205 m->newest_map = osdmap.get_epoch();
3206 m->maps[first] = bl;
3209 mon->send_reply(req, m);
3210 session->osd_epoch = first;
3213 session->con->send_message(m);
3214 session->osd_epoch = first;
3219 while (first <= osdmap.get_epoch()) {
3220 epoch_t last = MIN(first + g_conf->osd_map_message_max - 1,
3221 osdmap.get_epoch());
3222 MOSDMap *m = build_incremental(first, last);
3225 // send some maps. it may not be all of them, but it will get them
3227 mon->send_reply(req, m);
3229 session->con->send_message(m);
3232 session->osd_epoch = last;
3238 int OSDMonitor::get_version(version_t ver, bufferlist& bl)
3240 if (inc_osd_cache.lookup(ver, &bl)) {
3243 int ret = PaxosService::get_version(ver, bl);
3245 inc_osd_cache.add(ver, bl);
3250 int OSDMonitor::get_version_full(version_t ver, bufferlist& bl)
3252 if (full_osd_cache.lookup(ver, &bl)) {
3255 int ret = PaxosService::get_version_full(ver, bl);
3257 full_osd_cache.add(ver, bl);
3262 epoch_t OSDMonitor::blacklist(const entity_addr_t& a, utime_t until)
3264 dout(10) << "blacklist " << a << " until " << until << dendl;
3265 pending_inc.new_blacklist[a] = until;
3266 return pending_inc.epoch;
3270 void OSDMonitor::check_osdmap_subs()
3272 dout(10) << __func__ << dendl;
3273 if (!osdmap.get_epoch()) {
3276 auto osdmap_subs = mon->session_map.subs.find("osdmap");
3277 if (osdmap_subs == mon->session_map.subs.end()) {
3280 auto p = osdmap_subs->second->begin();
3284 check_osdmap_sub(sub);
3288 void OSDMonitor::check_osdmap_sub(Subscription *sub)
3290 dout(10) << __func__ << " " << sub << " next " << sub->next
3291 << (sub->onetime ? " (onetime)":" (ongoing)") << dendl;
3292 if (sub->next <= osdmap.get_epoch()) {
3294 send_incremental(sub->next, sub->session, sub->incremental_onetime);
3296 sub->session->con->send_message(build_latest_full());
3298 mon->session_map.remove_sub(sub);
3300 sub->next = osdmap.get_epoch() + 1;
3304 void OSDMonitor::check_pg_creates_subs()
3306 if (!mon->monmap->get_required_features().contains_all(
3307 ceph::features::mon::FEATURE_LUMINOUS)) {
3308 // PGMonitor takes care of this in pre-luminous era.
3311 if (!osdmap.get_num_up_osds()) {
3314 assert(osdmap.get_up_osd_features() & CEPH_FEATURE_MON_STATEFUL_SUB);
3315 mon->with_session_map([this](const MonSessionMap& session_map) {
3316 auto pg_creates_subs = session_map.subs.find("osd_pg_creates");
3317 if (pg_creates_subs == session_map.subs.end()) {
3320 for (auto sub : *pg_creates_subs->second) {
3321 check_pg_creates_sub(sub);
3326 void OSDMonitor::check_pg_creates_sub(Subscription *sub)
3328 dout(20) << __func__ << " .. " << sub->session->inst << dendl;
3329 assert(sub->type == "osd_pg_creates");
3330 // only send these if the OSD is up. we will check_subs() when they do
3331 // come up so they will get the creates then.
3332 if (sub->session->inst.name.is_osd() &&
3333 mon->osdmon()->osdmap.is_up(sub->session->inst.name.num())) {
3334 sub->next = send_pg_creates(sub->session->inst.name.num(),
3335 sub->session->con.get(),
3340 void OSDMonitor::do_application_enable(int64_t pool_id,
3341 const std::string &app_name)
3343 assert(paxos->is_plugged() && is_writeable());
3345 dout(20) << __func__ << ": pool_id=" << pool_id << ", app_name=" << app_name
3348 assert(osdmap.require_osd_release >= CEPH_RELEASE_LUMINOUS ||
3349 pending_inc.new_require_osd_release >= CEPH_RELEASE_LUMINOUS);
3351 auto pp = osdmap.get_pg_pool(pool_id);
3352 assert(pp != nullptr);
3355 if (pending_inc.new_pools.count(pool_id)) {
3356 p = pending_inc.new_pools[pool_id];
3359 p.application_metadata.insert({app_name, {}});
3360 p.last_change = pending_inc.epoch;
3361 pending_inc.new_pools[pool_id] = p;
3364 unsigned OSDMonitor::scan_for_creating_pgs(
3365 const mempool::osdmap::map<int64_t,pg_pool_t>& pools,
3366 const mempool::osdmap::set<int64_t>& removed_pools,
3368 creating_pgs_t* creating_pgs) const
3370 unsigned queued = 0;
3371 for (auto& p : pools) {
3372 int64_t poolid = p.first;
3373 const pg_pool_t& pool = p.second;
3374 int ruleno = osdmap.crush->find_rule(pool.get_crush_rule(),
3375 pool.get_type(), pool.get_size());
3376 if (ruleno < 0 || !osdmap.crush->rule_exists(ruleno))
3379 const auto last_scan_epoch = creating_pgs->last_scan_epoch;
3380 const auto created = pool.get_last_change();
3381 if (last_scan_epoch && created <= last_scan_epoch) {
3382 dout(10) << __func__ << " no change in pool " << poolid
3383 << " " << pool << dendl;
3386 if (removed_pools.count(poolid)) {
3387 dout(10) << __func__ << " pool is being removed: " << poolid
3388 << " " << pool << dendl;
3391 dout(10) << __func__ << " queueing pool create for " << poolid
3392 << " " << pool << dendl;
3393 if (creating_pgs->create_pool(poolid, pool.get_pg_num(),
3394 created, modified)) {
3401 void OSDMonitor::update_creating_pgs()
3403 dout(10) << __func__ << " " << creating_pgs.pgs.size() << " pgs creating, "
3404 << creating_pgs.queue.size() << " pools in queue" << dendl;
3405 decltype(creating_pgs_by_osd_epoch) new_pgs_by_osd_epoch;
3406 std::lock_guard<std::mutex> l(creating_pgs_lock);
3407 for (const auto& pg : creating_pgs.pgs) {
3408 int acting_primary = -1;
3409 auto pgid = pg.first;
3410 auto mapped = pg.second.first;
3411 dout(20) << __func__ << " looking up " << pgid << "@" << mapped << dendl;
3412 mapping.get(pgid, nullptr, nullptr, nullptr, &acting_primary);
3413 // check the previous creating_pgs, look for the target to whom the pg was
3414 // previously mapped
3415 for (const auto& pgs_by_epoch : creating_pgs_by_osd_epoch) {
3416 const auto last_acting_primary = pgs_by_epoch.first;
3417 for (auto& pgs: pgs_by_epoch.second) {
3418 if (pgs.second.count(pgid)) {
3419 if (last_acting_primary == acting_primary) {
3422 dout(20) << __func__ << " " << pgid << " "
3423 << " acting_primary:" << last_acting_primary
3424 << " -> " << acting_primary << dendl;
3425 // note epoch if the target of the create message changed.
3426 mapped = mapping.get_epoch();
3431 mapped = mapping.get_epoch();
3435 dout(10) << __func__ << " will instruct osd." << acting_primary
3436 << " to create " << pgid << "@" << mapped << dendl;
3437 new_pgs_by_osd_epoch[acting_primary][mapped].insert(pgid);
3439 creating_pgs_by_osd_epoch = std::move(new_pgs_by_osd_epoch);
3440 creating_pgs_epoch = mapping.get_epoch();
3443 epoch_t OSDMonitor::send_pg_creates(int osd, Connection *con, epoch_t next) const
3445 dout(30) << __func__ << " osd." << osd << " next=" << next
3446 << " " << creating_pgs_by_osd_epoch << dendl;
3447 std::lock_guard<std::mutex> l(creating_pgs_lock);
3448 if (creating_pgs_epoch <= creating_pgs.last_scan_epoch) {
3449 dout(20) << __func__
3450 << " not using stale creating_pgs@" << creating_pgs_epoch << dendl;
3451 // the subscribers will be updated when the mapping is completed anyway
3454 auto creating_pgs_by_epoch = creating_pgs_by_osd_epoch.find(osd);
3455 if (creating_pgs_by_epoch == creating_pgs_by_osd_epoch.end())
3457 assert(!creating_pgs_by_epoch->second.empty());
3459 MOSDPGCreate *m = nullptr;
3461 for (auto epoch_pgs = creating_pgs_by_epoch->second.lower_bound(next);
3462 epoch_pgs != creating_pgs_by_epoch->second.end(); ++epoch_pgs) {
3463 auto epoch = epoch_pgs->first;
3464 auto& pgs = epoch_pgs->second;
3465 dout(20) << __func__ << " osd." << osd << " from " << next
3466 << " : epoch " << epoch << " " << pgs.size() << " pgs" << dendl;
3468 for (auto& pg : pgs) {
3470 m = new MOSDPGCreate(creating_pgs_epoch);
3471 // Need the create time from the monitor using its clock to set
3472 // last_scrub_stamp upon pg creation.
3473 auto create = creating_pgs.pgs.find(pg);
3474 assert(create != creating_pgs.pgs.end());
3475 m->mkpg.emplace(pg, pg_create_t{create->second.first, pg, 0});
3476 m->ctimes.emplace(pg, create->second.second);
3477 dout(20) << __func__ << " will create " << pg
3478 << " at " << create->second.first << dendl;
3482 dout(20) << __func__ << " osd." << osd << " from " << next
3483 << " has nothing to send" << dendl;
3486 con->send_message(m);
3487 // sub is current through last + 1
3494 void OSDMonitor::tick()
3496 if (!is_active()) return;
3498 dout(10) << osdmap << dendl;
3500 if (!mon->is_leader()) return;
3502 bool do_propose = false;
3503 utime_t now = ceph_clock_now();
3505 if (osdmap.require_osd_release >= CEPH_RELEASE_LUMINOUS &&
3506 mon->monmap->get_required_features().contains_all(
3507 ceph::features::mon::FEATURE_LUMINOUS)) {
3508 if (handle_osd_timeouts(now, last_osd_report)) {
3512 if (!osdmap.test_flag(CEPH_OSDMAP_PURGED_SNAPDIRS) &&
3513 osdmap.require_osd_release >= CEPH_RELEASE_LUMINOUS &&
3514 mon->mgrstatmon()->is_readable() &&
3515 mon->mgrstatmon()->definitely_converted_snapsets()) {
3516 dout(1) << __func__ << " all snapsets converted, setting purged_snapdirs"
3518 add_flag(CEPH_OSDMAP_PURGED_SNAPDIRS);
3523 if (check_failures(now))
3526 // mark down osds out?
3528 /* can_mark_out() checks if we can mark osds as being out. The -1 has no
3529 * influence at all. The decision is made based on the ratio of "in" osds,
3530 * and the function returns false if this ratio is lower that the minimum
3531 * ratio set by g_conf->mon_osd_min_in_ratio. So it's not really up to us.
3533 if (can_mark_out(-1)) {
3534 set<int> down_cache; // quick cache of down subtrees
3536 map<int,utime_t>::iterator i = down_pending_out.begin();
3537 while (i != down_pending_out.end()) {
3543 if (osdmap.is_down(o) &&
3546 utime_t orig_grace(g_conf->mon_osd_down_out_interval, 0);
3547 utime_t grace = orig_grace;
3548 double my_grace = 0.0;
3550 if (g_conf->mon_osd_adjust_down_out_interval) {
3551 // scale grace period the same way we do the heartbeat grace.
3552 const osd_xinfo_t& xi = osdmap.get_xinfo(o);
3553 double halflife = (double)g_conf->mon_osd_laggy_halflife;
3554 double decay_k = ::log(.5) / halflife;
3555 double decay = exp((double)down * decay_k);
3556 dout(20) << "osd." << o << " laggy halflife " << halflife << " decay_k " << decay_k
3557 << " down for " << down << " decay " << decay << dendl;
3558 my_grace = decay * (double)xi.laggy_interval * xi.laggy_probability;
3562 // is this an entire large subtree down?
3563 if (g_conf->mon_osd_down_out_subtree_limit.length()) {
3564 int type = osdmap.crush->get_type_id(g_conf->mon_osd_down_out_subtree_limit);
3566 if (osdmap.containing_subtree_is_down(g_ceph_context, o, type, &down_cache)) {
3567 dout(10) << "tick entire containing " << g_conf->mon_osd_down_out_subtree_limit
3568 << " subtree for osd." << o << " is down; resetting timer" << dendl;
3569 // reset timer, too.
3570 down_pending_out[o] = now;
3576 bool down_out = !osdmap.is_destroyed(o) &&
3577 g_conf->mon_osd_down_out_interval > 0 && down.sec() >= grace;
3578 bool destroyed_out = osdmap.is_destroyed(o) &&
3579 g_conf->mon_osd_destroyed_out_interval > 0 &&
3580 // this is not precise enough as we did not make a note when this osd
3581 // was marked as destroyed, but let's not bother with that
3582 // complexity for now.
3583 down.sec() >= g_conf->mon_osd_destroyed_out_interval;
3584 if (down_out || destroyed_out) {
3585 dout(10) << "tick marking osd." << o << " OUT after " << down
3586 << " sec (target " << grace << " = " << orig_grace << " + " << my_grace << ")" << dendl;
3587 pending_inc.new_weight[o] = CEPH_OSD_OUT;
3589 // set the AUTOOUT bit.
3590 if (pending_inc.new_state.count(o) == 0)
3591 pending_inc.new_state[o] = 0;
3592 pending_inc.new_state[o] |= CEPH_OSD_AUTOOUT;
3594 // remember previous weight
3595 if (pending_inc.new_xinfo.count(o) == 0)
3596 pending_inc.new_xinfo[o] = osdmap.osd_xinfo[o];
3597 pending_inc.new_xinfo[o].old_weight = osdmap.osd_weight[o];
3601 mon->clog->info() << "Marking osd." << o << " out (has been down for "
3602 << int(down.sec()) << " seconds)";
3607 down_pending_out.erase(o);
3610 dout(10) << "tick NOOUT flag set, not checking down osds" << dendl;
3613 // expire blacklisted items?
3614 for (ceph::unordered_map<entity_addr_t,utime_t>::iterator p = osdmap.blacklist.begin();
3615 p != osdmap.blacklist.end();
3617 if (p->second < now) {
3618 dout(10) << "expiring blacklist item " << p->first << " expired " << p->second << " < now " << now << dendl;
3619 pending_inc.old_blacklist.push_back(p->first);
3624 // if map full setting has changed, get that info out there!
3625 if (osdmap.require_osd_release < CEPH_RELEASE_LUMINOUS &&
3626 mon->pgservice->is_readable()) {
3627 // for pre-luminous compat only!
3628 if (mon->pgservice->have_full_osds()) {
3629 dout(5) << "There are full osds, setting full flag" << dendl;
3630 add_flag(CEPH_OSDMAP_FULL);
3631 } else if (osdmap.test_flag(CEPH_OSDMAP_FULL)){
3632 dout(10) << "No full osds, removing full flag" << dendl;
3633 remove_flag(CEPH_OSDMAP_FULL);
3636 if (mon->pgservice->have_nearfull_osds()) {
3637 dout(5) << "There are near full osds, setting nearfull flag" << dendl;
3638 add_flag(CEPH_OSDMAP_NEARFULL);
3639 } else if (osdmap.test_flag(CEPH_OSDMAP_NEARFULL)){
3640 dout(10) << "No near full osds, removing nearfull flag" << dendl;
3641 remove_flag(CEPH_OSDMAP_NEARFULL);
3643 if (pending_inc.new_flags != -1 &&
3644 (pending_inc.new_flags ^ osdmap.flags) & (CEPH_OSDMAP_FULL | CEPH_OSDMAP_NEARFULL)) {
3645 dout(1) << "New setting for" <<
3646 (pending_inc.new_flags & CEPH_OSDMAP_FULL ? " CEPH_OSDMAP_FULL" : "") <<
3647 (pending_inc.new_flags & CEPH_OSDMAP_NEARFULL ? " CEPH_OSDMAP_NEARFULL" : "")
3648 << " -- doing propose" << dendl;
3653 if (update_pools_status())
3657 !pending_inc.new_pg_temp.empty()) // also propose if we adjusted pg_temp
3661 bool OSDMonitor::handle_osd_timeouts(const utime_t &now,
3662 std::map<int,utime_t> &last_osd_report)
3664 utime_t timeo(g_conf->mon_osd_report_timeout, 0);
3665 if (now - mon->get_leader_since() < timeo) {
3666 // We haven't been the leader for long enough to consider OSD timeouts
3670 int max_osd = osdmap.get_max_osd();
3671 bool new_down = false;
3673 for (int i=0; i < max_osd; ++i) {
3674 dout(30) << __func__ << ": checking up on osd " << i << dendl;
3675 if (!osdmap.exists(i)) {
3676 last_osd_report.erase(i); // if any
3679 if (!osdmap.is_up(i))
3681 const std::map<int,utime_t>::const_iterator t = last_osd_report.find(i);
3682 if (t == last_osd_report.end()) {
3683 // it wasn't in the map; start the timer.
3684 last_osd_report[i] = now;
3685 } else if (can_mark_down(i)) {
3686 utime_t diff = now - t->second;
3688 mon->clog->info() << "osd." << i << " marked down after no beacon for "
3689 << diff << " seconds";
3690 derr << "no beacon from osd." << i << " since " << t->second
3691 << ", " << diff << " seconds ago. marking down" << dendl;
3692 pending_inc.new_state[i] = CEPH_OSD_UP;
3700 void OSDMonitor::get_health(list<pair<health_status_t,string> >& summary,
3701 list<pair<health_status_t,string> > *detail,
3702 CephContext *cct) const
3704 int num_osds = osdmap.get_num_osds();
3706 if (num_osds == 0) {
3707 summary.push_back(make_pair(HEALTH_ERR, "no osds"));
3709 int num_in_osds = 0;
3710 int num_down_in_osds = 0;
3712 set<int> down_in_osds;
3713 set<int> up_in_osds;
3714 set<int> subtree_up;
3715 unordered_map<int, set<int> > subtree_type_down;
3716 unordered_map<int, int> num_osds_subtree;
3717 int max_type = osdmap.crush->get_max_type_id();
3719 for (int i = 0; i < osdmap.get_max_osd(); i++) {
3720 if (!osdmap.exists(i)) {
3721 if (osdmap.crush->item_exists(i)) {
3726 if (osdmap.is_out(i))
3729 if (down_in_osds.count(i) || up_in_osds.count(i))
3731 if (!osdmap.is_up(i)) {
3732 down_in_osds.insert(i);
3735 for (int type = 0; type <= max_type; type++) {
3736 if (!osdmap.crush->get_type_name(type))
3738 int r = osdmap.crush->get_immediate_parent_id(current, &parent_id);
3741 // break early if this parent is already marked as up
3742 if (subtree_up.count(parent_id))
3744 type = osdmap.crush->get_bucket_type(parent_id);
3745 if (!osdmap.subtree_type_is_down(
3746 g_ceph_context, parent_id, type,
3747 &down_in_osds, &up_in_osds, &subtree_up, &subtree_type_down))
3749 current = parent_id;
3754 // calculate the number of down osds in each down subtree and
3755 // store it in num_osds_subtree
3756 for (int type = 1; type <= max_type; type++) {
3757 if (!osdmap.crush->get_type_name(type))
3759 for (auto j = subtree_type_down[type].begin();
3760 j != subtree_type_down[type].end();
3764 int num = osdmap.crush->get_children(*j, &children);
3765 num_osds_subtree[*j] = num;
3769 int num_children = osdmap.crush->get_children(*j, &children);
3770 if (num_children == 0)
3772 for (auto l = children.begin(); l != children.end(); ++l) {
3773 if (num_osds_subtree[*l] > 0) {
3774 num = num + num_osds_subtree[*l];
3777 num_osds_subtree[*j] = num;
3781 num_down_in_osds = down_in_osds.size();
3782 assert(num_down_in_osds <= num_in_osds);
3783 if (num_down_in_osds > 0) {
3784 // summary of down subtree types and osds
3785 for (int type = max_type; type > 0; type--) {
3786 if (!osdmap.crush->get_type_name(type))
3788 if (subtree_type_down[type].size() > 0) {
3790 ss << subtree_type_down[type].size() << " "
3791 << osdmap.crush->get_type_name(type);
3792 if (subtree_type_down[type].size() > 1) {
3795 int sum_down_osds = 0;
3796 for (auto j = subtree_type_down[type].begin();
3797 j != subtree_type_down[type].end();
3799 sum_down_osds = sum_down_osds + num_osds_subtree[*j];
3801 ss << " (" << sum_down_osds << " osds) down";
3802 summary.push_back(make_pair(HEALTH_WARN, ss.str()));
3806 ss << down_in_osds.size() << " osds down";
3807 summary.push_back(make_pair(HEALTH_WARN, ss.str()));
3810 // details of down subtree types
3811 for (int type = max_type; type > 0; type--) {
3812 if (!osdmap.crush->get_type_name(type))
3814 for (auto j = subtree_type_down[type].rbegin();
3815 j != subtree_type_down[type].rend();
3818 ss << osdmap.crush->get_type_name(type);
3820 ss << osdmap.crush->get_item_name(*j);
3821 // at the top level, do not print location
3822 if (type != max_type) {
3824 ss << osdmap.crush->get_full_location_ordered_string(*j);
3827 int num = num_osds_subtree[*j];
3828 ss << " (" << num << " osds)";
3830 detail->push_back(make_pair(HEALTH_WARN, ss.str()));
3833 // details of down osds
3834 for (auto it = down_in_osds.begin(); it != down_in_osds.end(); ++it) {
3836 ss << "osd." << *it << " (";
3837 ss << osdmap.crush->get_full_location_ordered_string(*it);
3839 detail->push_back(make_pair(HEALTH_WARN, ss.str()));
3844 if (!osds.empty()) {
3846 ss << osds.size() << " osds exist in the crush map but not in the osdmap";
3847 summary.push_back(make_pair(HEALTH_WARN, ss.str()));
3849 ss << " (osds: " << osds << ")";
3850 detail->push_back(make_pair(HEALTH_WARN, ss.str()));
3854 // note: we leave it to ceph-mgr to generate details health warnings
3855 // with actual osd utilizations
3858 uint64_t warn_flags =
3860 CEPH_OSDMAP_PAUSERD |
3861 CEPH_OSDMAP_PAUSEWR |
3862 CEPH_OSDMAP_PAUSEREC |
3864 CEPH_OSDMAP_NODOWN |
3867 CEPH_OSDMAP_NOBACKFILL |
3868 CEPH_OSDMAP_NORECOVER |
3869 CEPH_OSDMAP_NOSCRUB |
3870 CEPH_OSDMAP_NODEEP_SCRUB |
3871 CEPH_OSDMAP_NOTIERAGENT |
3872 CEPH_OSDMAP_NOREBALANCE;
3873 if (osdmap.test_flag(warn_flags)) {
3875 ss << osdmap.get_flag_string(osdmap.get_flags() & warn_flags)
3877 summary.push_back(make_pair(HEALTH_WARN, ss.str()));
3879 detail->push_back(make_pair(HEALTH_WARN, ss.str()));
3882 // old crush tunables?
3883 if (g_conf->mon_warn_on_legacy_crush_tunables) {
3884 string min = osdmap.crush->get_min_required_version();
3885 if (min < g_conf->mon_crush_min_required_version) {
3887 ss << "crush map has legacy tunables (require " << min
3888 << ", min is " << g_conf->mon_crush_min_required_version << ")";
3889 summary.push_back(make_pair(HEALTH_WARN, ss.str()));
3891 ss << "; see http://docs.ceph.com/docs/master/rados/operations/crush-map/#tunables";
3892 detail->push_back(make_pair(HEALTH_WARN, ss.str()));
3896 if (g_conf->mon_warn_on_crush_straw_calc_version_zero) {
3897 if (osdmap.crush->get_straw_calc_version() == 0) {
3899 ss << "crush map has straw_calc_version=0";
3900 summary.push_back(make_pair(HEALTH_WARN, ss.str()));
3902 ss << "; see http://docs.ceph.com/docs/master/rados/operations/crush-map/#tunables";
3903 detail->push_back(make_pair(HEALTH_WARN, ss.str()));
3908 // hit_set-less cache_mode?
3909 if (g_conf->mon_warn_on_cache_pools_without_hit_sets) {
3910 int problem_cache_pools = 0;
3911 for (map<int64_t, pg_pool_t>::const_iterator p = osdmap.pools.begin();
3912 p != osdmap.pools.end();
3914 const pg_pool_t& info = p->second;
3915 if (info.cache_mode_requires_hit_set() &&
3916 info.hit_set_params.get_type() == HitSet::TYPE_NONE) {
3917 ++problem_cache_pools;
3920 ss << "pool '" << osdmap.get_pool_name(p->first)
3921 << "' with cache_mode " << info.get_cache_mode_name()
3922 << " needs hit_set_type to be set but it is not";
3923 detail->push_back(make_pair(HEALTH_WARN, ss.str()));
3927 if (problem_cache_pools) {
3929 ss << problem_cache_pools << " cache pools are missing hit_sets";
3930 summary.push_back(make_pair(HEALTH_WARN, ss.str()));
3934 // Not using 'sortbitwise' and should be?
3935 if (!osdmap.test_flag(CEPH_OSDMAP_SORTBITWISE) &&
3936 (osdmap.get_up_osd_features() &
3937 CEPH_FEATURE_OSD_BITWISE_HOBJ_SORT)) {
3939 ss << "no legacy OSD present but 'sortbitwise' flag is not set";
3940 summary.push_back(make_pair(HEALTH_WARN, ss.str()));
3943 // Warn if 'mon_osd_down_out_interval' is set to zero.
3944 // Having this option set to zero on the leader acts much like the
3945 // 'noout' flag. It's hard to figure out what's going wrong with clusters
3946 // without the 'noout' flag set but acting like that just the same, so
3947 // we report a HEALTH_WARN in case this option is set to zero.
3948 // This is an ugly hack to get the warning out, but until we find a way
3949 // to spread global options throughout the mon cluster and have all mons
3950 // using a base set of the same options, we need to work around this sort
3952 // There's also the obvious drawback that if this is set on a single
3953 // monitor on a 3-monitor cluster, this warning will only be shown every
3954 // third monitor connection.
3955 if (g_conf->mon_warn_on_osd_down_out_interval_zero &&
3956 g_conf->mon_osd_down_out_interval == 0) {
3958 ss << "mon." << mon->name << " has mon_osd_down_out_interval set to 0";
3959 summary.push_back(make_pair(HEALTH_WARN, ss.str()));
3961 ss << "; this has the same effect as the 'noout' flag";
3962 detail->push_back(make_pair(HEALTH_WARN, ss.str()));
3966 // warn about upgrade flags that can be set but are not.
3967 if (g_conf->mon_debug_no_require_luminous) {
3968 // ignore these checks
3969 } else if (HAVE_FEATURE(osdmap.get_up_osd_features(), SERVER_LUMINOUS) &&
3970 osdmap.require_osd_release < CEPH_RELEASE_LUMINOUS) {
3971 string msg = "all OSDs are running luminous or later but"
3972 " require_osd_release < luminous";
3973 summary.push_back(make_pair(HEALTH_WARN, msg));
3975 detail->push_back(make_pair(HEALTH_WARN, msg));
3977 } else if (HAVE_FEATURE(osdmap.get_up_osd_features(), SERVER_KRAKEN) &&
3978 osdmap.require_osd_release < CEPH_RELEASE_KRAKEN) {
3979 string msg = "all OSDs are running kraken or later but"
3980 " require_osd_release < kraken";
3981 summary.push_back(make_pair(HEALTH_WARN, msg));
3983 detail->push_back(make_pair(HEALTH_WARN, msg));
3985 } else if (HAVE_FEATURE(osdmap.get_up_osd_features(), SERVER_JEWEL) &&
3986 osdmap.require_osd_release < CEPH_RELEASE_JEWEL) {
3987 string msg = "all OSDs are running jewel or later but"
3988 " require_osd_release < jewel";
3989 summary.push_back(make_pair(HEALTH_WARN, msg));
3991 detail->push_back(make_pair(HEALTH_WARN, msg));
3995 for (auto it : osdmap.get_pools()) {
3996 const pg_pool_t &pool = it.second;
3997 if (pool.has_flag(pg_pool_t::FLAG_FULL)) {
3998 const string& pool_name = osdmap.get_pool_name(it.first);
4000 ss << "pool '" << pool_name << "' is full";
4001 summary.push_back(make_pair(HEALTH_WARN, ss.str()));
4003 detail->push_back(make_pair(HEALTH_WARN, ss.str()));
4009 void OSDMonitor::dump_info(Formatter *f)
4011 f->open_object_section("osdmap");
4015 f->open_array_section("osd_metadata");
4016 for (int i=0; i<osdmap.get_max_osd(); ++i) {
4017 if (osdmap.exists(i)) {
4018 f->open_object_section("osd");
4019 f->dump_unsigned("id", i);
4020 dump_osd_metadata(i, f, NULL);
4026 f->dump_unsigned("osdmap_first_committed", get_first_committed());
4027 f->dump_unsigned("osdmap_last_committed", get_last_committed());
4029 f->open_object_section("crushmap");
4030 osdmap.crush->dump(f);
4035 enum osd_pool_get_choices {
4036 SIZE, MIN_SIZE, CRASH_REPLAY_INTERVAL,
4037 PG_NUM, PGP_NUM, CRUSH_RULE, HASHPSPOOL,
4038 NODELETE, NOPGCHANGE, NOSIZECHANGE,
4039 WRITE_FADVISE_DONTNEED, NOSCRUB, NODEEP_SCRUB,
4040 HIT_SET_TYPE, HIT_SET_PERIOD, HIT_SET_COUNT, HIT_SET_FPP,
4041 USE_GMT_HITSET, AUID, TARGET_MAX_OBJECTS, TARGET_MAX_BYTES,
4042 CACHE_TARGET_DIRTY_RATIO, CACHE_TARGET_DIRTY_HIGH_RATIO,
4043 CACHE_TARGET_FULL_RATIO,
4044 CACHE_MIN_FLUSH_AGE, CACHE_MIN_EVICT_AGE,
4045 ERASURE_CODE_PROFILE, MIN_READ_RECENCY_FOR_PROMOTE,
4046 MIN_WRITE_RECENCY_FOR_PROMOTE, FAST_READ,
4047 HIT_SET_GRADE_DECAY_RATE, HIT_SET_SEARCH_LAST_N,
4048 SCRUB_MIN_INTERVAL, SCRUB_MAX_INTERVAL, DEEP_SCRUB_INTERVAL,
4049 RECOVERY_PRIORITY, RECOVERY_OP_PRIORITY, SCRUB_PRIORITY,
4050 COMPRESSION_MODE, COMPRESSION_ALGORITHM, COMPRESSION_REQUIRED_RATIO,
4051 COMPRESSION_MAX_BLOB_SIZE, COMPRESSION_MIN_BLOB_SIZE,
4052 CSUM_TYPE, CSUM_MAX_BLOCK, CSUM_MIN_BLOCK };
4054 std::set<osd_pool_get_choices>
4055 subtract_second_from_first(const std::set<osd_pool_get_choices>& first,
4056 const std::set<osd_pool_get_choices>& second)
4058 std::set<osd_pool_get_choices> result;
4059 std::set_difference(first.begin(), first.end(),
4060 second.begin(), second.end(),
4061 std::inserter(result, result.end()));
4067 bool OSDMonitor::preprocess_command(MonOpRequestRef op)
4069 op->mark_osdmon_event(__func__);
4070 MMonCommand *m = static_cast<MMonCommand*>(op->get_req());
4073 stringstream ss, ds;
4075 map<string, cmd_vartype> cmdmap;
4076 if (!cmdmap_from_json(m->cmd, &cmdmap, ss)) {
4077 string rs = ss.str();
4078 mon->reply_command(op, -EINVAL, rs, get_last_committed());
4082 MonSession *session = m->get_session();
4084 mon->reply_command(op, -EACCES, "access denied", get_last_committed());
4089 cmd_getval(g_ceph_context, cmdmap, "prefix", prefix);
4092 cmd_getval(g_ceph_context, cmdmap, "format", format, string("plain"));
4093 boost::scoped_ptr<Formatter> f(Formatter::create(format));
4095 if (prefix == "osd stat") {
4096 osdmap.print_summary(f.get(), ds, "");
4102 else if (prefix == "osd perf" ||
4103 prefix == "osd blocked-by") {
4104 r = mon->pgservice->process_pg_command(prefix, cmdmap,
4105 osdmap, f.get(), &ss, &rdata);
4107 else if (prefix == "osd dump" ||
4108 prefix == "osd tree" ||
4109 prefix == "osd ls" ||
4110 prefix == "osd getmap" ||
4111 prefix == "osd getcrushmap" ||
4112 prefix == "osd ls-tree") {
4117 cmd_getval(g_ceph_context, cmdmap, "epoch", epochnum, (int64_t)osdmap.get_epoch());
4120 bufferlist osdmap_bl;
4121 int err = get_version_full(epoch, osdmap_bl);
4122 if (err == -ENOENT) {
4124 ss << "there is no map for epoch " << epoch;
4128 assert(osdmap_bl.length());
4131 if (epoch == osdmap.get_epoch()) {
4135 p->decode(osdmap_bl);
4138 auto sg = make_scope_guard([&] {
4144 if (prefix == "osd dump") {
4147 f->open_object_section("osdmap");
4157 } else if (prefix == "osd ls") {
4159 f->open_array_section("osds");
4160 for (int i = 0; i < osdmap.get_max_osd(); i++) {
4161 if (osdmap.exists(i)) {
4162 f->dump_int("osd", i);
4169 for (int i = 0; i < osdmap.get_max_osd(); i++) {
4170 if (osdmap.exists(i)) {
4179 } else if (prefix == "osd tree") {
4180 vector<string> states;
4181 cmd_getval(g_ceph_context, cmdmap, "states", states);
4182 unsigned filter = 0;
4183 for (auto& s : states) {
4185 filter |= OSDMap::DUMP_UP;
4186 } else if (s == "down") {
4187 filter |= OSDMap::DUMP_DOWN;
4188 } else if (s == "in") {
4189 filter |= OSDMap::DUMP_IN;
4190 } else if (s == "out") {
4191 filter |= OSDMap::DUMP_OUT;
4192 } else if (s == "destroyed") {
4193 filter |= OSDMap::DUMP_DESTROYED;
4195 ss << "unrecognized state '" << s << "'";
4200 if ((filter & (OSDMap::DUMP_IN|OSDMap::DUMP_OUT)) ==
4201 (OSDMap::DUMP_IN|OSDMap::DUMP_OUT)) {
4202 ss << "cannot specify both 'in' and 'out'";
4206 if (((filter & (OSDMap::DUMP_UP|OSDMap::DUMP_DOWN)) ==
4207 (OSDMap::DUMP_UP|OSDMap::DUMP_DOWN)) ||
4208 ((filter & (OSDMap::DUMP_UP|OSDMap::DUMP_DESTROYED)) ==
4209 (OSDMap::DUMP_UP|OSDMap::DUMP_DESTROYED)) ||
4210 ((filter & (OSDMap::DUMP_DOWN|OSDMap::DUMP_DESTROYED)) ==
4211 (OSDMap::DUMP_DOWN|OSDMap::DUMP_DESTROYED))) {
4212 ss << "can specify only one of 'up', 'down' and 'destroyed'";
4217 f->open_object_section("tree");
4218 p->print_tree(f.get(), NULL, filter);
4222 p->print_tree(NULL, &ds, filter);
4225 } else if (prefix == "osd getmap") {
4226 rdata.append(osdmap_bl);
4227 ss << "got osdmap epoch " << p->get_epoch();
4228 } else if (prefix == "osd getcrushmap") {
4229 p->crush->encode(rdata, mon->get_quorum_con_features());
4230 ss << p->get_crush_version();
4231 } else if (prefix == "osd ls-tree") {
4233 cmd_getval(g_ceph_context, cmdmap, "name", bucket_name);
4235 r = p->get_osds_by_bucket_name(bucket_name, &osds);
4237 ss << "\"" << bucket_name << "\" does not exist";
4240 ss << "can not parse bucket name:\"" << bucket_name << "\"";
4245 f->open_array_section("osds");
4246 for (auto &i : osds) {
4247 if (osdmap.exists(i)) {
4248 f->dump_int("osd", i);
4255 for (auto &i : osds) {
4256 if (osdmap.exists(i)) {
4267 } else if (prefix == "osd df") {
4269 cmd_getval(g_ceph_context, cmdmap, "output_method", method);
4270 print_osd_utilization(osdmap, mon->pgservice, ds,
4271 f.get(), method == "tree");
4273 } else if (prefix == "osd getmaxosd") {
4275 f->open_object_section("getmaxosd");
4276 f->dump_unsigned("epoch", osdmap.get_epoch());
4277 f->dump_int("max_osd", osdmap.get_max_osd());
4281 ds << "max_osd = " << osdmap.get_max_osd() << " in epoch " << osdmap.get_epoch();
4284 } else if (prefix == "osd utilization") {
4286 osdmap.summarize_mapping_stats(NULL, NULL, &out, f.get());
4293 } else if (prefix == "osd find") {
4295 if (!cmd_getval(g_ceph_context, cmdmap, "id", osd)) {
4296 ss << "unable to parse osd id value '"
4297 << cmd_vartype_stringify(cmdmap["id"]) << "'";
4301 if (!osdmap.exists(osd)) {
4302 ss << "osd." << osd << " does not exist";
4307 cmd_getval(g_ceph_context, cmdmap, "format", format);
4308 boost::scoped_ptr<Formatter> f(Formatter::create(format, "json-pretty", "json-pretty"));
4309 f->open_object_section("osd_location");
4310 f->dump_int("osd", osd);
4311 f->dump_stream("ip") << osdmap.get_addr(osd);
4312 f->open_object_section("crush_location");
4313 map<string,string> loc = osdmap.crush->get_full_location(osd);
4314 for (map<string,string>::iterator p = loc.begin(); p != loc.end(); ++p)
4315 f->dump_string(p->first.c_str(), p->second);
4319 } else if (prefix == "osd metadata") {
4321 if (cmd_vartype_stringify(cmdmap["id"]).size() &&
4322 !cmd_getval(g_ceph_context, cmdmap, "id", osd)) {
4323 ss << "unable to parse osd id value '"
4324 << cmd_vartype_stringify(cmdmap["id"]) << "'";
4328 if (osd >= 0 && !osdmap.exists(osd)) {
4329 ss << "osd." << osd << " does not exist";
4334 cmd_getval(g_ceph_context, cmdmap, "format", format);
4335 boost::scoped_ptr<Formatter> f(Formatter::create(format, "json-pretty", "json-pretty"));
4337 f->open_object_section("osd_metadata");
4338 f->dump_unsigned("id", osd);
4339 r = dump_osd_metadata(osd, f.get(), &ss);
4345 f->open_array_section("osd_metadata");
4346 for (int i=0; i<osdmap.get_max_osd(); ++i) {
4347 if (osdmap.exists(i)) {
4348 f->open_object_section("osd");
4349 f->dump_unsigned("id", i);
4350 r = dump_osd_metadata(i, f.get(), NULL);
4351 if (r == -EINVAL || r == -ENOENT) {
4352 // Drop error, continue to get other daemons' metadata
4353 dout(4) << "No metadata for osd." << i << dendl;
4365 } else if (prefix == "osd versions") {
4367 f.reset(Formatter::create("json-pretty"));
4368 count_metadata("ceph_version", f.get());
4371 } else if (prefix == "osd count-metadata") {
4373 f.reset(Formatter::create("json-pretty"));
4375 cmd_getval(g_ceph_context, cmdmap, "property", field);
4376 count_metadata(field, f.get());
4379 } else if (prefix == "osd map") {
4380 string poolstr, objstr, namespacestr;
4381 cmd_getval(g_ceph_context, cmdmap, "pool", poolstr);
4382 cmd_getval(g_ceph_context, cmdmap, "object", objstr);
4383 cmd_getval(g_ceph_context, cmdmap, "nspace", namespacestr);
4385 int64_t pool = osdmap.lookup_pg_pool_name(poolstr.c_str());
4387 ss << "pool " << poolstr << " does not exist";
4391 object_locator_t oloc(pool, namespacestr);
4392 object_t oid(objstr);
4393 pg_t pgid = osdmap.object_locator_to_pg(oid, oloc);
4394 pg_t mpgid = osdmap.raw_pg_to_pg(pgid);
4395 vector<int> up, acting;
4397 osdmap.pg_to_up_acting_osds(mpgid, &up, &up_p, &acting, &acting_p);
4400 if (!namespacestr.empty())
4401 fullobjname = namespacestr + string("/") + oid.name;
4403 fullobjname = oid.name;
4405 f->open_object_section("osd_map");
4406 f->dump_unsigned("epoch", osdmap.get_epoch());
4407 f->dump_string("pool", poolstr);
4408 f->dump_int("pool_id", pool);
4409 f->dump_stream("objname") << fullobjname;
4410 f->dump_stream("raw_pgid") << pgid;
4411 f->dump_stream("pgid") << mpgid;
4412 f->open_array_section("up");
4413 for (vector<int>::iterator p = up.begin(); p != up.end(); ++p)
4414 f->dump_int("osd", *p);
4416 f->dump_int("up_primary", up_p);
4417 f->open_array_section("acting");
4418 for (vector<int>::iterator p = acting.begin(); p != acting.end(); ++p)
4419 f->dump_int("osd", *p);
4421 f->dump_int("acting_primary", acting_p);
4422 f->close_section(); // osd_map
4425 ds << "osdmap e" << osdmap.get_epoch()
4426 << " pool '" << poolstr << "' (" << pool << ")"
4427 << " object '" << fullobjname << "' ->"
4428 << " pg " << pgid << " (" << mpgid << ")"
4429 << " -> up (" << pg_vector_string(up) << ", p" << up_p << ") acting ("
4430 << pg_vector_string(acting) << ", p" << acting_p << ")";
4434 } else if (prefix == "pg map") {
4437 cmd_getval(g_ceph_context, cmdmap, "pgid", pgidstr);
4438 if (!pgid.parse(pgidstr.c_str())) {
4439 ss << "invalid pgid '" << pgidstr << "'";
4443 vector<int> up, acting;
4444 if (!osdmap.have_pg_pool(pgid.pool())) {
4445 ss << "pg '" << pgidstr << "' does not exist";
4449 pg_t mpgid = osdmap.raw_pg_to_pg(pgid);
4450 osdmap.pg_to_up_acting_osds(pgid, up, acting);
4452 f->open_object_section("pg_map");
4453 f->dump_unsigned("epoch", osdmap.get_epoch());
4454 f->dump_stream("raw_pgid") << pgid;
4455 f->dump_stream("pgid") << mpgid;
4456 f->open_array_section("up");
4457 for (auto osd : up) {
4458 f->dump_int("up_osd", osd);
4461 f->open_array_section("acting");
4462 for (auto osd : acting) {
4463 f->dump_int("acting_osd", osd);
4469 ds << "osdmap e" << osdmap.get_epoch()
4470 << " pg " << pgid << " (" << mpgid << ")"
4471 << " -> up " << up << " acting " << acting;
4476 } else if (prefix == "osd scrub" ||
4477 prefix == "osd deep-scrub" ||
4478 prefix == "osd repair") {
4480 cmd_getval(g_ceph_context, cmdmap, "who", whostr);
4481 vector<string> pvec;
4482 get_str_vec(prefix, pvec);
4484 if (whostr == "*" || whostr == "all" || whostr == "any") {
4487 for (int i = 0; i < osdmap.get_max_osd(); i++)
4488 if (osdmap.is_up(i)) {
4489 ss << (c++ ? "," : "") << i;
4490 mon->try_send_message(new MOSDScrub(osdmap.get_fsid(),
4491 pvec.back() == "repair",
4492 pvec.back() == "deep-scrub"),
4493 osdmap.get_inst(i));
4496 ss << " instructed to " << pvec.back();
4498 long osd = parse_osd_id(whostr.c_str(), &ss);
4501 } else if (osdmap.is_up(osd)) {
4502 mon->try_send_message(new MOSDScrub(osdmap.get_fsid(),
4503 pvec.back() == "repair",
4504 pvec.back() == "deep-scrub"),
4505 osdmap.get_inst(osd));
4506 ss << "osd." << osd << " instructed to " << pvec.back();
4508 ss << "osd." << osd << " is not up";
4512 } else if (prefix == "osd lspools") {
4514 cmd_getval(g_ceph_context, cmdmap, "auid", auid, int64_t(0));
4516 f->open_array_section("pools");
4517 for (map<int64_t, pg_pool_t>::iterator p = osdmap.pools.begin();
4518 p != osdmap.pools.end();
4520 if (!auid || p->second.auid == (uint64_t)auid) {
4522 f->open_object_section("pool");
4523 f->dump_int("poolnum", p->first);
4524 f->dump_string("poolname", osdmap.pool_name[p->first]);
4527 ds << p->first << ' ' << osdmap.pool_name[p->first] << ',';
4536 } else if (prefix == "osd blacklist ls") {
4538 f->open_array_section("blacklist");
4540 for (ceph::unordered_map<entity_addr_t,utime_t>::iterator p = osdmap.blacklist.begin();
4541 p != osdmap.blacklist.end();
4544 f->open_object_section("entry");
4545 f->dump_stream("addr") << p->first;
4546 f->dump_stream("until") << p->second;
4551 ss << p->first << " " << p->second;
4561 ss << "listed " << osdmap.blacklist.size() << " entries";
4563 } else if (prefix == "osd pool ls") {
4565 cmd_getval(g_ceph_context, cmdmap, "detail", detail);
4566 if (!f && detail == "detail") {
4568 osdmap.print_pools(ss);
4569 rdata.append(ss.str());
4572 f->open_array_section("pools");
4573 for (map<int64_t,pg_pool_t>::const_iterator it = osdmap.get_pools().begin();
4574 it != osdmap.get_pools().end();
4577 if (detail == "detail") {
4578 f->open_object_section("pool");
4579 f->dump_string("pool_name", osdmap.get_pool_name(it->first));
4580 it->second.dump(f.get());
4583 f->dump_string("pool_name", osdmap.get_pool_name(it->first));
4586 rdata.append(osdmap.get_pool_name(it->first) + "\n");
4595 } else if (prefix == "osd crush get-tunable") {
4597 cmd_getval(g_ceph_context, cmdmap, "tunable", tunable);
4600 f->open_object_section("tunable");
4601 if (tunable == "straw_calc_version") {
4603 f->dump_int(tunable.c_str(), osdmap.crush->get_straw_calc_version());
4605 rss << osdmap.crush->get_straw_calc_version() << "\n";
4614 rdata.append(rss.str());
4618 } else if (prefix == "osd pool get") {
4620 cmd_getval(g_ceph_context, cmdmap, "pool", poolstr);
4621 int64_t pool = osdmap.lookup_pg_pool_name(poolstr.c_str());
4623 ss << "unrecognized pool '" << poolstr << "'";
4628 const pg_pool_t *p = osdmap.get_pg_pool(pool);
4630 cmd_getval(g_ceph_context, cmdmap, "var", var);
4632 typedef std::map<std::string, osd_pool_get_choices> choices_map_t;
4633 const choices_map_t ALL_CHOICES = {
4635 {"min_size", MIN_SIZE},
4636 {"crash_replay_interval", CRASH_REPLAY_INTERVAL},
4637 {"pg_num", PG_NUM}, {"pgp_num", PGP_NUM},
4638 {"crush_rule", CRUSH_RULE},
4639 {"hashpspool", HASHPSPOOL}, {"nodelete", NODELETE},
4640 {"nopgchange", NOPGCHANGE}, {"nosizechange", NOSIZECHANGE},
4641 {"noscrub", NOSCRUB}, {"nodeep-scrub", NODEEP_SCRUB},
4642 {"write_fadvise_dontneed", WRITE_FADVISE_DONTNEED},
4643 {"hit_set_type", HIT_SET_TYPE}, {"hit_set_period", HIT_SET_PERIOD},
4644 {"hit_set_count", HIT_SET_COUNT}, {"hit_set_fpp", HIT_SET_FPP},
4645 {"use_gmt_hitset", USE_GMT_HITSET},
4646 {"auid", AUID}, {"target_max_objects", TARGET_MAX_OBJECTS},
4647 {"target_max_bytes", TARGET_MAX_BYTES},
4648 {"cache_target_dirty_ratio", CACHE_TARGET_DIRTY_RATIO},
4649 {"cache_target_dirty_high_ratio", CACHE_TARGET_DIRTY_HIGH_RATIO},
4650 {"cache_target_full_ratio", CACHE_TARGET_FULL_RATIO},
4651 {"cache_min_flush_age", CACHE_MIN_FLUSH_AGE},
4652 {"cache_min_evict_age", CACHE_MIN_EVICT_AGE},
4653 {"erasure_code_profile", ERASURE_CODE_PROFILE},
4654 {"min_read_recency_for_promote", MIN_READ_RECENCY_FOR_PROMOTE},
4655 {"min_write_recency_for_promote", MIN_WRITE_RECENCY_FOR_PROMOTE},
4656 {"fast_read", FAST_READ},
4657 {"hit_set_grade_decay_rate", HIT_SET_GRADE_DECAY_RATE},
4658 {"hit_set_search_last_n", HIT_SET_SEARCH_LAST_N},
4659 {"scrub_min_interval", SCRUB_MIN_INTERVAL},
4660 {"scrub_max_interval", SCRUB_MAX_INTERVAL},
4661 {"deep_scrub_interval", DEEP_SCRUB_INTERVAL},
4662 {"recovery_priority", RECOVERY_PRIORITY},
4663 {"recovery_op_priority", RECOVERY_OP_PRIORITY},
4664 {"scrub_priority", SCRUB_PRIORITY},
4665 {"compression_mode", COMPRESSION_MODE},
4666 {"compression_algorithm", COMPRESSION_ALGORITHM},
4667 {"compression_required_ratio", COMPRESSION_REQUIRED_RATIO},
4668 {"compression_max_blob_size", COMPRESSION_MAX_BLOB_SIZE},
4669 {"compression_min_blob_size", COMPRESSION_MIN_BLOB_SIZE},
4670 {"csum_type", CSUM_TYPE},
4671 {"csum_max_block", CSUM_MAX_BLOCK},
4672 {"csum_min_block", CSUM_MIN_BLOCK},
4675 typedef std::set<osd_pool_get_choices> choices_set_t;
4677 const choices_set_t ONLY_TIER_CHOICES = {
4678 HIT_SET_TYPE, HIT_SET_PERIOD, HIT_SET_COUNT, HIT_SET_FPP,
4679 TARGET_MAX_OBJECTS, TARGET_MAX_BYTES, CACHE_TARGET_FULL_RATIO,
4680 CACHE_TARGET_DIRTY_RATIO, CACHE_TARGET_DIRTY_HIGH_RATIO,
4681 CACHE_MIN_FLUSH_AGE, CACHE_MIN_EVICT_AGE,
4682 MIN_READ_RECENCY_FOR_PROMOTE,
4683 MIN_WRITE_RECENCY_FOR_PROMOTE,
4684 HIT_SET_GRADE_DECAY_RATE, HIT_SET_SEARCH_LAST_N
4686 const choices_set_t ONLY_ERASURE_CHOICES = {
4687 ERASURE_CODE_PROFILE
4690 choices_set_t selected_choices;
4692 for(choices_map_t::const_iterator it = ALL_CHOICES.begin();
4693 it != ALL_CHOICES.end(); ++it) {
4694 selected_choices.insert(it->second);
4698 selected_choices = subtract_second_from_first(selected_choices,
4702 if(!p->is_erasure()) {
4703 selected_choices = subtract_second_from_first(selected_choices,
4704 ONLY_ERASURE_CHOICES);
4706 } else /* var != "all" */ {
4707 choices_map_t::const_iterator found = ALL_CHOICES.find(var);
4708 osd_pool_get_choices selected = found->second;
4710 if (!p->is_tier() &&
4711 ONLY_TIER_CHOICES.find(selected) != ONLY_TIER_CHOICES.end()) {
4712 ss << "pool '" << poolstr
4713 << "' is not a tier pool: variable not applicable";
4718 if (!p->is_erasure() &&
4719 ONLY_ERASURE_CHOICES.find(selected)
4720 != ONLY_ERASURE_CHOICES.end()) {
4721 ss << "pool '" << poolstr
4722 << "' is not a erasure pool: variable not applicable";
4727 selected_choices.insert(selected);
4731 for(choices_set_t::const_iterator it = selected_choices.begin();
4732 it != selected_choices.end(); ++it) {
4733 choices_map_t::const_iterator i;
4734 for (i = ALL_CHOICES.begin(); i != ALL_CHOICES.end(); ++i) {
4735 if (i->second == *it) {
4739 assert(i != ALL_CHOICES.end());
4740 bool pool_opt = pool_opts_t::is_opt_name(i->first);
4742 f->open_object_section("pool");
4743 f->dump_string("pool", poolstr);
4744 f->dump_int("pool_id", pool);
4748 f->dump_int("pg_num", p->get_pg_num());
4751 f->dump_int("pgp_num", p->get_pgp_num());
4754 f->dump_int("auid", p->get_auid());
4757 f->dump_int("size", p->get_size());
4760 f->dump_int("min_size", p->get_min_size());
4762 case CRASH_REPLAY_INTERVAL:
4763 f->dump_int("crash_replay_interval",
4764 p->get_crash_replay_interval());
4767 if (osdmap.crush->rule_exists(p->get_crush_rule())) {
4768 f->dump_string("crush_rule", osdmap.crush->get_rule_name(
4769 p->get_crush_rule()));
4771 f->dump_string("crush_rule", stringify(p->get_crush_rule()));
4778 case WRITE_FADVISE_DONTNEED:
4781 f->dump_string(i->first.c_str(),
4782 p->has_flag(pg_pool_t::get_flag_by_name(i->first)) ?
4785 case HIT_SET_PERIOD:
4786 f->dump_int("hit_set_period", p->hit_set_period);
4789 f->dump_int("hit_set_count", p->hit_set_count);
4792 f->dump_string("hit_set_type",
4793 HitSet::get_type_name(p->hit_set_params.get_type()));
4797 if (p->hit_set_params.get_type() == HitSet::TYPE_BLOOM) {
4798 BloomHitSet::Params *bloomp =
4799 static_cast<BloomHitSet::Params*>(p->hit_set_params.impl.get());
4800 f->dump_float("hit_set_fpp", bloomp->get_fpp());
4801 } else if(var != "all") {
4803 ss << "hit set is not of type Bloom; " <<
4804 "invalid to get a false positive rate!";
4810 case USE_GMT_HITSET:
4811 f->dump_bool("use_gmt_hitset", p->use_gmt_hitset);
4813 case TARGET_MAX_OBJECTS:
4814 f->dump_unsigned("target_max_objects", p->target_max_objects);
4816 case TARGET_MAX_BYTES:
4817 f->dump_unsigned("target_max_bytes", p->target_max_bytes);
4819 case CACHE_TARGET_DIRTY_RATIO:
4820 f->dump_unsigned("cache_target_dirty_ratio_micro",
4821 p->cache_target_dirty_ratio_micro);
4822 f->dump_float("cache_target_dirty_ratio",
4823 ((float)p->cache_target_dirty_ratio_micro/1000000));
4825 case CACHE_TARGET_DIRTY_HIGH_RATIO:
4826 f->dump_unsigned("cache_target_dirty_high_ratio_micro",
4827 p->cache_target_dirty_high_ratio_micro);
4828 f->dump_float("cache_target_dirty_high_ratio",
4829 ((float)p->cache_target_dirty_high_ratio_micro/1000000));
4831 case CACHE_TARGET_FULL_RATIO:
4832 f->dump_unsigned("cache_target_full_ratio_micro",
4833 p->cache_target_full_ratio_micro);
4834 f->dump_float("cache_target_full_ratio",
4835 ((float)p->cache_target_full_ratio_micro/1000000));
4837 case CACHE_MIN_FLUSH_AGE:
4838 f->dump_unsigned("cache_min_flush_age", p->cache_min_flush_age);
4840 case CACHE_MIN_EVICT_AGE:
4841 f->dump_unsigned("cache_min_evict_age", p->cache_min_evict_age);
4843 case ERASURE_CODE_PROFILE:
4844 f->dump_string("erasure_code_profile", p->erasure_code_profile);
4846 case MIN_READ_RECENCY_FOR_PROMOTE:
4847 f->dump_int("min_read_recency_for_promote",
4848 p->min_read_recency_for_promote);
4850 case MIN_WRITE_RECENCY_FOR_PROMOTE:
4851 f->dump_int("min_write_recency_for_promote",
4852 p->min_write_recency_for_promote);
4855 f->dump_int("fast_read", p->fast_read);
4857 case HIT_SET_GRADE_DECAY_RATE:
4858 f->dump_int("hit_set_grade_decay_rate",
4859 p->hit_set_grade_decay_rate);
4861 case HIT_SET_SEARCH_LAST_N:
4862 f->dump_int("hit_set_search_last_n",
4863 p->hit_set_search_last_n);
4865 case SCRUB_MIN_INTERVAL:
4866 case SCRUB_MAX_INTERVAL:
4867 case DEEP_SCRUB_INTERVAL:
4868 case RECOVERY_PRIORITY:
4869 case RECOVERY_OP_PRIORITY:
4870 case SCRUB_PRIORITY:
4871 case COMPRESSION_MODE:
4872 case COMPRESSION_ALGORITHM:
4873 case COMPRESSION_REQUIRED_RATIO:
4874 case COMPRESSION_MAX_BLOB_SIZE:
4875 case COMPRESSION_MIN_BLOB_SIZE:
4877 case CSUM_MAX_BLOCK:
4878 case CSUM_MIN_BLOCK:
4879 pool_opts_t::key_t key = pool_opts_t::get_opt_desc(i->first).key;
4880 if (p->opts.is_set(key)) {
4881 f->open_object_section("pool");
4882 f->dump_string("pool", poolstr);
4883 f->dump_int("pool_id", pool);
4884 if(*it == CSUM_TYPE) {
4886 p->opts.get(pool_opts_t::CSUM_TYPE, &val);
4887 f->dump_string(i->first.c_str(), Checksummer::get_csum_type_string(val));
4889 p->opts.dump(i->first, f.get());
4903 for(choices_set_t::const_iterator it = selected_choices.begin();
4904 it != selected_choices.end(); ++it) {
4905 choices_map_t::const_iterator i;
4908 ss << "pg_num: " << p->get_pg_num() << "\n";
4911 ss << "pgp_num: " << p->get_pgp_num() << "\n";
4914 ss << "auid: " << p->get_auid() << "\n";
4917 ss << "size: " << p->get_size() << "\n";
4920 ss << "min_size: " << p->get_min_size() << "\n";
4922 case CRASH_REPLAY_INTERVAL:
4923 ss << "crash_replay_interval: " <<
4924 p->get_crash_replay_interval() << "\n";
4927 if (osdmap.crush->rule_exists(p->get_crush_rule())) {
4928 ss << "crush_rule: " << osdmap.crush->get_rule_name(
4929 p->get_crush_rule()) << "\n";
4931 ss << "crush_rule: " << p->get_crush_rule() << "\n";
4934 case HIT_SET_PERIOD:
4935 ss << "hit_set_period: " << p->hit_set_period << "\n";
4938 ss << "hit_set_count: " << p->hit_set_count << "\n";
4941 ss << "hit_set_type: " <<
4942 HitSet::get_type_name(p->hit_set_params.get_type()) << "\n";
4946 if (p->hit_set_params.get_type() == HitSet::TYPE_BLOOM) {
4947 BloomHitSet::Params *bloomp =
4948 static_cast<BloomHitSet::Params*>(p->hit_set_params.impl.get());
4949 ss << "hit_set_fpp: " << bloomp->get_fpp() << "\n";
4950 } else if(var != "all") {
4951 ss << "hit set is not of type Bloom; " <<
4952 "invalid to get a false positive rate!";
4958 case USE_GMT_HITSET:
4959 ss << "use_gmt_hitset: " << p->use_gmt_hitset << "\n";
4961 case TARGET_MAX_OBJECTS:
4962 ss << "target_max_objects: " << p->target_max_objects << "\n";
4964 case TARGET_MAX_BYTES:
4965 ss << "target_max_bytes: " << p->target_max_bytes << "\n";
4967 case CACHE_TARGET_DIRTY_RATIO:
4968 ss << "cache_target_dirty_ratio: "
4969 << ((float)p->cache_target_dirty_ratio_micro/1000000) << "\n";
4971 case CACHE_TARGET_DIRTY_HIGH_RATIO:
4972 ss << "cache_target_dirty_high_ratio: "
4973 << ((float)p->cache_target_dirty_high_ratio_micro/1000000) << "\n";
4975 case CACHE_TARGET_FULL_RATIO:
4976 ss << "cache_target_full_ratio: "
4977 << ((float)p->cache_target_full_ratio_micro/1000000) << "\n";
4979 case CACHE_MIN_FLUSH_AGE:
4980 ss << "cache_min_flush_age: " << p->cache_min_flush_age << "\n";
4982 case CACHE_MIN_EVICT_AGE:
4983 ss << "cache_min_evict_age: " << p->cache_min_evict_age << "\n";
4985 case ERASURE_CODE_PROFILE:
4986 ss << "erasure_code_profile: " << p->erasure_code_profile << "\n";
4988 case MIN_READ_RECENCY_FOR_PROMOTE:
4989 ss << "min_read_recency_for_promote: " <<
4990 p->min_read_recency_for_promote << "\n";
4992 case HIT_SET_GRADE_DECAY_RATE:
4993 ss << "hit_set_grade_decay_rate: " <<
4994 p->hit_set_grade_decay_rate << "\n";
4996 case HIT_SET_SEARCH_LAST_N:
4997 ss << "hit_set_search_last_n: " <<
4998 p->hit_set_search_last_n << "\n";
5004 case WRITE_FADVISE_DONTNEED:
5007 for (i = ALL_CHOICES.begin(); i != ALL_CHOICES.end(); ++i) {
5008 if (i->second == *it)
5011 assert(i != ALL_CHOICES.end());
5012 ss << i->first << ": " <<
5013 (p->has_flag(pg_pool_t::get_flag_by_name(i->first)) ?
5014 "true" : "false") << "\n";
5016 case MIN_WRITE_RECENCY_FOR_PROMOTE:
5017 ss << "min_write_recency_for_promote: " <<
5018 p->min_write_recency_for_promote << "\n";
5021 ss << "fast_read: " << p->fast_read << "\n";
5023 case SCRUB_MIN_INTERVAL:
5024 case SCRUB_MAX_INTERVAL:
5025 case DEEP_SCRUB_INTERVAL:
5026 case RECOVERY_PRIORITY:
5027 case RECOVERY_OP_PRIORITY:
5028 case SCRUB_PRIORITY:
5029 case COMPRESSION_MODE:
5030 case COMPRESSION_ALGORITHM:
5031 case COMPRESSION_REQUIRED_RATIO:
5032 case COMPRESSION_MAX_BLOB_SIZE:
5033 case COMPRESSION_MIN_BLOB_SIZE:
5035 case CSUM_MAX_BLOCK:
5036 case CSUM_MIN_BLOCK:
5037 for (i = ALL_CHOICES.begin(); i != ALL_CHOICES.end(); ++i) {
5038 if (i->second == *it)
5041 assert(i != ALL_CHOICES.end());
5043 pool_opts_t::key_t key = pool_opts_t::get_opt_desc(i->first).key;
5044 if (p->opts.is_set(key)) {
5045 if(key == pool_opts_t::CSUM_TYPE) {
5047 p->opts.get(key, &val);
5048 ss << i->first << ": " << Checksummer::get_csum_type_string(val) << "\n";
5050 ss << i->first << ": " << p->opts.get(key) << "\n";
5056 rdata.append(ss.str());
5061 } else if (prefix == "osd pool stats") {
5062 r = mon->pgservice->process_pg_command(prefix, cmdmap,
5063 osdmap, f.get(), &ss, &rdata);
5064 } else if (prefix == "osd pool get-quota") {
5066 cmd_getval(g_ceph_context, cmdmap, "pool", pool_name);
5068 int64_t poolid = osdmap.lookup_pg_pool_name(pool_name);
5070 assert(poolid == -ENOENT);
5071 ss << "unrecognized pool '" << pool_name << "'";
5075 const pg_pool_t *p = osdmap.get_pg_pool(poolid);
5078 f->open_object_section("pool_quotas");
5079 f->dump_string("pool_name", pool_name);
5080 f->dump_unsigned("pool_id", poolid);
5081 f->dump_unsigned("quota_max_objects", p->quota_max_objects);
5082 f->dump_unsigned("quota_max_bytes", p->quota_max_bytes);
5087 rs << "quotas for pool '" << pool_name << "':\n"
5088 << " max objects: ";
5089 if (p->quota_max_objects == 0)
5092 rs << si_t(p->quota_max_objects) << " objects";
5095 if (p->quota_max_bytes == 0)
5098 rs << si_t(p->quota_max_bytes) << "B";
5099 rdata.append(rs.str());
5103 } else if (prefix == "osd crush rule list" ||
5104 prefix == "osd crush rule ls") {
5106 f->open_array_section("rules");
5107 osdmap.crush->list_rules(f.get());
5112 osdmap.crush->list_rules(&ss);
5113 rdata.append(ss.str());
5115 } else if (prefix == "osd crush rule ls-by-class") {
5117 cmd_getval(g_ceph_context, cmdmap, "class", class_name);
5118 if (class_name.empty()) {
5119 ss << "no class specified";
5124 r = osdmap.crush->get_rules_by_class(class_name, &rules);
5126 ss << "failed to get rules by class '" << class_name << "'";
5130 f->open_array_section("rules");
5131 for (auto &rule: rules) {
5132 f->dump_string("name", osdmap.crush->get_rule_name(rule));
5138 for (auto &rule: rules) {
5139 rs << osdmap.crush->get_rule_name(rule) << "\n";
5141 rdata.append(rs.str());
5143 } else if (prefix == "osd crush rule dump") {
5145 cmd_getval(g_ceph_context, cmdmap, "name", name);
5147 cmd_getval(g_ceph_context, cmdmap, "format", format);
5148 boost::scoped_ptr<Formatter> f(Formatter::create(format, "json-pretty", "json-pretty"));
5150 f->open_array_section("rules");
5151 osdmap.crush->dump_rules(f.get());
5154 int ruleno = osdmap.crush->get_rule_id(name);
5156 ss << "unknown crush rule '" << name << "'";
5160 osdmap.crush->dump_rule(ruleno, f.get());
5165 rdata.append(rs.str());
5166 } else if (prefix == "osd crush dump") {
5168 cmd_getval(g_ceph_context, cmdmap, "format", format);
5169 boost::scoped_ptr<Formatter> f(Formatter::create(format, "json-pretty", "json-pretty"));
5170 f->open_object_section("crush_map");
5171 osdmap.crush->dump(f.get());
5176 rdata.append(rs.str());
5177 } else if (prefix == "osd crush show-tunables") {
5179 cmd_getval(g_ceph_context, cmdmap, "format", format);
5180 boost::scoped_ptr<Formatter> f(Formatter::create(format, "json-pretty", "json-pretty"));
5181 f->open_object_section("crush_map_tunables");
5182 osdmap.crush->dump_tunables(f.get());
5187 rdata.append(rs.str());
5188 } else if (prefix == "osd crush tree") {
5190 cmd_getval(g_ceph_context, cmdmap, "shadow", shadow);
5191 bool show_shadow = shadow == "--show-shadow";
5192 boost::scoped_ptr<Formatter> f(Formatter::create(format));
5194 osdmap.crush->dump_tree(nullptr,
5196 osdmap.get_pool_names(),
5201 osdmap.crush->dump_tree(&ss,
5203 osdmap.get_pool_names(),
5205 rdata.append(ss.str());
5207 } else if (prefix == "osd crush ls") {
5209 if (!cmd_getval(g_ceph_context, cmdmap, "node", name)) {
5210 ss << "no node specified";
5214 if (!osdmap.crush->name_exists(name)) {
5215 ss << "node '" << name << "' does not exist";
5219 int id = osdmap.crush->get_item_id(name);
5222 result.push_back(id);
5224 int num = osdmap.crush->get_bucket_size(id);
5225 for (int i = 0; i < num; ++i) {
5226 result.push_back(osdmap.crush->get_bucket_item(id, i));
5230 f->open_array_section("items");
5231 for (auto i : result) {
5232 f->dump_string("item", osdmap.crush->get_item_name(i));
5238 for (auto i : result) {
5239 ss << osdmap.crush->get_item_name(i) << "\n";
5241 rdata.append(ss.str());
5244 } else if (prefix == "osd crush class ls") {
5245 boost::scoped_ptr<Formatter> f(Formatter::create(format, "json-pretty", "json-pretty"));
5246 f->open_array_section("crush_classes");
5247 for (auto i : osdmap.crush->class_name)
5248 f->dump_string("class", i.second);
5251 } else if (prefix == "osd crush class ls-osd") {
5253 cmd_getval(g_ceph_context, cmdmap, "class", name);
5255 osdmap.crush->get_devices_by_class(name, &osds);
5257 f->open_array_section("osds");
5258 for (auto &osd: osds)
5259 f->dump_int("osd", osd);
5264 for (auto &osd : osds) {
5272 } else if (prefix == "osd erasure-code-profile ls") {
5273 const auto &profiles = osdmap.get_erasure_code_profiles();
5275 f->open_array_section("erasure-code-profiles");
5276 for (auto i = profiles.begin(); i != profiles.end(); ++i) {
5278 f->dump_string("profile", i->first.c_str());
5280 rdata.append(i->first + "\n");
5287 rdata.append(rs.str());
5289 } else if (prefix == "osd crush weight-set ls") {
5290 boost::scoped_ptr<Formatter> f(Formatter::create(format));
5292 f->open_array_section("weight_sets");
5293 if (osdmap.crush->have_choose_args(CrushWrapper::DEFAULT_CHOOSE_ARGS)) {
5294 f->dump_string("pool", "(compat)");
5296 for (auto& i : osdmap.crush->choose_args) {
5298 f->dump_string("pool", osdmap.get_pool_name(i.first));
5305 if (osdmap.crush->have_choose_args(CrushWrapper::DEFAULT_CHOOSE_ARGS)) {
5308 for (auto& i : osdmap.crush->choose_args) {
5310 rs << osdmap.get_pool_name(i.first) << "\n";
5313 rdata.append(rs.str());
5315 } else if (prefix == "osd crush weight-set dump") {
5316 boost::scoped_ptr<Formatter> f(Formatter::create(format, "json-pretty",
5318 osdmap.crush->dump_choose_args(f.get());
5320 } else if (prefix == "osd erasure-code-profile get") {
5322 cmd_getval(g_ceph_context, cmdmap, "name", name);
5323 if (!osdmap.has_erasure_code_profile(name)) {
5324 ss << "unknown erasure code profile '" << name << "'";
5328 const map<string,string> &profile = osdmap.get_erasure_code_profile(name);
5330 f->open_object_section("profile");
5331 for (map<string,string>::const_iterator i = profile.begin();
5335 f->dump_string(i->first.c_str(), i->second.c_str());
5337 rdata.append(i->first + "=" + i->second + "\n");
5344 rdata.append(rs.str());
5346 } else if (prefix == "osd pool application get") {
5347 boost::scoped_ptr<Formatter> f(Formatter::create(format, "json-pretty",
5350 cmd_getval(g_ceph_context, cmdmap, "pool", pool_name);
5352 cmd_getval(g_ceph_context, cmdmap, "app", app);
5354 cmd_getval(g_ceph_context, cmdmap, "key", key);
5356 if (pool_name.empty()) {
5358 f->open_object_section("pools");
5359 for (const auto &pool : osdmap.pools) {
5360 std::string name("<unknown>");
5361 const auto &pni = osdmap.pool_name.find(pool.first);
5362 if (pni != osdmap.pool_name.end())
5364 f->open_object_section(name.c_str());
5365 for (auto &app_pair : pool.second.application_metadata) {
5366 f->open_object_section(app_pair.first.c_str());
5367 for (auto &kv_pair : app_pair.second) {
5368 f->dump_string(kv_pair.first.c_str(), kv_pair.second);
5372 f->close_section(); // name
5374 f->close_section(); // pools
5377 int64_t pool = osdmap.lookup_pg_pool_name(pool_name.c_str());
5379 ss << "unrecognized pool '" << pool_name << "'";
5383 auto p = osdmap.get_pg_pool(pool);
5386 f->open_object_section(pool_name.c_str());
5387 for (auto &app_pair : p->application_metadata) {
5388 f->open_object_section(app_pair.first.c_str());
5389 for (auto &kv_pair : app_pair.second) {
5390 f->dump_string(kv_pair.first.c_str(), kv_pair.second);
5392 f->close_section(); // application
5394 f->close_section(); // pool_name
5399 auto app_it = p->application_metadata.find(app);
5400 if (app_it == p->application_metadata.end()) {
5401 ss << "pool '" << pool_name << "' has no application '" << app << "'";
5405 // filter by pool + app
5407 f->open_object_section(app_it->first.c_str());
5408 for (auto &kv_pair : app_it->second) {
5409 f->dump_string(kv_pair.first.c_str(), kv_pair.second);
5411 f->close_section(); // application
5415 // filter by pool + app + key
5416 auto key_it = app_it->second.find(key);
5417 if (key_it == app_it->second.end()) {
5418 ss << "application '" << app << "' on pool '" << pool_name
5419 << "' does not have key '" << key << "'";
5423 ss << key_it->second << "\n";
5424 rdata.append(ss.str());
5428 // try prepare update
5435 mon->reply_command(op, r, rs, rdata, get_last_committed());
5439 void OSDMonitor::set_pool_flags(int64_t pool_id, uint64_t flags)
5441 pg_pool_t *pool = pending_inc.get_new_pool(pool_id,
5442 osdmap.get_pg_pool(pool_id));
5444 pool->set_flag(flags);
5447 void OSDMonitor::clear_pool_flags(int64_t pool_id, uint64_t flags)
5449 pg_pool_t *pool = pending_inc.get_new_pool(pool_id,
5450 osdmap.get_pg_pool(pool_id));
5452 pool->unset_flag(flags);
5455 bool OSDMonitor::update_pools_status()
5457 if (!mon->pgservice->is_readable())
5462 auto& pools = osdmap.get_pools();
5463 for (auto it = pools.begin(); it != pools.end(); ++it) {
5464 const pool_stat_t *pstat = mon->pgservice->get_pool_stat(it->first);
5467 const object_stat_sum_t& sum = pstat->stats.sum;
5468 const pg_pool_t &pool = it->second;
5469 const string& pool_name = osdmap.get_pool_name(it->first);
5472 (pool.quota_max_bytes > 0 && (uint64_t)sum.num_bytes >= pool.quota_max_bytes) ||
5473 (pool.quota_max_objects > 0 && (uint64_t)sum.num_objects >= pool.quota_max_objects);
5475 if (pool.has_flag(pg_pool_t::FLAG_FULL_NO_QUOTA)) {
5479 mon->clog->info() << "pool '" << pool_name
5480 << "' no longer out of quota; removing NO_QUOTA flag";
5481 // below we cancel FLAG_FULL too, we'll set it again in
5482 // OSDMonitor::encode_pending if it still fails the osd-full checking.
5483 clear_pool_flags(it->first,
5484 pg_pool_t::FLAG_FULL_NO_QUOTA | pg_pool_t::FLAG_FULL);
5490 if (pool.quota_max_bytes > 0 &&
5491 (uint64_t)sum.num_bytes >= pool.quota_max_bytes) {
5492 mon->clog->warn() << "pool '" << pool_name << "' is full"
5493 << " (reached quota's max_bytes: "
5494 << si_t(pool.quota_max_bytes) << ")";
5496 if (pool.quota_max_objects > 0 &&
5497 (uint64_t)sum.num_objects >= pool.quota_max_objects) {
5498 mon->clog->warn() << "pool '" << pool_name << "' is full"
5499 << " (reached quota's max_objects: "
5500 << pool.quota_max_objects << ")";
5502 // set both FLAG_FULL_NO_QUOTA and FLAG_FULL
5503 // note that below we try to cancel FLAG_BACKFILLFULL/NEARFULL too
5504 // since FLAG_FULL should always take precedence
5505 set_pool_flags(it->first,
5506 pg_pool_t::FLAG_FULL_NO_QUOTA | pg_pool_t::FLAG_FULL);
5507 clear_pool_flags(it->first,
5508 pg_pool_t::FLAG_NEARFULL |
5509 pg_pool_t::FLAG_BACKFILLFULL);
5516 int OSDMonitor::prepare_new_pool(MonOpRequestRef op)
5518 op->mark_osdmon_event(__func__);
5519 MPoolOp *m = static_cast<MPoolOp*>(op->get_req());
5520 dout(10) << "prepare_new_pool from " << m->get_connection() << dendl;
5521 MonSession *session = m->get_session();
5524 string erasure_code_profile;
5528 return prepare_new_pool(m->name, m->auid, m->crush_rule, rule_name,
5530 erasure_code_profile,
5531 pg_pool_t::TYPE_REPLICATED, 0, FAST_READ_OFF, &ss);
5533 return prepare_new_pool(m->name, session->auid, m->crush_rule, rule_name,
5535 erasure_code_profile,
5536 pg_pool_t::TYPE_REPLICATED, 0, FAST_READ_OFF, &ss);
5539 int OSDMonitor::crush_rename_bucket(const string& srcname,
5540 const string& dstname,
5545 // Avoid creating a pending crush if it does not already exists and
5546 // the rename would fail.
5548 if (!_have_pending_crush()) {
5549 ret = _get_stable_crush().can_rename_bucket(srcname,
5556 CrushWrapper newcrush;
5557 _get_pending_crush(newcrush);
5559 ret = newcrush.rename_bucket(srcname,
5565 pending_inc.crush.clear();
5566 newcrush.encode(pending_inc.crush, mon->get_quorum_con_features());
5567 *ss << "renamed bucket " << srcname << " into " << dstname;
5571 void OSDMonitor::check_legacy_ec_plugin(const string& plugin, const string& profile) const
5573 string replacement = "";
5575 if (plugin == "jerasure_generic" ||
5576 plugin == "jerasure_sse3" ||
5577 plugin == "jerasure_sse4" ||
5578 plugin == "jerasure_neon") {
5579 replacement = "jerasure";
5580 } else if (plugin == "shec_generic" ||
5581 plugin == "shec_sse3" ||
5582 plugin == "shec_sse4" ||
5583 plugin == "shec_neon") {
5584 replacement = "shec";
5587 if (replacement != "") {
5588 dout(0) << "WARNING: erasure coding profile " << profile << " uses plugin "
5589 << plugin << " that has been deprecated. Please use "
5590 << replacement << " instead." << dendl;
5594 int OSDMonitor::normalize_profile(const string& profilename,
5595 ErasureCodeProfile &profile,
5599 ErasureCodeInterfaceRef erasure_code;
5600 ErasureCodePluginRegistry &instance = ErasureCodePluginRegistry::instance();
5601 ErasureCodeProfile::const_iterator plugin = profile.find("plugin");
5602 check_legacy_ec_plugin(plugin->second, profilename);
5603 int err = instance.factory(plugin->second,
5604 g_conf->get_val<std::string>("erasure_code_dir"),
5605 profile, &erasure_code, ss);
5610 err = erasure_code->init(profile, ss);
5615 auto it = profile.find("stripe_unit");
5616 if (it != profile.end()) {
5618 uint32_t stripe_unit = strict_si_cast<uint32_t>(it->second.c_str(), &err_str);
5619 if (!err_str.empty()) {
5620 *ss << "could not parse stripe_unit '" << it->second
5621 << "': " << err_str << std::endl;
5624 uint32_t data_chunks = erasure_code->get_data_chunk_count();
5625 uint32_t chunk_size = erasure_code->get_chunk_size(stripe_unit * data_chunks);
5626 if (chunk_size != stripe_unit) {
5627 *ss << "stripe_unit " << stripe_unit << " does not match ec profile "
5628 << "alignment. Would be padded to " << chunk_size
5632 if ((stripe_unit % 4096) != 0 && !force) {
5633 *ss << "stripe_unit should be a multiple of 4096 bytes for best performance."
5634 << "use --force to override this check" << std::endl;
5641 int OSDMonitor::crush_rule_create_erasure(const string &name,
5642 const string &profile,
5646 int ruleid = osdmap.crush->get_rule_id(name);
5647 if (ruleid != -ENOENT) {
5648 *rule = osdmap.crush->get_rule_mask_ruleset(ruleid);
5652 CrushWrapper newcrush;
5653 _get_pending_crush(newcrush);
5655 ruleid = newcrush.get_rule_id(name);
5656 if (ruleid != -ENOENT) {
5657 *rule = newcrush.get_rule_mask_ruleset(ruleid);
5660 ErasureCodeInterfaceRef erasure_code;
5661 int err = get_erasure_code(profile, &erasure_code, ss);
5663 *ss << "failed to load plugin using profile " << profile << std::endl;
5667 err = erasure_code->create_rule(name, newcrush, ss);
5668 erasure_code.reset();
5672 pending_inc.crush.clear();
5673 newcrush.encode(pending_inc.crush, mon->get_quorum_con_features());
5678 int OSDMonitor::get_erasure_code(const string &erasure_code_profile,
5679 ErasureCodeInterfaceRef *erasure_code,
5682 if (pending_inc.has_erasure_code_profile(erasure_code_profile))
5684 ErasureCodeProfile profile =
5685 osdmap.get_erasure_code_profile(erasure_code_profile);
5686 ErasureCodeProfile::const_iterator plugin =
5687 profile.find("plugin");
5688 if (plugin == profile.end()) {
5689 *ss << "cannot determine the erasure code plugin"
5690 << " because there is no 'plugin' entry in the erasure_code_profile "
5691 << profile << std::endl;
5694 check_legacy_ec_plugin(plugin->second, erasure_code_profile);
5695 ErasureCodePluginRegistry &instance = ErasureCodePluginRegistry::instance();
5696 return instance.factory(plugin->second,
5697 g_conf->get_val<std::string>("erasure_code_dir"),
5698 profile, erasure_code, ss);
5701 int OSDMonitor::check_cluster_features(uint64_t features,
5704 stringstream unsupported_ss;
5705 int unsupported_count = 0;
5706 if ((mon->get_quorum_con_features() & features) != features) {
5707 unsupported_ss << "the monitor cluster";
5708 ++unsupported_count;
5711 set<int32_t> up_osds;
5712 osdmap.get_up_osds(up_osds);
5713 for (set<int32_t>::iterator it = up_osds.begin();
5714 it != up_osds.end(); ++it) {
5715 const osd_xinfo_t &xi = osdmap.get_xinfo(*it);
5716 if ((xi.features & features) != features) {
5717 if (unsupported_count > 0)
5718 unsupported_ss << ", ";
5719 unsupported_ss << "osd." << *it;
5720 unsupported_count ++;
5724 if (unsupported_count > 0) {
5725 ss << "features " << features << " unsupported by: "
5726 << unsupported_ss.str();
5730 // check pending osd state, too!
5731 for (map<int32_t,osd_xinfo_t>::const_iterator p =
5732 pending_inc.new_xinfo.begin();
5733 p != pending_inc.new_xinfo.end(); ++p) {
5734 const osd_xinfo_t &xi = p->second;
5735 if ((xi.features & features) != features) {
5736 dout(10) << __func__ << " pending osd." << p->first
5737 << " features are insufficient; retry" << dendl;
5745 bool OSDMonitor::validate_crush_against_features(const CrushWrapper *newcrush,
5748 OSDMap::Incremental new_pending = pending_inc;
5749 ::encode(*newcrush, new_pending.crush, mon->get_quorum_con_features());
5751 newmap.deepish_copy_from(osdmap);
5752 newmap.apply_incremental(new_pending);
5755 if (newmap.require_min_compat_client > 0) {
5756 auto mv = newmap.get_min_compat_client();
5757 if (mv > newmap.require_min_compat_client) {
5758 ss << "new crush map requires client version " << ceph_release_name(mv)
5759 << " but require_min_compat_client is "
5760 << ceph_release_name(newmap.require_min_compat_client);
5767 newmap.get_features(CEPH_ENTITY_TYPE_MON, NULL) |
5768 newmap.get_features(CEPH_ENTITY_TYPE_OSD, NULL);
5769 stringstream features_ss;
5770 int r = check_cluster_features(features, features_ss);
5772 ss << "Could not change CRUSH: " << features_ss.str();
5779 bool OSDMonitor::erasure_code_profile_in_use(
5780 const mempool::osdmap::map<int64_t, pg_pool_t> &pools,
5781 const string &profile,
5785 for (map<int64_t, pg_pool_t>::const_iterator p = pools.begin();
5788 if (p->second.erasure_code_profile == profile) {
5789 *ss << osdmap.pool_name[p->first] << " ";
5794 *ss << "pool(s) are using the erasure code profile '" << profile << "'";
5799 int OSDMonitor::parse_erasure_code_profile(const vector<string> &erasure_code_profile,
5800 map<string,string> *erasure_code_profile_map,
5803 int r = get_json_str_map(g_conf->osd_pool_default_erasure_code_profile,
5805 erasure_code_profile_map);
5808 assert((*erasure_code_profile_map).count("plugin"));
5809 string default_plugin = (*erasure_code_profile_map)["plugin"];
5810 map<string,string> user_map;
5811 for (vector<string>::const_iterator i = erasure_code_profile.begin();
5812 i != erasure_code_profile.end();
5814 size_t equal = i->find('=');
5815 if (equal == string::npos) {
5816 user_map[*i] = string();
5817 (*erasure_code_profile_map)[*i] = string();
5819 string key = i->substr(0, equal);
5821 const string value = i->substr(equal);
5822 if (osdmap.require_osd_release >= CEPH_RELEASE_LUMINOUS &&
5823 key.find("ruleset-") == 0) {
5824 if (g_conf->get_val<bool>("mon_fixup_legacy_erasure_code_profiles")) {
5825 mon->clog->warn() << "erasure code profile property '" << key
5826 << "' is no longer supported; try "
5827 << "'crush-" << key.substr(8) << "' instead";
5828 key = string("crush-") + key.substr(8);
5830 *ss << "property '" << key << "' is no longer supported; try "
5831 << "'crush-" << key.substr(8) << "' instead";
5835 user_map[key] = value;
5836 (*erasure_code_profile_map)[key] = value;
5840 if (user_map.count("plugin") && user_map["plugin"] != default_plugin)
5841 (*erasure_code_profile_map) = user_map;
5846 int OSDMonitor::prepare_pool_size(const unsigned pool_type,
5847 const string &erasure_code_profile,
5848 unsigned *size, unsigned *min_size,
5852 switch (pool_type) {
5853 case pg_pool_t::TYPE_REPLICATED:
5854 *size = g_conf->osd_pool_default_size;
5855 *min_size = g_conf->get_osd_pool_default_min_size();
5857 case pg_pool_t::TYPE_ERASURE:
5859 ErasureCodeInterfaceRef erasure_code;
5860 err = get_erasure_code(erasure_code_profile, &erasure_code, ss);
5862 *size = erasure_code->get_chunk_count();
5863 *min_size = MIN(erasure_code->get_data_chunk_count() + 1, *size);
5868 *ss << "prepare_pool_size: " << pool_type << " is not a known pool type";
5875 int OSDMonitor::prepare_pool_stripe_width(const unsigned pool_type,
5876 const string &erasure_code_profile,
5877 uint32_t *stripe_width,
5881 switch (pool_type) {
5882 case pg_pool_t::TYPE_REPLICATED:
5885 case pg_pool_t::TYPE_ERASURE:
5887 ErasureCodeProfile profile =
5888 osdmap.get_erasure_code_profile(erasure_code_profile);
5889 ErasureCodeInterfaceRef erasure_code;
5890 err = get_erasure_code(erasure_code_profile, &erasure_code, ss);
5893 uint32_t data_chunks = erasure_code->get_data_chunk_count();
5894 uint32_t stripe_unit = g_conf->osd_pool_erasure_code_stripe_unit;
5895 auto it = profile.find("stripe_unit");
5896 if (it != profile.end()) {
5898 stripe_unit = strict_si_cast<uint32_t>(it->second.c_str(), &err_str);
5899 assert(err_str.empty());
5901 *stripe_width = data_chunks *
5902 erasure_code->get_chunk_size(stripe_unit * data_chunks);
5906 *ss << "prepare_pool_stripe_width: "
5907 << pool_type << " is not a known pool type";
5914 int OSDMonitor::prepare_pool_crush_rule(const unsigned pool_type,
5915 const string &erasure_code_profile,
5916 const string &rule_name,
5921 if (*crush_rule < 0) {
5922 switch (pool_type) {
5923 case pg_pool_t::TYPE_REPLICATED:
5925 if (rule_name == "") {
5927 *crush_rule = osdmap.crush->get_osd_pool_default_crush_replicated_ruleset(g_ceph_context);
5928 if (*crush_rule < 0) {
5929 // Errors may happen e.g. if no valid rule is available
5930 *ss << "No suitable CRUSH rule exists, check "
5931 << "'osd pool default crush *' config options";
5935 return get_crush_rule(rule_name, crush_rule, ss);
5939 case pg_pool_t::TYPE_ERASURE:
5941 int err = crush_rule_create_erasure(rule_name,
5942 erasure_code_profile,
5946 dout(20) << "prepare_pool_crush_rule: rule "
5947 << rule_name << " try again" << dendl;
5950 // need to wait for the crush rule to be proposed before proceeding
5961 *ss << "prepare_pool_crush_rule: " << pool_type
5962 << " is not a known pool type";
5967 if (!osdmap.crush->ruleset_exists(*crush_rule)) {
5968 *ss << "CRUSH rule " << *crush_rule << " not found";
5976 int OSDMonitor::get_crush_rule(const string &rule_name,
5981 ret = osdmap.crush->get_rule_id(rule_name);
5982 if (ret != -ENOENT) {
5986 CrushWrapper newcrush;
5987 _get_pending_crush(newcrush);
5989 ret = newcrush.get_rule_id(rule_name);
5990 if (ret != -ENOENT) {
5991 // found it, wait for it to be proposed
5992 dout(20) << __func__ << ": rule " << rule_name
5993 << " try again" << dendl;
5996 // Cannot find it , return error
5997 *ss << "specified rule " << rule_name << " doesn't exist";
6004 int OSDMonitor::check_pg_num(int64_t pool, int pg_num, int size, ostream *ss)
6006 auto max_pgs_per_osd = g_conf->get_val<uint64_t>("mon_max_pg_per_osd");
6007 auto num_osds = std::max(osdmap.get_num_in_osds(), 3u); // assume min cluster size 3
6008 auto max_pgs = max_pgs_per_osd * num_osds;
6009 uint64_t projected = 0;
6011 projected += pg_num * size;
6013 for (const auto& i : osdmap.get_pools()) {
6014 if (i.first == pool) {
6015 projected += pg_num * size;
6017 projected += i.second.get_pg_num() * i.second.get_size();
6020 if (projected > max_pgs) {
6022 *ss << "pool id " << pool;
6024 *ss << " pg_num " << pg_num << " size " << size
6025 << " would mean " << projected
6026 << " total pgs, which exceeds max " << max_pgs
6027 << " (mon_max_pg_per_osd " << max_pgs_per_osd
6028 << " * num_in_osds " << num_osds << ")";
6035 * @param name The name of the new pool
6036 * @param auid The auid of the pool owner. Can be -1
6037 * @param crush_rule The crush rule to use. If <0, will use the system default
6038 * @param crush_rule_name The crush rule to use, if crush_rulset <0
6039 * @param pg_num The pg_num to use. If set to 0, will use the system default
6040 * @param pgp_num The pgp_num to use. If set to 0, will use the system default
6041 * @param erasure_code_profile The profile name in OSDMap to be used for erasure code
6042 * @param pool_type TYPE_ERASURE, or TYPE_REP
6043 * @param expected_num_objects expected number of objects on the pool
6044 * @param fast_read fast read type.
6045 * @param ss human readable error message, if any.
6047 * @return 0 on success, negative errno on failure.
6049 int OSDMonitor::prepare_new_pool(string& name, uint64_t auid,
6051 const string &crush_rule_name,
6052 unsigned pg_num, unsigned pgp_num,
6053 const string &erasure_code_profile,
6054 const unsigned pool_type,
6055 const uint64_t expected_num_objects,
6056 FastReadType fast_read,
6059 if (name.length() == 0)
6062 pg_num = g_conf->osd_pool_default_pg_num;
6064 pgp_num = g_conf->osd_pool_default_pgp_num;
6065 if (pg_num > (unsigned)g_conf->mon_max_pool_pg_num) {
6066 *ss << "'pg_num' must be greater than 0 and less than or equal to "
6067 << g_conf->mon_max_pool_pg_num
6068 << " (you may adjust 'mon max pool pg num' for higher values)";
6071 if (pgp_num > pg_num) {
6072 *ss << "'pgp_num' must be greater than 0 and lower or equal than 'pg_num'"
6073 << ", which in this case is " << pg_num;
6076 if (pool_type == pg_pool_t::TYPE_REPLICATED && fast_read == FAST_READ_ON) {
6077 *ss << "'fast_read' can only apply to erasure coding pool";
6081 r = prepare_pool_crush_rule(pool_type, erasure_code_profile,
6082 crush_rule_name, &crush_rule, ss);
6084 dout(10) << " prepare_pool_crush_rule returns " << r << dendl;
6087 if (g_conf->mon_osd_crush_smoke_test) {
6088 CrushWrapper newcrush;
6089 _get_pending_crush(newcrush);
6091 CrushTester tester(newcrush, err);
6092 tester.set_min_x(0);
6093 tester.set_max_x(50);
6094 tester.set_rule(crush_rule);
6095 auto start = ceph::coarse_mono_clock::now();
6096 r = tester.test_with_fork(g_conf->mon_lease);
6097 auto duration = ceph::coarse_mono_clock::now() - start;
6099 dout(10) << " tester.test_with_fork returns " << r
6100 << ": " << err.str() << dendl;
6101 *ss << "crush test failed with " << r << ": " << err.str();
6104 dout(10) << __func__ << " crush smoke test duration: "
6105 << duration << dendl;
6107 unsigned size, min_size;
6108 r = prepare_pool_size(pool_type, erasure_code_profile, &size, &min_size, ss);
6110 dout(10) << " prepare_pool_size returns " << r << dendl;
6113 r = check_pg_num(-1, pg_num, size, ss);
6115 dout(10) << " prepare_pool_size returns " << r << dendl;
6119 if (!osdmap.crush->check_crush_rule(crush_rule, pool_type, size, *ss)) {
6123 uint32_t stripe_width = 0;
6124 r = prepare_pool_stripe_width(pool_type, erasure_code_profile, &stripe_width, ss);
6126 dout(10) << " prepare_pool_stripe_width returns " << r << dendl;
6131 if (pool_type == pg_pool_t::TYPE_ERASURE) {
6132 switch (fast_read) {
6139 case FAST_READ_DEFAULT:
6140 fread = g_conf->mon_osd_pool_ec_fast_read;
6143 *ss << "invalid fast_read setting: " << fast_read;
6148 for (map<int64_t,string>::iterator p = pending_inc.new_pool_names.begin();
6149 p != pending_inc.new_pool_names.end();
6151 if (p->second == name)
6155 if (-1 == pending_inc.new_pool_max)
6156 pending_inc.new_pool_max = osdmap.pool_max;
6157 int64_t pool = ++pending_inc.new_pool_max;
6159 pg_pool_t *pi = pending_inc.get_new_pool(pool, &empty);
6160 pi->type = pool_type;
6161 pi->fast_read = fread;
6162 pi->flags = g_conf->osd_pool_default_flags;
6163 if (g_conf->osd_pool_default_flag_hashpspool)
6164 pi->set_flag(pg_pool_t::FLAG_HASHPSPOOL);
6165 if (g_conf->osd_pool_default_flag_nodelete)
6166 pi->set_flag(pg_pool_t::FLAG_NODELETE);
6167 if (g_conf->osd_pool_default_flag_nopgchange)
6168 pi->set_flag(pg_pool_t::FLAG_NOPGCHANGE);
6169 if (g_conf->osd_pool_default_flag_nosizechange)
6170 pi->set_flag(pg_pool_t::FLAG_NOSIZECHANGE);
6171 if (g_conf->osd_pool_use_gmt_hitset &&
6172 (osdmap.get_up_osd_features() & CEPH_FEATURE_OSD_HITSET_GMT))
6173 pi->use_gmt_hitset = true;
6175 pi->use_gmt_hitset = false;
6178 pi->min_size = min_size;
6179 pi->crush_rule = crush_rule;
6180 pi->expected_num_objects = expected_num_objects;
6181 pi->object_hash = CEPH_STR_HASH_RJENKINS;
6182 pi->set_pg_num(pg_num);
6183 pi->set_pgp_num(pgp_num);
6184 pi->last_change = pending_inc.epoch;
6186 pi->erasure_code_profile = erasure_code_profile;
6187 pi->stripe_width = stripe_width;
6188 pi->cache_target_dirty_ratio_micro =
6189 g_conf->osd_pool_default_cache_target_dirty_ratio * 1000000;
6190 pi->cache_target_dirty_high_ratio_micro =
6191 g_conf->osd_pool_default_cache_target_dirty_high_ratio * 1000000;
6192 pi->cache_target_full_ratio_micro =
6193 g_conf->osd_pool_default_cache_target_full_ratio * 1000000;
6194 pi->cache_min_flush_age = g_conf->osd_pool_default_cache_min_flush_age;
6195 pi->cache_min_evict_age = g_conf->osd_pool_default_cache_min_evict_age;
6196 pending_inc.new_pool_names[pool] = name;
6200 bool OSDMonitor::prepare_set_flag(MonOpRequestRef op, int flag)
6202 op->mark_osdmon_event(__func__);
6204 if (pending_inc.new_flags < 0)
6205 pending_inc.new_flags = osdmap.get_flags();
6206 pending_inc.new_flags |= flag;
6207 ss << OSDMap::get_flag_string(flag) << " is set";
6208 wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, ss.str(),
6209 get_last_committed() + 1));
6213 bool OSDMonitor::prepare_unset_flag(MonOpRequestRef op, int flag)
6215 op->mark_osdmon_event(__func__);
6217 if (pending_inc.new_flags < 0)
6218 pending_inc.new_flags = osdmap.get_flags();
6219 pending_inc.new_flags &= ~flag;
6220 ss << OSDMap::get_flag_string(flag) << " is unset";
6221 wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, ss.str(),
6222 get_last_committed() + 1));
6226 int OSDMonitor::prepare_command_pool_set(map<string,cmd_vartype> &cmdmap,
6230 cmd_getval(g_ceph_context, cmdmap, "pool", poolstr);
6231 int64_t pool = osdmap.lookup_pg_pool_name(poolstr.c_str());
6233 ss << "unrecognized pool '" << poolstr << "'";
6237 cmd_getval(g_ceph_context, cmdmap, "var", var);
6239 pg_pool_t p = *osdmap.get_pg_pool(pool);
6240 if (pending_inc.new_pools.count(pool))
6241 p = pending_inc.new_pools[pool];
6243 // accept val as a json string in the normal case (current
6244 // generation monitor). parse out int or float values from the
6245 // string as needed. however, if it is not a string, try to pull
6246 // out an int, in case an older monitor with an older json schema is
6247 // forwarding a request.
6249 string interr, floaterr;
6252 int64_t uf = 0; // micro-f
6253 if (!cmd_getval(g_ceph_context, cmdmap, "val", val)) {
6254 // wasn't a string; maybe an older mon forwarded json with an int?
6255 if (!cmd_getval(g_ceph_context, cmdmap, "val", n))
6256 return -EINVAL; // no value!
6258 // we got a string. see if it contains an int.
6259 n = strict_strtoll(val.c_str(), 10, &interr);
6261 f = strict_strtod(val.c_str(), &floaterr);
6262 uf = llrintl(f * (double)1000000.0);
6266 (var == "hit_set_type" || var == "hit_set_period" ||
6267 var == "hit_set_count" || var == "hit_set_fpp" ||
6268 var == "target_max_objects" || var == "target_max_bytes" ||
6269 var == "cache_target_full_ratio" || var == "cache_target_dirty_ratio" ||
6270 var == "cache_target_dirty_high_ratio" || var == "use_gmt_hitset" ||
6271 var == "cache_min_flush_age" || var == "cache_min_evict_age" ||
6272 var == "hit_set_grade_decay_rate" || var == "hit_set_search_last_n" ||
6273 var == "min_read_recency_for_promote" || var == "min_write_recency_for_promote")) {
6277 if (var == "size") {
6278 if (p.has_flag(pg_pool_t::FLAG_NOSIZECHANGE)) {
6279 ss << "pool size change is disabled; you must unset nosizechange flag for the pool first";
6282 if (p.type == pg_pool_t::TYPE_ERASURE) {
6283 ss << "can not change the size of an erasure-coded pool";
6286 if (interr.length()) {
6287 ss << "error parsing integer value '" << val << "': " << interr;
6290 if (n <= 0 || n > 10) {
6291 ss << "pool size must be between 1 and 10";
6294 int r = check_pg_num(pool, p.get_pg_num(), n, &ss);
6301 } else if (var == "min_size") {
6302 if (p.has_flag(pg_pool_t::FLAG_NOSIZECHANGE)) {
6303 ss << "pool min size change is disabled; you must unset nosizechange flag for the pool first";
6306 if (interr.length()) {
6307 ss << "error parsing integer value '" << val << "': " << interr;
6311 if (p.type != pg_pool_t::TYPE_ERASURE) {
6312 if (n < 1 || n > p.size) {
6313 ss << "pool min_size must be between 1 and " << (int)p.size;
6317 ErasureCodeInterfaceRef erasure_code;
6320 int err = get_erasure_code(p.erasure_code_profile, &erasure_code, &tmp);
6322 k = erasure_code->get_data_chunk_count();
6324 ss << __func__ << " get_erasure_code failed: " << tmp.rdbuf();
6328 if (n < k || n > p.size) {
6329 ss << "pool min_size must be between " << k << " and " << (int)p.size;
6334 } else if (var == "auid") {
6335 if (interr.length()) {
6336 ss << "error parsing integer value '" << val << "': " << interr;
6340 } else if (var == "crash_replay_interval") {
6341 if (interr.length()) {
6342 ss << "error parsing integer value '" << val << "': " << interr;
6345 p.crash_replay_interval = n;
6346 } else if (var == "pg_num") {
6347 if (p.has_flag(pg_pool_t::FLAG_NOPGCHANGE)) {
6348 ss << "pool pg_num change is disabled; you must unset nopgchange flag for the pool first";
6351 if (interr.length()) {
6352 ss << "error parsing integer value '" << val << "': " << interr;
6355 if (n <= (int)p.get_pg_num()) {
6356 ss << "specified pg_num " << n << " <= current " << p.get_pg_num();
6357 if (n < (int)p.get_pg_num())
6361 if (n > (unsigned)g_conf->mon_max_pool_pg_num) {
6362 ss << "'pg_num' must be greater than 0 and less than or equal to "
6363 << g_conf->mon_max_pool_pg_num
6364 << " (you may adjust 'mon max pool pg num' for higher values)";
6367 int r = check_pg_num(pool, n, p.get_size(), &ss);
6372 cmd_getval(g_ceph_context,cmdmap, "force", force);
6373 if (p.cache_mode != pg_pool_t::CACHEMODE_NONE &&
6374 force != "--yes-i-really-mean-it") {
6375 ss << "splits in cache pools must be followed by scrubs and leave sufficient free space to avoid overfilling. use --yes-i-really-mean-it to force.";
6378 int expected_osds = MIN(p.get_pg_num(), osdmap.get_num_osds());
6379 int64_t new_pgs = n - p.get_pg_num();
6380 if (new_pgs > g_conf->mon_osd_max_split_count * expected_osds) {
6381 ss << "specified pg_num " << n << " is too large (creating "
6382 << new_pgs << " new PGs on ~" << expected_osds
6383 << " OSDs exceeds per-OSD max of " << g_conf->mon_osd_max_split_count
6388 // force pre-luminous clients to resend their ops, since they
6389 // don't understand that split PGs now form a new interval.
6390 p.last_force_op_resend_preluminous = pending_inc.epoch;
6391 } else if (var == "pgp_num") {
6392 if (p.has_flag(pg_pool_t::FLAG_NOPGCHANGE)) {
6393 ss << "pool pgp_num change is disabled; you must unset nopgchange flag for the pool first";
6396 if (interr.length()) {
6397 ss << "error parsing integer value '" << val << "': " << interr;
6401 ss << "specified pgp_num must > 0, but you set to " << n;
6404 if (n > (int)p.get_pg_num()) {
6405 ss << "specified pgp_num " << n << " > pg_num " << p.get_pg_num();
6409 } else if (var == "crush_rule") {
6410 int id = osdmap.crush->get_rule_id(val);
6411 if (id == -ENOENT) {
6412 ss << "crush rule " << val << " does not exist";
6416 ss << cpp_strerror(id);
6419 if (!osdmap.crush->check_crush_rule(id, p.get_type(), p.get_size(), ss)) {
6423 } else if (var == "nodelete" || var == "nopgchange" ||
6424 var == "nosizechange" || var == "write_fadvise_dontneed" ||
6425 var == "noscrub" || var == "nodeep-scrub") {
6426 uint64_t flag = pg_pool_t::get_flag_by_name(var);
6427 // make sure we only compare against 'n' if we didn't receive a string
6428 if (val == "true" || (interr.empty() && n == 1)) {
6430 } else if (val == "false" || (interr.empty() && n == 0)) {
6433 ss << "expecting value 'true', 'false', '0', or '1'";
6436 } else if (var == "hashpspool") {
6437 uint64_t flag = pg_pool_t::get_flag_by_name(var);
6439 cmd_getval(g_ceph_context, cmdmap, "force", force);
6440 if (force != "--yes-i-really-mean-it") {
6441 ss << "are you SURE? this will remap all placement groups in this pool,"
6442 " this triggers large data movement,"
6443 " pass --yes-i-really-mean-it if you really do.";
6446 // make sure we only compare against 'n' if we didn't receive a string
6447 if (val == "true" || (interr.empty() && n == 1)) {
6449 } else if (val == "false" || (interr.empty() && n == 0)) {
6452 ss << "expecting value 'true', 'false', '0', or '1'";
6455 } else if (var == "hit_set_type") {
6457 p.hit_set_params = HitSet::Params();
6459 int err = check_cluster_features(CEPH_FEATURE_OSD_CACHEPOOL, ss);
6462 if (val == "bloom") {
6463 BloomHitSet::Params *bsp = new BloomHitSet::Params;
6464 bsp->set_fpp(g_conf->osd_pool_default_hit_set_bloom_fpp);
6465 p.hit_set_params = HitSet::Params(bsp);
6466 } else if (val == "explicit_hash")
6467 p.hit_set_params = HitSet::Params(new ExplicitHashHitSet::Params);
6468 else if (val == "explicit_object")
6469 p.hit_set_params = HitSet::Params(new ExplicitObjectHitSet::Params);
6471 ss << "unrecognized hit_set type '" << val << "'";
6475 } else if (var == "hit_set_period") {
6476 if (interr.length()) {
6477 ss << "error parsing integer value '" << val << "': " << interr;
6480 p.hit_set_period = n;
6481 } else if (var == "hit_set_count") {
6482 if (interr.length()) {
6483 ss << "error parsing integer value '" << val << "': " << interr;
6486 p.hit_set_count = n;
6487 } else if (var == "hit_set_fpp") {
6488 if (floaterr.length()) {
6489 ss << "error parsing floating point value '" << val << "': " << floaterr;
6492 if (p.hit_set_params.get_type() != HitSet::TYPE_BLOOM) {
6493 ss << "hit set is not of type Bloom; invalid to set a false positive rate!";
6496 BloomHitSet::Params *bloomp = static_cast<BloomHitSet::Params*>(p.hit_set_params.impl.get());
6498 } else if (var == "use_gmt_hitset") {
6499 if (val == "true" || (interr.empty() && n == 1)) {
6501 cmd_getval(g_ceph_context, cmdmap, "force", force);
6502 if (!osdmap.get_num_up_osds() && force != "--yes-i-really-mean-it") {
6503 ss << "Not advisable to continue since no OSDs are up. Pass "
6504 << "--yes-i-really-mean-it if you really wish to continue.";
6507 if (!(osdmap.get_up_osd_features() & CEPH_FEATURE_OSD_HITSET_GMT)
6508 && force != "--yes-i-really-mean-it") {
6509 ss << "not all OSDs support GMT hit set.";
6512 p.use_gmt_hitset = true;
6514 ss << "expecting value 'true' or '1'";
6517 } else if (var == "allow_ec_overwrites") {
6518 if (!p.is_erasure()) {
6519 ss << "ec overwrites can only be enabled for an erasure coded pool";
6523 if (!g_conf->mon_debug_no_require_bluestore_for_ec_overwrites &&
6524 !is_pool_currently_all_bluestore(pool, p, &err)) {
6525 ss << "pool must only be stored on bluestore for scrubbing to work: " << err.str();
6528 if (val == "true" || (interr.empty() && n == 1)) {
6529 p.flags |= pg_pool_t::FLAG_EC_OVERWRITES;
6530 } else if (val == "false" || (interr.empty() && n == 0)) {
6531 ss << "ec overwrites cannot be disabled once enabled";
6534 ss << "expecting value 'true', 'false', '0', or '1'";
6537 } else if (var == "target_max_objects") {
6538 if (interr.length()) {
6539 ss << "error parsing int '" << val << "': " << interr;
6542 p.target_max_objects = n;
6543 } else if (var == "target_max_bytes") {
6544 if (interr.length()) {
6545 ss << "error parsing int '" << val << "': " << interr;
6548 p.target_max_bytes = n;
6549 } else if (var == "cache_target_dirty_ratio") {
6550 if (floaterr.length()) {
6551 ss << "error parsing float '" << val << "': " << floaterr;
6554 if (f < 0 || f > 1.0) {
6555 ss << "value must be in the range 0..1";
6558 p.cache_target_dirty_ratio_micro = uf;
6559 } else if (var == "cache_target_dirty_high_ratio") {
6560 if (floaterr.length()) {
6561 ss << "error parsing float '" << val << "': " << floaterr;
6564 if (f < 0 || f > 1.0) {
6565 ss << "value must be in the range 0..1";
6568 p.cache_target_dirty_high_ratio_micro = uf;
6569 } else if (var == "cache_target_full_ratio") {
6570 if (floaterr.length()) {
6571 ss << "error parsing float '" << val << "': " << floaterr;
6574 if (f < 0 || f > 1.0) {
6575 ss << "value must be in the range 0..1";
6578 p.cache_target_full_ratio_micro = uf;
6579 } else if (var == "cache_min_flush_age") {
6580 if (interr.length()) {
6581 ss << "error parsing int '" << val << "': " << interr;
6584 p.cache_min_flush_age = n;
6585 } else if (var == "cache_min_evict_age") {
6586 if (interr.length()) {
6587 ss << "error parsing int '" << val << "': " << interr;
6590 p.cache_min_evict_age = n;
6591 } else if (var == "min_read_recency_for_promote") {
6592 if (interr.length()) {
6593 ss << "error parsing integer value '" << val << "': " << interr;
6596 p.min_read_recency_for_promote = n;
6597 } else if (var == "hit_set_grade_decay_rate") {
6598 if (interr.length()) {
6599 ss << "error parsing integer value '" << val << "': " << interr;
6602 if (n > 100 || n < 0) {
6603 ss << "value out of range,valid range is 0 - 100";
6606 p.hit_set_grade_decay_rate = n;
6607 } else if (var == "hit_set_search_last_n") {
6608 if (interr.length()) {
6609 ss << "error parsing integer value '" << val << "': " << interr;
6612 if (n > p.hit_set_count || n < 0) {
6613 ss << "value out of range,valid range is 0 - hit_set_count";
6616 p.hit_set_search_last_n = n;
6617 } else if (var == "min_write_recency_for_promote") {
6618 if (interr.length()) {
6619 ss << "error parsing integer value '" << val << "': " << interr;
6622 p.min_write_recency_for_promote = n;
6623 } else if (var == "fast_read") {
6624 if (p.is_replicated()) {
6625 ss << "fast read is not supported in replication pool";
6628 if (val == "true" || (interr.empty() && n == 1)) {
6630 } else if (val == "false" || (interr.empty() && n == 0)) {
6631 p.fast_read = false;
6633 ss << "expecting value 'true', 'false', '0', or '1'";
6636 } else if (pool_opts_t::is_opt_name(var)) {
6637 bool unset = val == "unset";
6638 if (var == "compression_mode") {
6640 auto cmode = Compressor::get_comp_mode_type(val);
6642 ss << "unrecognized compression mode '" << val << "'";
6646 } else if (var == "compression_algorithm") {
6648 auto alg = Compressor::get_comp_alg_type(val);
6650 ss << "unrecognized compression_algorithm '" << val << "'";
6654 } else if (var == "compression_required_ratio") {
6655 if (floaterr.length()) {
6656 ss << "error parsing float value '" << val << "': " << floaterr;
6659 if (f < 0 || f > 1) {
6660 ss << "compression_required_ratio is out of range (0-1): '" << val << "'";
6663 } else if (var == "csum_type") {
6664 auto t = unset ? 0 : Checksummer::get_csum_string_type(val);
6666 ss << "unrecognized csum_type '" << val << "'";
6669 //preserve csum_type numeric value
6672 } else if (var == "compression_max_blob_size" ||
6673 var == "compression_min_blob_size" ||
6674 var == "csum_max_block" ||
6675 var == "csum_min_block") {
6676 if (interr.length()) {
6677 ss << "error parsing int value '" << val << "': " << interr;
6682 pool_opts_t::opt_desc_t desc = pool_opts_t::get_opt_desc(var);
6683 switch (desc.type) {
6684 case pool_opts_t::STR:
6686 p.opts.unset(desc.key);
6688 p.opts.set(desc.key, static_cast<std::string>(val));
6691 case pool_opts_t::INT:
6692 if (interr.length()) {
6693 ss << "error parsing integer value '" << val << "': " << interr;
6697 p.opts.unset(desc.key);
6699 p.opts.set(desc.key, static_cast<int>(n));
6702 case pool_opts_t::DOUBLE:
6703 if (floaterr.length()) {
6704 ss << "error parsing floating point value '" << val << "': " << floaterr;
6708 p.opts.unset(desc.key);
6710 p.opts.set(desc.key, static_cast<double>(f));
6714 assert(!"unknown type");
6717 ss << "unrecognized variable '" << var << "'";
6720 if (val != "unset") {
6721 ss << "set pool " << pool << " " << var << " to " << val;
6723 ss << "unset pool " << pool << " " << var;
6725 p.last_change = pending_inc.epoch;
6726 pending_inc.new_pools[pool] = p;
6730 int OSDMonitor::prepare_command_pool_application(const string &prefix,
6731 map<string,cmd_vartype> &cmdmap,
6735 cmd_getval(g_ceph_context, cmdmap, "pool", pool_name);
6736 int64_t pool = osdmap.lookup_pg_pool_name(pool_name.c_str());
6738 ss << "unrecognized pool '" << pool_name << "'";
6742 pg_pool_t p = *osdmap.get_pg_pool(pool);
6743 if (pending_inc.new_pools.count(pool)) {
6744 p = pending_inc.new_pools[pool];
6748 cmd_getval(g_ceph_context, cmdmap, "app", app);
6749 bool app_exists = (p.application_metadata.count(app) > 0);
6751 if (boost::algorithm::ends_with(prefix, "enable")) {
6753 ss << "application name must be provided";
6758 ss << "application must be enabled on base tier";
6763 cmd_getval(g_ceph_context, cmdmap, "force", force);
6765 if (!app_exists && !p.application_metadata.empty() &&
6766 force != "--yes-i-really-mean-it") {
6767 ss << "Are you SURE? Pool '" << pool_name << "' already has an enabled "
6768 << "application; pass --yes-i-really-mean-it to proceed anyway";
6772 if (!app_exists && p.application_metadata.size() >= MAX_POOL_APPLICATIONS) {
6773 ss << "too many enabled applications on pool '" << pool_name << "'; "
6774 << "max " << MAX_POOL_APPLICATIONS;
6778 if (app.length() > MAX_POOL_APPLICATION_LENGTH) {
6779 ss << "application name '" << app << "' too long; max length "
6780 << MAX_POOL_APPLICATION_LENGTH;
6785 p.application_metadata[app] = {};
6787 ss << "enabled application '" << app << "' on pool '" << pool_name << "'";
6789 } else if (boost::algorithm::ends_with(prefix, "disable")) {
6791 cmd_getval(g_ceph_context, cmdmap, "force", force);
6793 if (force != "--yes-i-really-mean-it") {
6794 ss << "Are you SURE? Disabling an application within a pool might result "
6795 << "in loss of application functionality; pass "
6796 << "--yes-i-really-mean-it to proceed anyway";
6801 ss << "application '" << app << "' is not enabled on pool '" << pool_name
6803 return 0; // idempotent
6806 p.application_metadata.erase(app);
6807 ss << "disable application '" << app << "' on pool '" << pool_name << "'";
6809 } else if (boost::algorithm::ends_with(prefix, "set")) {
6811 ss << "application metadata must be set on base tier";
6816 ss << "application '" << app << "' is not enabled on pool '" << pool_name
6822 cmd_getval(g_ceph_context, cmdmap, "key", key);
6825 ss << "key must be provided";
6829 auto &app_keys = p.application_metadata[app];
6830 if (app_keys.count(key) == 0 &&
6831 app_keys.size() >= MAX_POOL_APPLICATION_KEYS) {
6832 ss << "too many keys set for application '" << app << "' on pool '"
6833 << pool_name << "'; max " << MAX_POOL_APPLICATION_KEYS;
6837 if (key.length() > MAX_POOL_APPLICATION_LENGTH) {
6838 ss << "key '" << app << "' too long; max length "
6839 << MAX_POOL_APPLICATION_LENGTH;
6844 cmd_getval(g_ceph_context, cmdmap, "value", value);
6845 if (value.length() > MAX_POOL_APPLICATION_LENGTH) {
6846 ss << "value '" << value << "' too long; max length "
6847 << MAX_POOL_APPLICATION_LENGTH;
6851 p.application_metadata[app][key] = value;
6852 ss << "set application '" << app << "' key '" << key << "' to '"
6853 << value << "' on pool '" << pool_name << "'";
6854 } else if (boost::algorithm::ends_with(prefix, "rm")) {
6856 ss << "application '" << app << "' is not enabled on pool '" << pool_name
6862 cmd_getval(g_ceph_context, cmdmap, "key", key);
6863 auto it = p.application_metadata[app].find(key);
6864 if (it == p.application_metadata[app].end()) {
6865 ss << "application '" << app << "' on pool '" << pool_name
6866 << "' does not have key '" << key << "'";
6867 return 0; // idempotent
6870 p.application_metadata[app].erase(it);
6871 ss << "removed application '" << app << "' key '" << key << "' on pool '"
6872 << pool_name << "'";
6877 p.last_change = pending_inc.epoch;
6878 pending_inc.new_pools[pool] = p;
6882 int OSDMonitor::_prepare_command_osd_crush_remove(
6883 CrushWrapper &newcrush,
6892 err = newcrush.remove_item_under(g_ceph_context, id, ancestor,
6895 err = newcrush.remove_item(g_ceph_context, id, unlink_only);
6900 void OSDMonitor::do_osd_crush_remove(CrushWrapper& newcrush)
6902 pending_inc.crush.clear();
6903 newcrush.encode(pending_inc.crush, mon->get_quorum_con_features());
6906 int OSDMonitor::prepare_command_osd_crush_remove(
6907 CrushWrapper &newcrush,
6913 int err = _prepare_command_osd_crush_remove(
6914 newcrush, id, ancestor,
6915 has_ancestor, unlink_only);
6921 do_osd_crush_remove(newcrush);
6926 int OSDMonitor::prepare_command_osd_remove(int32_t id)
6928 if (osdmap.is_up(id)) {
6932 pending_inc.new_state[id] = osdmap.get_state(id);
6933 pending_inc.new_uuid[id] = uuid_d();
6934 pending_metadata_rm.insert(id);
6935 pending_metadata.erase(id);
6940 int32_t OSDMonitor::_allocate_osd_id(int32_t* existing_id)
6942 assert(existing_id);
6945 for (int32_t i = 0; i < osdmap.get_max_osd(); ++i) {
6946 if (!osdmap.exists(i) &&
6947 pending_inc.new_up_client.count(i) == 0 &&
6948 (pending_inc.new_state.count(i) == 0 ||
6949 (pending_inc.new_state[i] & CEPH_OSD_EXISTS) == 0)) {
6955 if (pending_inc.new_max_osd < 0) {
6956 return osdmap.get_max_osd();
6958 return pending_inc.new_max_osd;
6961 void OSDMonitor::do_osd_create(
6966 dout(10) << __func__ << " uuid " << uuid << dendl;
6969 // We presume validation has been performed prior to calling this
6970 // function. We assert with prejudice.
6972 int32_t allocated_id = -1; // declare here so we can jump
6973 int32_t existing_id = -1;
6974 if (!uuid.is_zero()) {
6975 existing_id = osdmap.identify_osd(uuid);
6976 if (existing_id >= 0) {
6977 assert(id < 0 || id == existing_id);
6978 *new_id = existing_id;
6980 } else if (id >= 0) {
6981 // uuid does not exist, and id has been provided, so just create
6988 // allocate a new id
6989 allocated_id = _allocate_osd_id(&existing_id);
6990 dout(10) << __func__ << " allocated id " << allocated_id
6991 << " existing id " << existing_id << dendl;
6992 if (existing_id >= 0) {
6993 assert(existing_id < osdmap.get_max_osd());
6994 assert(allocated_id < 0);
6995 pending_inc.new_weight[existing_id] = CEPH_OSD_OUT;
6996 *new_id = existing_id;
6998 } else if (allocated_id >= 0) {
6999 assert(existing_id < 0);
7001 if (pending_inc.new_max_osd < 0) {
7002 pending_inc.new_max_osd = osdmap.get_max_osd() + 1;
7004 ++pending_inc.new_max_osd;
7006 *new_id = pending_inc.new_max_osd - 1;
7007 assert(*new_id == allocated_id);
7009 assert(0 == "unexpected condition");
7013 dout(10) << __func__ << " using id " << *new_id << dendl;
7014 if (osdmap.get_max_osd() <= *new_id && pending_inc.new_max_osd <= *new_id) {
7015 pending_inc.new_max_osd = *new_id + 1;
7018 pending_inc.new_state[*new_id] |= CEPH_OSD_EXISTS | CEPH_OSD_NEW;
7019 if (!uuid.is_zero())
7020 pending_inc.new_uuid[*new_id] = uuid;
7023 int OSDMonitor::validate_osd_create(
7026 const bool check_osd_exists,
7027 int32_t* existing_id,
7031 dout(10) << __func__ << " id " << id << " uuid " << uuid
7032 << " check_osd_exists " << check_osd_exists << dendl;
7034 assert(existing_id);
7036 if (id < 0 && uuid.is_zero()) {
7037 // we have nothing to validate
7040 } else if (uuid.is_zero()) {
7041 // we have an id but we will ignore it - because that's what
7042 // `osd create` does.
7047 * This function will be used to validate whether we are able to
7048 * create a new osd when the `uuid` is specified.
7050 * It will be used by both `osd create` and `osd new`, as the checks
7051 * are basically the same when it pertains to osd id and uuid validation.
7052 * However, `osd create` presumes an `uuid` is optional, for legacy
7053 * reasons, while `osd new` requires the `uuid` to be provided. This
7054 * means that `osd create` will not be idempotent if an `uuid` is not
7055 * provided, but we will always guarantee the idempotency of `osd new`.
7058 assert(!uuid.is_zero());
7059 if (pending_inc.identify_osd(uuid) >= 0) {
7060 // osd is about to exist
7064 int32_t i = osdmap.identify_osd(uuid);
7066 // osd already exists
7067 if (id >= 0 && i != id) {
7068 ss << "uuid " << uuid << " already in use for different id " << i;
7071 // return a positive errno to distinguish between a blocking error
7072 // and an error we consider to not be a problem (i.e., this would be
7073 // an idempotent operation).
7079 if (pending_inc.new_state.count(id)) {
7080 // osd is about to exist
7083 // we may not care if an osd exists if we are recreating a previously
7085 if (check_osd_exists && osdmap.exists(id)) {
7086 ss << "id " << id << " already in use and does not match uuid "
7094 int OSDMonitor::prepare_command_osd_create(
7097 int32_t* existing_id,
7100 dout(10) << __func__ << " id " << id << " uuid " << uuid << dendl;
7101 assert(existing_id);
7102 if (osdmap.is_destroyed(id)) {
7103 ss << "ceph osd create has been deprecated. Please use ceph osd new "
7108 if (uuid.is_zero()) {
7109 dout(10) << __func__ << " no uuid; assuming legacy `osd create`" << dendl;
7112 return validate_osd_create(id, uuid, true, existing_id, ss);
7115 int OSDMonitor::prepare_command_osd_new(
7117 const map<string,cmd_vartype>& cmdmap,
7118 const map<string,string>& secrets,
7126 assert(paxos->is_plugged());
7128 dout(10) << __func__ << " " << op << dendl;
7130 /* validate command. abort now if something's wrong. */
7132 /* `osd new` will expect a `uuid` to be supplied; `id` is optional.
7134 * If `id` is not specified, we will identify any existing osd based
7135 * on `uuid`. Operation will be idempotent iff secrets match.
7137 * If `id` is specified, we will identify any existing osd based on
7138 * `uuid` and match against `id`. If they match, operation will be
7139 * idempotent iff secrets match.
7141 * `-i secrets.json` will be optional. If supplied, will be used
7142 * to check for idempotency when `id` and `uuid` match.
7144 * If `id` is not specified, and `uuid` does not exist, an id will
7145 * be found or allocated for the osd.
7147 * If `id` is specified, and the osd has been previously marked
7148 * as destroyed, then the `id` will be reused.
7150 if (!cmd_getval(g_ceph_context, cmdmap, "uuid", uuidstr)) {
7151 ss << "requires the OSD's UUID to be specified.";
7153 } else if (!uuid.parse(uuidstr.c_str())) {
7154 ss << "invalid UUID value '" << uuidstr << "'.";
7158 if (cmd_getval(g_ceph_context, cmdmap, "id", id) &&
7160 ss << "invalid OSD id; must be greater or equal than zero.";
7164 // are we running an `osd create`-like command, or recreating
7165 // a previously destroyed osd?
7167 bool is_recreate_destroyed = (id >= 0 && osdmap.is_destroyed(id));
7169 // we will care about `id` to assess whether osd is `destroyed`, or
7170 // to create a new osd.
7171 // we will need an `id` by the time we reach auth.
7173 int32_t existing_id = -1;
7174 int err = validate_osd_create(id, uuid, !is_recreate_destroyed,
7177 bool may_be_idempotent = false;
7178 if (err == EEXIST) {
7179 // this is idempotent from the osdmon's point-of-view
7180 may_be_idempotent = true;
7181 assert(existing_id >= 0);
7183 } else if (err < 0) {
7187 if (!may_be_idempotent) {
7188 // idempotency is out of the window. We are either creating a new
7189 // osd or recreating a destroyed osd.
7191 // We now need to figure out if we have an `id` (and if it's valid),
7192 // of find an `id` if we don't have one.
7194 // NOTE: we need to consider the case where the `id` is specified for
7195 // `osd create`, and we must honor it. So this means checking if
7196 // the `id` is destroyed, and if so assume the destroy; otherwise,
7197 // check if it `exists` - in which case we complain about not being
7198 // `destroyed`. In the end, if nothing fails, we must allow the
7199 // creation, so that we are compatible with `create`.
7200 if (id >= 0 && osdmap.exists(id) && !osdmap.is_destroyed(id)) {
7201 dout(10) << __func__ << " osd." << id << " isn't destroyed" << dendl;
7202 ss << "OSD " << id << " has not yet been destroyed";
7204 } else if (id < 0) {
7206 id = _allocate_osd_id(&existing_id);
7208 assert(existing_id >= 0);
7211 dout(10) << __func__ << " found id " << id << " to use" << dendl;
7212 } else if (id >= 0 && osdmap.is_destroyed(id)) {
7213 dout(10) << __func__ << " recreating osd." << id << dendl;
7215 dout(10) << __func__ << " creating new osd." << id << dendl;
7219 assert(osdmap.exists(id));
7222 // we are now able to either create a brand new osd or reuse an existing
7223 // osd that has been previously destroyed.
7225 dout(10) << __func__ << " id " << id << " uuid " << uuid << dendl;
7227 if (may_be_idempotent && secrets.empty()) {
7228 // nothing to do, really.
7229 dout(10) << __func__ << " idempotent and no secrets -- no op." << dendl;
7232 f->open_object_section("created_osd");
7233 f->dump_int("osdid", id);
7241 string cephx_secret, lockbox_secret, dmcrypt_key;
7242 bool has_lockbox = false;
7243 bool has_secrets = (!secrets.empty());
7245 ConfigKeyService *svc = nullptr;
7246 AuthMonitor::auth_entity_t cephx_entity, lockbox_entity;
7249 if (secrets.count("cephx_secret") == 0) {
7250 ss << "requires a cephx secret.";
7253 cephx_secret = secrets.at("cephx_secret");
7255 bool has_lockbox_secret = (secrets.count("cephx_lockbox_secret") > 0);
7256 bool has_dmcrypt_key = (secrets.count("dmcrypt_key") > 0);
7258 dout(10) << __func__ << " has lockbox " << has_lockbox_secret
7259 << " dmcrypt " << has_dmcrypt_key << dendl;
7261 if (has_lockbox_secret && has_dmcrypt_key) {
7263 lockbox_secret = secrets.at("cephx_lockbox_secret");
7264 dmcrypt_key = secrets.at("dmcrypt_key");
7265 } else if (!has_lockbox_secret != !has_dmcrypt_key) {
7266 ss << "requires both a cephx lockbox secret and a dm-crypt key.";
7270 dout(10) << __func__ << " validate secrets using osd id " << id << dendl;
7272 err = mon->authmon()->validate_osd_new(id, uuid,
7280 } else if (may_be_idempotent && err != EEXIST) {
7281 // for this to be idempotent, `id` should already be >= 0; no need
7282 // to use validate_id.
7284 ss << "osd." << id << " exists but secrets do not match";
7289 svc = (ConfigKeyService*)mon->config_key_service;
7290 err = svc->validate_osd_new(uuid, dmcrypt_key, ss);
7293 } else if (may_be_idempotent && err != EEXIST) {
7295 ss << "osd." << id << " exists but dm-crypt key does not match.";
7300 assert(!has_secrets || !cephx_secret.empty());
7301 assert(!has_lockbox || !lockbox_secret.empty());
7303 if (may_be_idempotent) {
7304 // we have nothing to do for either the osdmon or the authmon,
7305 // and we have no lockbox - so the config key service will not be
7306 // touched. This is therefore an idempotent operation, and we can
7307 // just return right away.
7308 dout(10) << __func__ << " idempotent -- no op." << dendl;
7311 f->open_object_section("created_osd");
7312 f->dump_int("osdid", id);
7319 assert(!may_be_idempotent);
7323 assert(!cephx_secret.empty());
7324 assert((lockbox_secret.empty() && dmcrypt_key.empty()) ||
7325 (!lockbox_secret.empty() && !dmcrypt_key.empty()));
7327 err = mon->authmon()->do_osd_new(cephx_entity,
7333 assert(nullptr != svc);
7334 svc->do_osd_new(uuid, dmcrypt_key);
7338 if (is_recreate_destroyed) {
7340 assert(osdmap.is_destroyed(id));
7341 pending_inc.new_weight[id] = CEPH_OSD_OUT;
7342 pending_inc.new_state[id] |= CEPH_OSD_DESTROYED | CEPH_OSD_NEW;
7343 if (osdmap.get_state(id) & CEPH_OSD_UP) {
7344 // due to http://tracker.ceph.com/issues/20751 some clusters may
7345 // have UP set for non-existent OSDs; make sure it is cleared
7346 // for a newly created osd.
7347 pending_inc.new_state[id] |= CEPH_OSD_UP;
7349 pending_inc.new_uuid[id] = uuid;
7352 int32_t new_id = -1;
7353 do_osd_create(id, uuid, &new_id);
7354 assert(new_id >= 0);
7355 assert(id == new_id);
7359 f->open_object_section("created_osd");
7360 f->dump_int("osdid", id);
7369 bool OSDMonitor::prepare_command(MonOpRequestRef op)
7371 op->mark_osdmon_event(__func__);
7372 MMonCommand *m = static_cast<MMonCommand*>(op->get_req());
7374 map<string, cmd_vartype> cmdmap;
7375 if (!cmdmap_from_json(m->cmd, &cmdmap, ss)) {
7376 string rs = ss.str();
7377 mon->reply_command(op, -EINVAL, rs, get_last_committed());
7381 MonSession *session = m->get_session();
7383 mon->reply_command(op, -EACCES, "access denied", get_last_committed());
7387 return prepare_command_impl(op, cmdmap);
7390 static int parse_reweights(CephContext *cct,
7391 const map<string,cmd_vartype> &cmdmap,
7392 const OSDMap& osdmap,
7393 map<int32_t, uint32_t>* weights)
7396 if (!cmd_getval(g_ceph_context, cmdmap, "weights", weights_str)) {
7399 std::replace(begin(weights_str), end(weights_str), '\'', '"');
7400 json_spirit::mValue json_value;
7401 if (!json_spirit::read(weights_str, json_value)) {
7404 if (json_value.type() != json_spirit::obj_type) {
7407 const auto obj = json_value.get_obj();
7409 for (auto& osd_weight : obj) {
7410 auto osd_id = std::stoi(osd_weight.first);
7411 if (!osdmap.exists(osd_id)) {
7414 if (osd_weight.second.type() != json_spirit::str_type) {
7417 auto weight = std::stoul(osd_weight.second.get_str());
7418 weights->insert({osd_id, weight});
7420 } catch (const std::logic_error& e) {
7426 int OSDMonitor::prepare_command_osd_destroy(
7430 assert(paxos->is_plugged());
7432 // we check if the osd exists for the benefit of `osd purge`, which may
7433 // have previously removed the osd. If the osd does not exist, return
7434 // -ENOENT to convey this, and let the caller deal with it.
7436 // we presume that all auth secrets and config keys were removed prior
7437 // to this command being called. if they exist by now, we also assume
7438 // they must have been created by some other command and do not pertain
7439 // to this non-existent osd.
7440 if (!osdmap.exists(id)) {
7441 dout(10) << __func__ << " osd." << id << " does not exist." << dendl;
7445 uuid_d uuid = osdmap.get_uuid(id);
7446 dout(10) << __func__ << " destroying osd." << id
7447 << " uuid " << uuid << dendl;
7449 // if it has been destroyed, we assume our work here is done.
7450 if (osdmap.is_destroyed(id)) {
7451 ss << "destroyed osd." << id;
7455 EntityName cephx_entity, lockbox_entity;
7456 bool idempotent_auth = false, idempotent_cks = false;
7458 int err = mon->authmon()->validate_osd_destroy(id, uuid,
7463 if (err == -ENOENT) {
7464 idempotent_auth = true;
7470 ConfigKeyService *svc = (ConfigKeyService*)mon->config_key_service;
7471 err = svc->validate_osd_destroy(id, uuid);
7473 assert(err == -ENOENT);
7475 idempotent_cks = true;
7478 if (!idempotent_auth) {
7479 err = mon->authmon()->do_osd_destroy(cephx_entity, lockbox_entity);
7483 if (!idempotent_cks) {
7484 svc->do_osd_destroy(id, uuid);
7487 pending_inc.new_state[id] = CEPH_OSD_DESTROYED;
7488 pending_inc.new_uuid[id] = uuid_d();
7490 // we can only propose_pending() once per service, otherwise we'll be
7491 // defying PaxosService and all laws of nature. Therefore, as we may
7492 // be used during 'osd purge', let's keep the caller responsible for
7498 int OSDMonitor::prepare_command_osd_purge(
7502 assert(paxos->is_plugged());
7503 dout(10) << __func__ << " purging osd." << id << dendl;
7505 assert(!osdmap.is_up(id));
7508 * This may look a bit weird, but this is what's going to happen:
7510 * 1. we make sure that removing from crush works
7511 * 2. we call `prepare_command_osd_destroy()`. If it returns an
7512 * error, then we abort the whole operation, as no updates
7513 * have been made. However, we this function will have
7514 * side-effects, thus we need to make sure that all operations
7515 * performed henceforth will *always* succeed.
7516 * 3. we call `prepare_command_osd_remove()`. Although this
7517 * function can return an error, it currently only checks if the
7518 * osd is up - and we have made sure that it is not so, so there
7519 * is no conflict, and it is effectively an update.
7520 * 4. finally, we call `do_osd_crush_remove()`, which will perform
7521 * the crush update we delayed from before.
7524 CrushWrapper newcrush;
7525 _get_pending_crush(newcrush);
7527 bool may_be_idempotent = false;
7529 int err = _prepare_command_osd_crush_remove(newcrush, id, 0, false, false);
7530 if (err == -ENOENT) {
7532 may_be_idempotent = true;
7533 } else if (err < 0) {
7534 ss << "error removing osd." << id << " from crush";
7538 // no point destroying the osd again if it has already been marked destroyed
7539 if (!osdmap.is_destroyed(id)) {
7540 err = prepare_command_osd_destroy(id, ss);
7542 if (err == -ENOENT) {
7548 may_be_idempotent = false;
7553 if (may_be_idempotent && !osdmap.exists(id)) {
7554 dout(10) << __func__ << " osd." << id << " does not exist and "
7555 << "we are idempotent." << dendl;
7559 err = prepare_command_osd_remove(id);
7560 // we should not be busy, as we should have made sure this id is not up.
7563 do_osd_crush_remove(newcrush);
7567 bool OSDMonitor::prepare_command_impl(MonOpRequestRef op,
7568 map<string,cmd_vartype> &cmdmap)
7570 op->mark_osdmon_event(__func__);
7571 MMonCommand *m = static_cast<MMonCommand*>(op->get_req());
7579 cmd_getval(g_ceph_context, cmdmap, "format", format, string("plain"));
7580 boost::scoped_ptr<Formatter> f(Formatter::create(format));
7583 cmd_getval(g_ceph_context, cmdmap, "prefix", prefix);
7587 bool osdid_present = cmd_getval(g_ceph_context, cmdmap, "id", osdid);
7588 if (osdid_present) {
7590 oss << "osd." << osdid;
7594 // Even if there's a pending state with changes that could affect
7595 // a command, considering that said state isn't yet committed, we
7596 // just don't care about those changes if the command currently being
7597 // handled acts as a no-op against the current committed state.
7598 // In a nutshell, we assume this command happens *before*.
7600 // Let me make this clearer:
7602 // - If we have only one client, and that client issues some
7603 // operation that would conflict with this operation but is
7604 // still on the pending state, then we would be sure that said
7605 // operation wouldn't have returned yet, so the client wouldn't
7606 // issue this operation (unless the client didn't wait for the
7607 // operation to finish, and that would be the client's own fault).
7609 // - If we have more than one client, each client will observe
7610 // whatever is the state at the moment of the commit. So, if we
7611 // have two clients, one issuing an unlink and another issuing a
7612 // link, and if the link happens while the unlink is still on the
7613 // pending state, from the link's point-of-view this is a no-op.
7614 // If different clients are issuing conflicting operations and
7615 // they care about that, then the clients should make sure they
7616 // enforce some kind of concurrency mechanism -- from our
7617 // perspective that's what Douglas Adams would call an SEP.
7619 // This should be used as a general guideline for most commands handled
7620 // in this function. Adapt as you see fit, but please bear in mind that
7621 // this is the expected behavior.
7624 if (prefix == "osd setcrushmap" ||
7625 (prefix == "osd crush set" && !osdid_present)) {
7626 if (pending_inc.crush.length()) {
7627 dout(10) << __func__ << " waiting for pending crush update " << dendl;
7628 wait_for_finished_proposal(op, new C_RetryMessage(this, op));
7631 dout(10) << "prepare_command setting new crush map" << dendl;
7632 bufferlist data(m->get_data());
7635 bufferlist::iterator bl(data.begin());
7638 catch (const std::exception &e) {
7640 ss << "Failed to parse crushmap: " << e.what();
7644 int64_t prior_version = 0;
7645 if (cmd_getval(g_ceph_context, cmdmap, "prior_version", prior_version)) {
7646 if (prior_version == osdmap.get_crush_version() - 1) {
7647 // see if we are a resend of the last update. this is imperfect
7648 // (multiple racing updaters may not both get reliable success)
7649 // but we expect crush updaters (via this interface) to be rare-ish.
7650 bufferlist current, proposed;
7651 osdmap.crush->encode(current, mon->get_quorum_con_features());
7652 crush.encode(proposed, mon->get_quorum_con_features());
7653 if (current.contents_equal(proposed)) {
7654 dout(10) << __func__
7655 << " proposed matches current and version equals previous"
7658 ss << osdmap.get_crush_version();
7662 if (prior_version != osdmap.get_crush_version()) {
7664 ss << "prior_version " << prior_version << " != crush version "
7665 << osdmap.get_crush_version();
7670 if (crush.has_legacy_rule_ids()) {
7672 ss << "crush maps with ruleset != ruleid are no longer allowed";
7675 if (!validate_crush_against_features(&crush, ss)) {
7680 err = osdmap.validate_crush_rules(&crush, &ss);
7685 if (g_conf->mon_osd_crush_smoke_test) {
7686 // sanity check: test some inputs to make sure this map isn't
7688 dout(10) << " testing map" << dendl;
7690 CrushTester tester(crush, ess);
7691 tester.set_min_x(0);
7692 tester.set_max_x(50);
7693 auto start = ceph::coarse_mono_clock::now();
7694 int r = tester.test_with_fork(g_conf->mon_lease);
7695 auto duration = ceph::coarse_mono_clock::now() - start;
7697 dout(10) << " tester.test_with_fork returns " << r
7698 << ": " << ess.str() << dendl;
7699 ss << "crush smoke test failed with " << r << ": " << ess.str();
7703 dout(10) << __func__ << " crush somke test duration: "
7704 << duration << ", result: " << ess.str() << dendl;
7707 pending_inc.crush = data;
7708 ss << osdmap.get_crush_version() + 1;
7711 } else if (prefix == "osd crush set-all-straw-buckets-to-straw2") {
7712 CrushWrapper newcrush;
7713 _get_pending_crush(newcrush);
7714 for (int b = 0; b < newcrush.get_max_buckets(); ++b) {
7716 if (newcrush.bucket_exists(bid) &&
7717 newcrush.get_bucket_alg(bid)) {
7718 dout(20) << " bucket " << bid << " is straw, can convert" << dendl;
7719 newcrush.bucket_set_alg(bid, CRUSH_BUCKET_STRAW2);
7722 if (!validate_crush_against_features(&newcrush, ss)) {
7726 pending_inc.crush.clear();
7727 newcrush.encode(pending_inc.crush, mon->get_quorum_con_features());
7728 wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs,
7729 get_last_committed() + 1));
7731 } else if (prefix == "osd crush set-device-class") {
7732 if (osdmap.require_osd_release < CEPH_RELEASE_LUMINOUS) {
7733 ss << "you must complete the upgrade and 'ceph osd require-osd-release "
7734 << "luminous' before using crush device classes";
7739 string device_class;
7740 if (!cmd_getval(g_ceph_context, cmdmap, "class", device_class)) {
7741 err = -EINVAL; // no value!
7746 vector<string> idvec;
7747 cmd_getval(g_ceph_context, cmdmap, "ids", idvec);
7748 CrushWrapper newcrush;
7749 _get_pending_crush(newcrush);
7751 for (unsigned j = 0; j < idvec.size() && !stop; j++) {
7755 (idvec[0] == "any" || idvec[0] == "all" || idvec[0] == "*")) {
7756 osdmap.get_all_osds(osds);
7759 // try traditional single osd way
7760 long osd = parse_osd_id(idvec[j].c_str(), &ss);
7762 // ss has reason for failure
7763 ss << ", unable to parse osd id:\"" << idvec[j] << "\". ";
7770 for (auto &osd : osds) {
7771 if (!osdmap.exists(osd)) {
7772 ss << "osd." << osd << " does not exist. ";
7777 oss << "osd." << osd;
7778 string name = oss.str();
7781 if (newcrush.item_exists(osd)) {
7782 action = "updating";
7784 action = "creating";
7785 newcrush.set_item_name(osd, name);
7788 dout(5) << action << " crush item id " << osd << " name '" << name
7789 << "' device_class '" << device_class << "'"
7791 err = newcrush.update_device_class(osd, device_class, name, &ss);
7795 if (err == 0 && !_have_pending_crush()) {
7797 // for single osd only, wildcard makes too much noise
7798 ss << "set-device-class item id " << osd << " name '" << name
7799 << "' device_class '" << device_class << "': no change";
7802 updated.insert(osd);
7807 if (!updated.empty()) {
7808 pending_inc.crush.clear();
7809 newcrush.encode(pending_inc.crush, mon->get_quorum_con_features());
7810 ss << "set osd(s) " << updated << " to class '" << device_class << "'";
7812 wait_for_finished_proposal(op,
7813 new Monitor::C_Command(mon,op, 0, rs, get_last_committed() + 1));
7817 } else if (prefix == "osd crush rm-device-class") {
7819 vector<string> idvec;
7820 cmd_getval(g_ceph_context, cmdmap, "ids", idvec);
7821 CrushWrapper newcrush;
7822 _get_pending_crush(newcrush);
7825 for (unsigned j = 0; j < idvec.size() && !stop; j++) {
7830 (idvec[0] == "any" || idvec[0] == "all" || idvec[0] == "*")) {
7831 osdmap.get_all_osds(osds);
7834 // try traditional single osd way
7835 long osd = parse_osd_id(idvec[j].c_str(), &ss);
7837 // ss has reason for failure
7838 ss << ", unable to parse osd id:\"" << idvec[j] << "\". ";
7845 for (auto &osd : osds) {
7846 if (!osdmap.exists(osd)) {
7847 ss << "osd." << osd << " does not exist. ";
7851 auto class_name = newcrush.get_item_class(osd);
7853 ss << "osd." << osd << " belongs to no class, ";
7856 // note that we do not verify if class_is_in_use here
7857 // in case the device is misclassified and user wants
7858 // to overridely reset...
7860 err = newcrush.remove_device_class(g_ceph_context, osd, &ss);
7862 // ss has reason for failure
7865 updated.insert(osd);
7869 if (!updated.empty()) {
7870 pending_inc.crush.clear();
7871 newcrush.encode(pending_inc.crush, mon->get_quorum_con_features());
7872 ss << "done removing class of osd(s): " << updated;
7874 wait_for_finished_proposal(op,
7875 new Monitor::C_Command(mon,op, 0, rs, get_last_committed() + 1));
7878 } else if (prefix == "osd crush class rename") {
7879 string srcname, dstname;
7880 if (!cmd_getval(g_ceph_context, cmdmap, "srcname", srcname)) {
7884 if (!cmd_getval(g_ceph_context, cmdmap, "dstname", dstname)) {
7889 CrushWrapper newcrush;
7890 _get_pending_crush(newcrush);
7891 if (!newcrush.class_exists(srcname) && newcrush.class_exists(dstname)) {
7892 // suppose this is a replay and return success
7893 // so command is idempotent
7894 ss << "already renamed to '" << dstname << "'";
7899 err = newcrush.rename_class(srcname, dstname);
7901 ss << "fail to rename '" << srcname << "' to '" << dstname << "' : "
7902 << cpp_strerror(err);
7906 pending_inc.crush.clear();
7907 newcrush.encode(pending_inc.crush, mon->get_quorum_con_features());
7908 ss << "rename class '" << srcname << "' to '" << dstname << "'";
7910 } else if (prefix == "osd crush add-bucket") {
7911 // os crush add-bucket <name> <type>
7912 string name, typestr;
7913 cmd_getval(g_ceph_context, cmdmap, "name", name);
7914 cmd_getval(g_ceph_context, cmdmap, "type", typestr);
7916 if (!_have_pending_crush() &&
7917 _get_stable_crush().name_exists(name)) {
7918 ss << "bucket '" << name << "' already exists";
7922 CrushWrapper newcrush;
7923 _get_pending_crush(newcrush);
7925 if (newcrush.name_exists(name)) {
7926 ss << "bucket '" << name << "' already exists";
7929 int type = newcrush.get_type_id(typestr);
7931 ss << "type '" << typestr << "' does not exist";
7936 ss << "type '" << typestr << "' is for devices, not buckets";
7941 err = newcrush.add_bucket(0, 0,
7942 CRUSH_HASH_DEFAULT, type, 0, NULL,
7945 ss << "add_bucket error: '" << cpp_strerror(err) << "'";
7948 err = newcrush.set_item_name(bucketno, name);
7950 ss << "error setting bucket name to '" << name << "'";
7954 pending_inc.crush.clear();
7955 newcrush.encode(pending_inc.crush, mon->get_quorum_con_features());
7956 ss << "added bucket " << name << " type " << typestr
7959 } else if (prefix == "osd crush rename-bucket") {
7960 string srcname, dstname;
7961 cmd_getval(g_ceph_context, cmdmap, "srcname", srcname);
7962 cmd_getval(g_ceph_context, cmdmap, "dstname", dstname);
7964 err = crush_rename_bucket(srcname, dstname, &ss);
7965 if (err == -EALREADY) // equivalent to success for idempotency
7971 } else if (prefix == "osd crush weight-set create" ||
7972 prefix == "osd crush weight-set create-compat") {
7973 CrushWrapper newcrush;
7974 _get_pending_crush(newcrush);
7977 if (newcrush.has_non_straw2_buckets()) {
7978 ss << "crush map contains one or more bucket(s) that are not straw2";
7982 if (prefix == "osd crush weight-set create") {
7983 if (osdmap.require_min_compat_client > 0 &&
7984 osdmap.require_min_compat_client < CEPH_RELEASE_LUMINOUS) {
7985 ss << "require_min_compat_client "
7986 << ceph_release_name(osdmap.require_min_compat_client)
7987 << " < luminous, which is required for per-pool weight-sets. "
7988 << "Try 'ceph osd set-require-min-compat-client luminous' "
7989 << "before using the new interface";
7993 string poolname, mode;
7994 cmd_getval(g_ceph_context, cmdmap, "pool", poolname);
7995 pool = osdmap.lookup_pg_pool_name(poolname.c_str());
7997 ss << "pool '" << poolname << "' not found";
8001 cmd_getval(g_ceph_context, cmdmap, "mode", mode);
8002 if (mode != "flat" && mode != "positional") {
8003 ss << "unrecognized weight-set mode '" << mode << "'";
8007 positions = mode == "flat" ? 1 : osdmap.get_pg_pool(pool)->get_size();
8009 pool = CrushWrapper::DEFAULT_CHOOSE_ARGS;
8012 newcrush.create_choose_args(pool, positions);
8013 pending_inc.crush.clear();
8014 newcrush.encode(pending_inc.crush, mon->get_quorum_con_features());
8017 } else if (prefix == "osd crush weight-set rm" ||
8018 prefix == "osd crush weight-set rm-compat") {
8019 CrushWrapper newcrush;
8020 _get_pending_crush(newcrush);
8022 if (prefix == "osd crush weight-set rm") {
8024 cmd_getval(g_ceph_context, cmdmap, "pool", poolname);
8025 pool = osdmap.lookup_pg_pool_name(poolname.c_str());
8027 ss << "pool '" << poolname << "' not found";
8032 pool = CrushWrapper::DEFAULT_CHOOSE_ARGS;
8034 newcrush.rm_choose_args(pool);
8035 pending_inc.crush.clear();
8036 newcrush.encode(pending_inc.crush, mon->get_quorum_con_features());
8039 } else if (prefix == "osd crush weight-set reweight" ||
8040 prefix == "osd crush weight-set reweight-compat") {
8041 string poolname, item;
8042 vector<double> weight;
8043 cmd_getval(g_ceph_context, cmdmap, "pool", poolname);
8044 cmd_getval(g_ceph_context, cmdmap, "item", item);
8045 cmd_getval(g_ceph_context, cmdmap, "weight", weight);
8046 CrushWrapper newcrush;
8047 _get_pending_crush(newcrush);
8049 if (prefix == "osd crush weight-set reweight") {
8050 pool = osdmap.lookup_pg_pool_name(poolname.c_str());
8052 ss << "pool '" << poolname << "' not found";
8056 if (!newcrush.have_choose_args(pool)) {
8057 ss << "no weight-set for pool '" << poolname << "'";
8061 auto arg_map = newcrush.choose_args_get(pool);
8062 int positions = newcrush.get_choose_args_positions(arg_map);
8063 if (weight.size() != (size_t)positions) {
8064 ss << "must specify exact " << positions << " weight values";
8069 pool = CrushWrapper::DEFAULT_CHOOSE_ARGS;
8070 if (!newcrush.have_choose_args(pool)) {
8071 ss << "no backward-compatible weight-set";
8076 if (!newcrush.name_exists(item)) {
8077 ss << "item '" << item << "' does not exist";
8081 err = newcrush.choose_args_adjust_item_weightf(
8083 newcrush.choose_args_get(pool),
8084 newcrush.get_item_id(item),
8091 pending_inc.crush.clear();
8092 newcrush.encode(pending_inc.crush, mon->get_quorum_con_features());
8094 } else if (osdid_present &&
8095 (prefix == "osd crush set" || prefix == "osd crush add")) {
8096 // <OsdName> is 'osd.<id>' or '<id>', passed as int64_t id
8097 // osd crush set <OsdName> <weight> <loc1> [<loc2> ...]
8098 // osd crush add <OsdName> <weight> <loc1> [<loc2> ...]
8100 if (!osdmap.exists(osdid)) {
8102 ss << name << " does not exist. Create it before updating the crush map";
8107 if (!cmd_getval(g_ceph_context, cmdmap, "weight", weight)) {
8108 ss << "unable to parse weight value '"
8109 << cmd_vartype_stringify(cmdmap["weight"]) << "'";
8115 vector<string> argvec;
8116 cmd_getval(g_ceph_context, cmdmap, "args", argvec);
8117 map<string,string> loc;
8118 CrushWrapper::parse_loc_map(argvec, &loc);
8120 if (prefix == "osd crush set"
8121 && !_get_stable_crush().item_exists(osdid)) {
8123 ss << "unable to set item id " << osdid << " name '" << name
8124 << "' weight " << weight << " at location " << loc
8125 << ": does not exist";
8129 dout(5) << "adding/updating crush item id " << osdid << " name '"
8130 << name << "' weight " << weight << " at location "
8132 CrushWrapper newcrush;
8133 _get_pending_crush(newcrush);
8136 if (prefix == "osd crush set" ||
8137 newcrush.check_item_loc(g_ceph_context, osdid, loc, (int *)NULL)) {
8139 err = newcrush.update_item(g_ceph_context, osdid, weight, name, loc);
8142 err = newcrush.insert_item(g_ceph_context, osdid, weight, name, loc);
8150 if (err == 0 && !_have_pending_crush()) {
8151 ss << action << " item id " << osdid << " name '" << name << "' weight "
8152 << weight << " at location " << loc << ": no change";
8156 pending_inc.crush.clear();
8157 newcrush.encode(pending_inc.crush, mon->get_quorum_con_features());
8158 ss << action << " item id " << osdid << " name '" << name << "' weight "
8159 << weight << " at location " << loc << " to crush map";
8161 wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs,
8162 get_last_committed() + 1));
8165 } else if (prefix == "osd crush create-or-move") {
8167 // osd crush create-or-move <OsdName> <initial_weight> <loc1> [<loc2> ...]
8168 if (!osdmap.exists(osdid)) {
8170 ss << name << " does not exist. create it before updating the crush map";
8175 if (!cmd_getval(g_ceph_context, cmdmap, "weight", weight)) {
8176 ss << "unable to parse weight value '"
8177 << cmd_vartype_stringify(cmdmap["weight"]) << "'";
8183 vector<string> argvec;
8184 cmd_getval(g_ceph_context, cmdmap, "args", argvec);
8185 map<string,string> loc;
8186 CrushWrapper::parse_loc_map(argvec, &loc);
8188 dout(0) << "create-or-move crush item name '" << name << "' initial_weight " << weight
8189 << " at location " << loc << dendl;
8191 CrushWrapper newcrush;
8192 _get_pending_crush(newcrush);
8194 err = newcrush.create_or_move_item(g_ceph_context, osdid, weight, name, loc);
8196 ss << "create-or-move updated item name '" << name << "' weight " << weight
8197 << " at location " << loc << " to crush map";
8201 pending_inc.crush.clear();
8202 newcrush.encode(pending_inc.crush, mon->get_quorum_con_features());
8203 ss << "create-or-move updating item name '" << name << "' weight " << weight
8204 << " at location " << loc << " to crush map";
8206 wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs,
8207 get_last_committed() + 1));
8212 } else if (prefix == "osd crush move") {
8214 // osd crush move <name> <loc1> [<loc2> ...]
8217 vector<string> argvec;
8218 cmd_getval(g_ceph_context, cmdmap, "name", name);
8219 cmd_getval(g_ceph_context, cmdmap, "args", argvec);
8220 map<string,string> loc;
8221 CrushWrapper::parse_loc_map(argvec, &loc);
8223 dout(0) << "moving crush item name '" << name << "' to location " << loc << dendl;
8224 CrushWrapper newcrush;
8225 _get_pending_crush(newcrush);
8227 if (!newcrush.name_exists(name)) {
8229 ss << "item " << name << " does not exist";
8232 int id = newcrush.get_item_id(name);
8234 if (!newcrush.check_item_loc(g_ceph_context, id, loc, (int *)NULL)) {
8236 err = newcrush.create_or_move_item(g_ceph_context, id, 0, name, loc);
8238 err = newcrush.move_bucket(g_ceph_context, id, loc);
8241 ss << "moved item id " << id << " name '" << name << "' to location " << loc << " in crush map";
8242 pending_inc.crush.clear();
8243 newcrush.encode(pending_inc.crush, mon->get_quorum_con_features());
8245 wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs,
8246 get_last_committed() + 1));
8250 ss << "no need to move item id " << id << " name '" << name << "' to location " << loc << " in crush map";
8254 } else if (prefix == "osd crush swap-bucket") {
8255 string source, dest, force;
8256 cmd_getval(g_ceph_context, cmdmap, "source", source);
8257 cmd_getval(g_ceph_context, cmdmap, "dest", dest);
8258 cmd_getval(g_ceph_context, cmdmap, "force", force);
8259 CrushWrapper newcrush;
8260 _get_pending_crush(newcrush);
8261 if (!newcrush.name_exists(source)) {
8262 ss << "source item " << source << " does not exist";
8266 if (!newcrush.name_exists(dest)) {
8267 ss << "dest item " << dest << " does not exist";
8271 int sid = newcrush.get_item_id(source);
8272 int did = newcrush.get_item_id(dest);
8274 if (newcrush.get_immediate_parent_id(sid, &sparent) == 0 &&
8275 force != "--yes-i-really-mean-it") {
8276 ss << "source item " << source << " is not an orphan bucket; pass --yes-i-really-mean-it to proceed anyway";
8280 if (newcrush.get_bucket_alg(sid) != newcrush.get_bucket_alg(did) &&
8281 force != "--yes-i-really-mean-it") {
8282 ss << "source bucket alg " << crush_alg_name(newcrush.get_bucket_alg(sid)) << " != "
8283 << "dest bucket alg " << crush_alg_name(newcrush.get_bucket_alg(did))
8284 << "; pass --yes-i-really-mean-it to proceed anyway";
8288 int r = newcrush.swap_bucket(g_ceph_context, sid, did);
8290 ss << "failed to swap bucket contents: " << cpp_strerror(r);
8294 ss << "swapped bucket of " << source << " to " << dest;
8295 pending_inc.crush.clear();
8296 newcrush.encode(pending_inc.crush, mon->get_quorum_con_features());
8297 wait_for_finished_proposal(op,
8298 new Monitor::C_Command(mon, op, err, ss.str(),
8299 get_last_committed() + 1));
8301 } else if (prefix == "osd crush link") {
8302 // osd crush link <name> <loc1> [<loc2> ...]
8304 cmd_getval(g_ceph_context, cmdmap, "name", name);
8305 vector<string> argvec;
8306 cmd_getval(g_ceph_context, cmdmap, "args", argvec);
8307 map<string,string> loc;
8308 CrushWrapper::parse_loc_map(argvec, &loc);
8310 // Need an explicit check for name_exists because get_item_id returns
8312 int id = osdmap.crush->get_item_id(name);
8313 if (!osdmap.crush->name_exists(name)) {
8315 ss << "item " << name << " does not exist";
8318 dout(5) << "resolved crush name '" << name << "' to id " << id << dendl;
8320 if (osdmap.crush->check_item_loc(g_ceph_context, id, loc, (int*) NULL)) {
8321 ss << "no need to move item id " << id << " name '" << name
8322 << "' to location " << loc << " in crush map";
8327 dout(5) << "linking crush item name '" << name << "' at location " << loc << dendl;
8328 CrushWrapper newcrush;
8329 _get_pending_crush(newcrush);
8331 if (!newcrush.name_exists(name)) {
8333 ss << "item " << name << " does not exist";
8336 int id = newcrush.get_item_id(name);
8337 if (!newcrush.check_item_loc(g_ceph_context, id, loc, (int *)NULL)) {
8338 err = newcrush.link_bucket(g_ceph_context, id, loc);
8340 ss << "linked item id " << id << " name '" << name
8341 << "' to location " << loc << " in crush map";
8342 pending_inc.crush.clear();
8343 newcrush.encode(pending_inc.crush, mon->get_quorum_con_features());
8345 ss << "cannot link item id " << id << " name '" << name
8346 << "' to location " << loc;
8350 ss << "no need to move item id " << id << " name '" << name
8351 << "' to location " << loc << " in crush map";
8355 wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, err, ss.str(),
8356 get_last_committed() + 1));
8358 } else if (prefix == "osd crush rm" ||
8359 prefix == "osd crush remove" ||
8360 prefix == "osd crush unlink") {
8362 // osd crush rm <id> [ancestor]
8363 CrushWrapper newcrush;
8364 _get_pending_crush(newcrush);
8367 cmd_getval(g_ceph_context, cmdmap, "name", name);
8369 if (!osdmap.crush->name_exists(name)) {
8371 ss << "device '" << name << "' does not appear in the crush map";
8374 if (!newcrush.name_exists(name)) {
8376 ss << "device '" << name << "' does not appear in the crush map";
8378 wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs,
8379 get_last_committed() + 1));
8382 int id = newcrush.get_item_id(name);
8385 bool unlink_only = prefix == "osd crush unlink";
8386 string ancestor_str;
8387 if (cmd_getval(g_ceph_context, cmdmap, "ancestor", ancestor_str)) {
8388 if (!newcrush.name_exists(ancestor_str)) {
8390 ss << "ancestor item '" << ancestor_str
8391 << "' does not appear in the crush map";
8394 ancestor = newcrush.get_item_id(ancestor_str);
8397 err = prepare_command_osd_crush_remove(
8400 (ancestor < 0), unlink_only);
8402 if (err == -ENOENT) {
8403 ss << "item " << id << " does not appear in that position";
8408 ss << "removed item id " << id << " name '" << name << "' from crush map";
8410 wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs,
8411 get_last_committed() + 1));
8416 } else if (prefix == "osd crush reweight-all") {
8417 CrushWrapper newcrush;
8418 _get_pending_crush(newcrush);
8420 newcrush.reweight(g_ceph_context);
8421 pending_inc.crush.clear();
8422 newcrush.encode(pending_inc.crush, mon->get_quorum_con_features());
8423 ss << "reweighted crush hierarchy";
8425 wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs,
8426 get_last_committed() + 1));
8428 } else if (prefix == "osd crush reweight") {
8429 // osd crush reweight <name> <weight>
8430 CrushWrapper newcrush;
8431 _get_pending_crush(newcrush);
8434 cmd_getval(g_ceph_context, cmdmap, "name", name);
8435 if (!newcrush.name_exists(name)) {
8437 ss << "device '" << name << "' does not appear in the crush map";
8441 int id = newcrush.get_item_id(name);
8443 ss << "device '" << name << "' is not a leaf in the crush map";
8448 if (!cmd_getval(g_ceph_context, cmdmap, "weight", w)) {
8449 ss << "unable to parse weight value '"
8450 << cmd_vartype_stringify(cmdmap["weight"]) << "'";
8455 err = newcrush.adjust_item_weightf(g_ceph_context, id, w);
8458 pending_inc.crush.clear();
8459 newcrush.encode(pending_inc.crush, mon->get_quorum_con_features());
8460 ss << "reweighted item id " << id << " name '" << name << "' to " << w
8463 wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs,
8464 get_last_committed() + 1));
8466 } else if (prefix == "osd crush reweight-subtree") {
8467 // osd crush reweight <name> <weight>
8468 CrushWrapper newcrush;
8469 _get_pending_crush(newcrush);
8472 cmd_getval(g_ceph_context, cmdmap, "name", name);
8473 if (!newcrush.name_exists(name)) {
8475 ss << "device '" << name << "' does not appear in the crush map";
8479 int id = newcrush.get_item_id(name);
8481 ss << "device '" << name << "' is not a subtree in the crush map";
8486 if (!cmd_getval(g_ceph_context, cmdmap, "weight", w)) {
8487 ss << "unable to parse weight value '"
8488 << cmd_vartype_stringify(cmdmap["weight"]) << "'";
8493 err = newcrush.adjust_subtree_weightf(g_ceph_context, id, w);
8496 pending_inc.crush.clear();
8497 newcrush.encode(pending_inc.crush, mon->get_quorum_con_features());
8498 ss << "reweighted subtree id " << id << " name '" << name << "' to " << w
8501 wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs,
8502 get_last_committed() + 1));
8504 } else if (prefix == "osd crush tunables") {
8505 CrushWrapper newcrush;
8506 _get_pending_crush(newcrush);
8510 cmd_getval(g_ceph_context, cmdmap, "profile", profile);
8511 if (profile == "legacy" || profile == "argonaut") {
8512 newcrush.set_tunables_legacy();
8513 } else if (profile == "bobtail") {
8514 newcrush.set_tunables_bobtail();
8515 } else if (profile == "firefly") {
8516 newcrush.set_tunables_firefly();
8517 } else if (profile == "hammer") {
8518 newcrush.set_tunables_hammer();
8519 } else if (profile == "jewel") {
8520 newcrush.set_tunables_jewel();
8521 } else if (profile == "optimal") {
8522 newcrush.set_tunables_optimal();
8523 } else if (profile == "default") {
8524 newcrush.set_tunables_default();
8526 ss << "unrecognized profile '" << profile << "'";
8531 if (!validate_crush_against_features(&newcrush, ss)) {
8536 pending_inc.crush.clear();
8537 newcrush.encode(pending_inc.crush, mon->get_quorum_con_features());
8538 ss << "adjusted tunables profile to " << profile;
8540 wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs,
8541 get_last_committed() + 1));
8543 } else if (prefix == "osd crush set-tunable") {
8544 CrushWrapper newcrush;
8545 _get_pending_crush(newcrush);
8549 cmd_getval(g_ceph_context, cmdmap, "tunable", tunable);
8552 if (!cmd_getval(g_ceph_context, cmdmap, "value", value)) {
8554 ss << "failed to parse integer value " << cmd_vartype_stringify(cmdmap["value"]);
8558 if (tunable == "straw_calc_version") {
8559 if (value != 0 && value != 1) {
8560 ss << "value must be 0 or 1; got " << value;
8564 newcrush.set_straw_calc_version(value);
8566 ss << "unrecognized tunable '" << tunable << "'";
8571 if (!validate_crush_against_features(&newcrush, ss)) {
8576 pending_inc.crush.clear();
8577 newcrush.encode(pending_inc.crush, mon->get_quorum_con_features());
8578 ss << "adjusted tunable " << tunable << " to " << value;
8580 wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs,
8581 get_last_committed() + 1));
8584 } else if (prefix == "osd crush rule create-simple") {
8585 string name, root, type, mode;
8586 cmd_getval(g_ceph_context, cmdmap, "name", name);
8587 cmd_getval(g_ceph_context, cmdmap, "root", root);
8588 cmd_getval(g_ceph_context, cmdmap, "type", type);
8589 cmd_getval(g_ceph_context, cmdmap, "mode", mode);
8593 if (osdmap.crush->rule_exists(name)) {
8594 // The name is uniquely associated to a ruleid and the rule it contains
8595 // From the user point of view, the rule is more meaningfull.
8596 ss << "rule " << name << " already exists";
8601 CrushWrapper newcrush;
8602 _get_pending_crush(newcrush);
8604 if (newcrush.rule_exists(name)) {
8605 // The name is uniquely associated to a ruleid and the rule it contains
8606 // From the user point of view, the rule is more meaningfull.
8607 ss << "rule " << name << " already exists";
8610 int ruleno = newcrush.add_simple_rule(name, root, type, "", mode,
8611 pg_pool_t::TYPE_REPLICATED, &ss);
8617 pending_inc.crush.clear();
8618 newcrush.encode(pending_inc.crush, mon->get_quorum_con_features());
8621 wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs,
8622 get_last_committed() + 1));
8625 } else if (prefix == "osd crush rule create-replicated") {
8626 string name, root, type, device_class;
8627 cmd_getval(g_ceph_context, cmdmap, "name", name);
8628 cmd_getval(g_ceph_context, cmdmap, "root", root);
8629 cmd_getval(g_ceph_context, cmdmap, "type", type);
8630 cmd_getval(g_ceph_context, cmdmap, "class", device_class);
8632 if (!device_class.empty()) {
8633 if (osdmap.require_osd_release < CEPH_RELEASE_LUMINOUS) {
8634 ss << "you must complete the upgrade and 'ceph osd require-osd-release "
8635 << "luminous' before using crush device classes";
8641 if (osdmap.crush->rule_exists(name)) {
8642 // The name is uniquely associated to a ruleid and the rule it contains
8643 // From the user point of view, the rule is more meaningfull.
8644 ss << "rule " << name << " already exists";
8649 CrushWrapper newcrush;
8650 _get_pending_crush(newcrush);
8652 if (newcrush.rule_exists(name)) {
8653 // The name is uniquely associated to a ruleid and the rule it contains
8654 // From the user point of view, the rule is more meaningfull.
8655 ss << "rule " << name << " already exists";
8658 int ruleno = newcrush.add_simple_rule(
8659 name, root, type, device_class,
8660 "firstn", pg_pool_t::TYPE_REPLICATED, &ss);
8666 pending_inc.crush.clear();
8667 newcrush.encode(pending_inc.crush, mon->get_quorum_con_features());
8670 wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs,
8671 get_last_committed() + 1));
8674 } else if (prefix == "osd erasure-code-profile rm") {
8676 cmd_getval(g_ceph_context, cmdmap, "name", name);
8678 if (erasure_code_profile_in_use(pending_inc.new_pools, name, &ss))
8681 if (erasure_code_profile_in_use(osdmap.pools, name, &ss)) {
8686 if (osdmap.has_erasure_code_profile(name) ||
8687 pending_inc.new_erasure_code_profiles.count(name)) {
8688 if (osdmap.has_erasure_code_profile(name)) {
8689 pending_inc.old_erasure_code_profiles.push_back(name);
8691 dout(20) << "erasure code profile rm " << name << ": creation canceled" << dendl;
8692 pending_inc.new_erasure_code_profiles.erase(name);
8696 wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs,
8697 get_last_committed() + 1));
8700 ss << "erasure-code-profile " << name << " does not exist";
8705 } else if (prefix == "osd erasure-code-profile set") {
8707 cmd_getval(g_ceph_context, cmdmap, "name", name);
8708 vector<string> profile;
8709 cmd_getval(g_ceph_context, cmdmap, "profile", profile);
8711 if (profile.size() > 0 && profile.back() == "--force") {
8717 map<string,string> profile_map;
8718 err = parse_erasure_code_profile(profile, &profile_map, &ss);
8721 if (profile_map.find("plugin") == profile_map.end()) {
8722 ss << "erasure-code-profile " << profile_map
8723 << " must contain a plugin entry" << std::endl;
8727 string plugin = profile_map["plugin"];
8729 if (pending_inc.has_erasure_code_profile(name)) {
8730 dout(20) << "erasure code profile " << name << " try again" << dendl;
8733 if (plugin == "isa" || plugin == "lrc") {
8734 err = check_cluster_features(CEPH_FEATURE_ERASURE_CODE_PLUGINS_V2, ss);
8739 } else if (plugin == "shec") {
8740 err = check_cluster_features(CEPH_FEATURE_ERASURE_CODE_PLUGINS_V3, ss);
8746 err = normalize_profile(name, profile_map, force, &ss);
8750 if (osdmap.has_erasure_code_profile(name)) {
8751 ErasureCodeProfile existing_profile_map =
8752 osdmap.get_erasure_code_profile(name);
8753 err = normalize_profile(name, existing_profile_map, force, &ss);
8757 if (existing_profile_map == profile_map) {
8763 ss << "will not override erasure code profile " << name
8764 << " because the existing profile "
8765 << existing_profile_map
8766 << " is different from the proposed profile "
8772 dout(20) << "erasure code profile set " << name << "="
8773 << profile_map << dendl;
8774 pending_inc.set_erasure_code_profile(name, profile_map);
8778 wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs,
8779 get_last_committed() + 1));
8782 } else if (prefix == "osd crush rule create-erasure") {
8783 err = check_cluster_features(CEPH_FEATURE_CRUSH_V2, ss);
8788 string name, poolstr;
8789 cmd_getval(g_ceph_context, cmdmap, "name", name);
8791 cmd_getval(g_ceph_context, cmdmap, "profile", profile);
8793 profile = "default";
8794 if (profile == "default") {
8795 if (!osdmap.has_erasure_code_profile(profile)) {
8796 if (pending_inc.has_erasure_code_profile(profile)) {
8797 dout(20) << "erasure code profile " << profile << " already pending" << dendl;
8801 map<string,string> profile_map;
8802 err = osdmap.get_erasure_code_profile_default(g_ceph_context,
8807 err = normalize_profile(name, profile_map, true, &ss);
8810 dout(20) << "erasure code profile set " << profile << "="
8811 << profile_map << dendl;
8812 pending_inc.set_erasure_code_profile(profile, profile_map);
8818 err = crush_rule_create_erasure(name, profile, &rule, &ss);
8821 case -EEXIST: // return immediately
8822 ss << "rule " << name << " already exists";
8826 case -EALREADY: // wait for pending to be proposed
8827 ss << "rule " << name << " already exists";
8830 default: // non recoverable error
8835 ss << "created rule " << name << " at " << rule;
8839 wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs,
8840 get_last_committed() + 1));
8843 } else if (prefix == "osd crush rule rm") {
8845 cmd_getval(g_ceph_context, cmdmap, "name", name);
8847 if (!osdmap.crush->rule_exists(name)) {
8848 ss << "rule " << name << " does not exist";
8853 CrushWrapper newcrush;
8854 _get_pending_crush(newcrush);
8856 if (!newcrush.rule_exists(name)) {
8857 ss << "rule " << name << " does not exist";
8860 int ruleno = newcrush.get_rule_id(name);
8861 assert(ruleno >= 0);
8863 // make sure it is not in use.
8864 // FIXME: this is ok in some situations, but let's not bother with that
8866 int ruleset = newcrush.get_rule_mask_ruleset(ruleno);
8867 if (osdmap.crush_rule_in_use(ruleset)) {
8868 ss << "crush ruleset " << name << " " << ruleset << " is in use";
8873 err = newcrush.remove_rule(ruleno);
8878 pending_inc.crush.clear();
8879 newcrush.encode(pending_inc.crush, mon->get_quorum_con_features());
8882 wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs,
8883 get_last_committed() + 1));
8886 } else if (prefix == "osd crush rule rename") {
8889 cmd_getval(g_ceph_context, cmdmap, "srcname", srcname);
8890 cmd_getval(g_ceph_context, cmdmap, "dstname", dstname);
8891 if (srcname.empty() || dstname.empty()) {
8892 ss << "must specify both source rule name and destination rule name";
8896 if (srcname == dstname) {
8897 ss << "destination rule name is equal to source rule name";
8902 CrushWrapper newcrush;
8903 _get_pending_crush(newcrush);
8904 if (!newcrush.rule_exists(srcname) && newcrush.rule_exists(dstname)) {
8905 // srcname does not exist and dstname already exists
8906 // suppose this is a replay and return success
8907 // (so this command is idempotent)
8908 ss << "already renamed to '" << dstname << "'";
8913 err = newcrush.rename_rule(srcname, dstname, &ss);
8915 // ss has reason for failure
8918 pending_inc.crush.clear();
8919 newcrush.encode(pending_inc.crush, mon->get_quorum_con_features());
8921 wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs,
8922 get_last_committed() + 1));
8925 } else if (prefix == "osd setmaxosd") {
8927 if (!cmd_getval(g_ceph_context, cmdmap, "newmax", newmax)) {
8928 ss << "unable to parse 'newmax' value '"
8929 << cmd_vartype_stringify(cmdmap["newmax"]) << "'";
8934 if (newmax > g_conf->mon_max_osd) {
8936 ss << "cannot set max_osd to " << newmax << " which is > conf.mon_max_osd ("
8937 << g_conf->mon_max_osd << ")";
8941 // Don't allow shrinking OSD number as this will cause data loss
8942 // and may cause kernel crashes.
8943 // Note: setmaxosd sets the maximum OSD number and not the number of OSDs
8944 if (newmax < osdmap.get_max_osd()) {
8945 // Check if the OSDs exist between current max and new value.
8946 // If there are any OSDs exist, then don't allow shrinking number
8948 for (int i = newmax; i < osdmap.get_max_osd(); i++) {
8949 if (osdmap.exists(i)) {
8951 ss << "cannot shrink max_osd to " << newmax
8952 << " because osd." << i << " (and possibly others) still in use";
8958 pending_inc.new_max_osd = newmax;
8959 ss << "set new max_osd = " << pending_inc.new_max_osd;
8961 wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs,
8962 get_last_committed() + 1));
8965 } else if (prefix == "osd set-full-ratio" ||
8966 prefix == "osd set-backfillfull-ratio" ||
8967 prefix == "osd set-nearfull-ratio") {
8968 if (osdmap.require_osd_release < CEPH_RELEASE_LUMINOUS) {
8969 ss << "you must complete the upgrade and 'ceph osd require-osd-release "
8970 << "luminous' before using the new interface";
8975 if (!cmd_getval(g_ceph_context, cmdmap, "ratio", n)) {
8976 ss << "unable to parse 'ratio' value '"
8977 << cmd_vartype_stringify(cmdmap["ratio"]) << "'";
8981 if (prefix == "osd set-full-ratio")
8982 pending_inc.new_full_ratio = n;
8983 else if (prefix == "osd set-backfillfull-ratio")
8984 pending_inc.new_backfillfull_ratio = n;
8985 else if (prefix == "osd set-nearfull-ratio")
8986 pending_inc.new_nearfull_ratio = n;
8987 ss << prefix << " " << n;
8989 wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs,
8990 get_last_committed() + 1));
8992 } else if (prefix == "osd set-require-min-compat-client") {
8993 if (osdmap.require_osd_release < CEPH_RELEASE_LUMINOUS) {
8994 ss << "you must complete the upgrade and 'ceph osd require-osd-release "
8995 << "luminous' before using the new interface";
9000 cmd_getval(g_ceph_context, cmdmap, "version", v);
9001 int vno = ceph_release_from_name(v.c_str());
9003 ss << "version " << v << " is not recognized";
9008 newmap.deepish_copy_from(osdmap);
9009 newmap.apply_incremental(pending_inc);
9010 newmap.require_min_compat_client = vno;
9011 auto mvno = newmap.get_min_compat_client();
9013 ss << "osdmap current utilizes features that require "
9014 << ceph_release_name(mvno)
9015 << "; cannot set require_min_compat_client below that to "
9016 << ceph_release_name(vno);
9021 cmd_getval(g_ceph_context, cmdmap, "sure", sure);
9022 if (sure != "--yes-i-really-mean-it") {
9024 mon->get_combined_feature_map(&m);
9025 uint64_t features = ceph_release_features(vno);
9029 CEPH_ENTITY_TYPE_CLIENT,
9030 CEPH_ENTITY_TYPE_MDS,
9031 CEPH_ENTITY_TYPE_MGR }) {
9032 auto p = m.m.find(type);
9033 if (p == m.m.end()) {
9036 for (auto& q : p->second) {
9037 uint64_t missing = ~q.first & features;
9040 ss << "cannot set require_min_compat_client to " << v << ": ";
9045 ss << q.second << " connected " << ceph_entity_type_name(type)
9046 << "(s) look like " << ceph_release_name(
9047 ceph_release_from_features(q.first))
9048 << " (missing 0x" << std::hex << missing << std::dec << ")";
9054 ss << "; add --yes-i-really-mean-it to do it anyway";
9059 ss << "set require_min_compat_client to " << ceph_release_name(vno);
9060 pending_inc.new_require_min_compat_client = vno;
9062 wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs,
9063 get_last_committed() + 1));
9065 } else if (prefix == "osd pause") {
9066 return prepare_set_flag(op, CEPH_OSDMAP_PAUSERD | CEPH_OSDMAP_PAUSEWR);
9068 } else if (prefix == "osd unpause") {
9069 return prepare_unset_flag(op, CEPH_OSDMAP_PAUSERD | CEPH_OSDMAP_PAUSEWR);
9071 } else if (prefix == "osd set") {
9073 cmd_getval(g_ceph_context, cmdmap, "sure", sure);
9075 cmd_getval(g_ceph_context, cmdmap, "key", key);
9077 return prepare_set_flag(op, CEPH_OSDMAP_FULL);
9078 else if (key == "pause")
9079 return prepare_set_flag(op, CEPH_OSDMAP_PAUSERD | CEPH_OSDMAP_PAUSEWR);
9080 else if (key == "noup")
9081 return prepare_set_flag(op, CEPH_OSDMAP_NOUP);
9082 else if (key == "nodown")
9083 return prepare_set_flag(op, CEPH_OSDMAP_NODOWN);
9084 else if (key == "noout")
9085 return prepare_set_flag(op, CEPH_OSDMAP_NOOUT);
9086 else if (key == "noin")
9087 return prepare_set_flag(op, CEPH_OSDMAP_NOIN);
9088 else if (key == "nobackfill")
9089 return prepare_set_flag(op, CEPH_OSDMAP_NOBACKFILL);
9090 else if (key == "norebalance")
9091 return prepare_set_flag(op, CEPH_OSDMAP_NOREBALANCE);
9092 else if (key == "norecover")
9093 return prepare_set_flag(op, CEPH_OSDMAP_NORECOVER);
9094 else if (key == "noscrub")
9095 return prepare_set_flag(op, CEPH_OSDMAP_NOSCRUB);
9096 else if (key == "nodeep-scrub")
9097 return prepare_set_flag(op, CEPH_OSDMAP_NODEEP_SCRUB);
9098 else if (key == "notieragent")
9099 return prepare_set_flag(op, CEPH_OSDMAP_NOTIERAGENT);
9100 else if (key == "sortbitwise") {
9101 if (!osdmap.get_num_up_osds() && sure != "--yes-i-really-mean-it") {
9102 ss << "Not advisable to continue since no OSDs are up. Pass "
9103 << "--yes-i-really-mean-it if you really wish to continue.";
9107 if ((osdmap.get_up_osd_features() & CEPH_FEATURE_OSD_BITWISE_HOBJ_SORT)
9108 || sure == "--yes-i-really-mean-it") {
9109 return prepare_set_flag(op, CEPH_OSDMAP_SORTBITWISE);
9111 ss << "not all up OSDs have OSD_BITWISE_HOBJ_SORT feature";
9115 } else if (key == "recovery_deletes") {
9116 if (!osdmap.get_num_up_osds() && sure != "--yes-i-really-mean-it") {
9117 ss << "Not advisable to continue since no OSDs are up. Pass "
9118 << "--yes-i-really-mean-it if you really wish to continue.";
9122 if (HAVE_FEATURE(osdmap.get_up_osd_features(), OSD_RECOVERY_DELETES)
9123 || sure == "--yes-i-really-mean-it") {
9124 return prepare_set_flag(op, CEPH_OSDMAP_RECOVERY_DELETES);
9126 ss << "not all up OSDs have OSD_RECOVERY_DELETES feature";
9130 } else if (key == "require_jewel_osds") {
9131 if (!osdmap.get_num_up_osds() && sure != "--yes-i-really-mean-it") {
9132 ss << "Not advisable to continue since no OSDs are up. Pass "
9133 << "--yes-i-really-mean-it if you really wish to continue.";
9137 if (!osdmap.test_flag(CEPH_OSDMAP_SORTBITWISE)) {
9138 ss << "the sortbitwise flag must be set before require_jewel_osds";
9141 } else if (osdmap.require_osd_release >= CEPH_RELEASE_JEWEL) {
9142 ss << "require_osd_release is already >= jewel";
9145 } else if (HAVE_FEATURE(osdmap.get_up_osd_features(), SERVER_JEWEL)
9146 || sure == "--yes-i-really-mean-it") {
9147 return prepare_set_flag(op, CEPH_OSDMAP_REQUIRE_JEWEL);
9149 ss << "not all up OSDs have CEPH_FEATURE_SERVER_JEWEL feature";
9152 } else if (key == "require_kraken_osds") {
9153 if (!osdmap.get_num_up_osds() && sure != "--yes-i-really-mean-it") {
9154 ss << "Not advisable to continue since no OSDs are up. Pass "
9155 << "--yes-i-really-mean-it if you really wish to continue.";
9159 if (!osdmap.test_flag(CEPH_OSDMAP_SORTBITWISE)) {
9160 ss << "the sortbitwise flag must be set before require_kraken_osds";
9163 } else if (osdmap.require_osd_release >= CEPH_RELEASE_KRAKEN) {
9164 ss << "require_osd_release is already >= kraken";
9167 } else if (HAVE_FEATURE(osdmap.get_up_osd_features(), SERVER_KRAKEN)
9168 || sure == "--yes-i-really-mean-it") {
9169 bool r = prepare_set_flag(op, CEPH_OSDMAP_REQUIRE_KRAKEN);
9170 // ensure JEWEL is also set
9171 pending_inc.new_flags |= CEPH_OSDMAP_REQUIRE_JEWEL;
9174 ss << "not all up OSDs have CEPH_FEATURE_SERVER_KRAKEN feature";
9178 ss << "unrecognized flag '" << key << "'";
9182 } else if (prefix == "osd unset") {
9184 cmd_getval(g_ceph_context, cmdmap, "key", key);
9186 return prepare_unset_flag(op, CEPH_OSDMAP_FULL);
9187 else if (key == "pause")
9188 return prepare_unset_flag(op, CEPH_OSDMAP_PAUSERD | CEPH_OSDMAP_PAUSEWR);
9189 else if (key == "noup")
9190 return prepare_unset_flag(op, CEPH_OSDMAP_NOUP);
9191 else if (key == "nodown")
9192 return prepare_unset_flag(op, CEPH_OSDMAP_NODOWN);
9193 else if (key == "noout")
9194 return prepare_unset_flag(op, CEPH_OSDMAP_NOOUT);
9195 else if (key == "noin")
9196 return prepare_unset_flag(op, CEPH_OSDMAP_NOIN);
9197 else if (key == "nobackfill")
9198 return prepare_unset_flag(op, CEPH_OSDMAP_NOBACKFILL);
9199 else if (key == "norebalance")
9200 return prepare_unset_flag(op, CEPH_OSDMAP_NOREBALANCE);
9201 else if (key == "norecover")
9202 return prepare_unset_flag(op, CEPH_OSDMAP_NORECOVER);
9203 else if (key == "noscrub")
9204 return prepare_unset_flag(op, CEPH_OSDMAP_NOSCRUB);
9205 else if (key == "nodeep-scrub")
9206 return prepare_unset_flag(op, CEPH_OSDMAP_NODEEP_SCRUB);
9207 else if (key == "notieragent")
9208 return prepare_unset_flag(op, CEPH_OSDMAP_NOTIERAGENT);
9210 ss << "unrecognized flag '" << key << "'";
9214 } else if (prefix == "osd require-osd-release") {
9216 cmd_getval(g_ceph_context, cmdmap, "release", release);
9218 cmd_getval(g_ceph_context, cmdmap, "sure", sure);
9219 if (!osdmap.test_flag(CEPH_OSDMAP_SORTBITWISE)) {
9220 ss << "the sortbitwise flag must be set first";
9224 int rel = ceph_release_from_name(release.c_str());
9226 ss << "unrecognized release " << release;
9230 if (rel < CEPH_RELEASE_LUMINOUS) {
9231 ss << "use this command only for luminous and later";
9235 if (rel == osdmap.require_osd_release) {
9240 if (rel == CEPH_RELEASE_LUMINOUS) {
9241 if (!HAVE_FEATURE(osdmap.get_up_osd_features(), SERVER_LUMINOUS)) {
9242 ss << "not all up OSDs have CEPH_FEATURE_SERVER_LUMINOUS feature";
9247 ss << "not supported for this release yet";
9251 if (rel < osdmap.require_osd_release) {
9252 ss << "require_osd_release cannot be lowered once it has been set";
9256 pending_inc.new_require_osd_release = rel;
9257 if (rel >= CEPH_RELEASE_LUMINOUS &&
9258 !osdmap.test_flag(CEPH_OSDMAP_RECOVERY_DELETES)) {
9259 return prepare_set_flag(op, CEPH_OSDMAP_RECOVERY_DELETES);
9262 } else if (prefix == "osd cluster_snap") {
9263 // ** DISABLE THIS FOR NOW **
9264 ss << "cluster snapshot currently disabled (broken implementation)";
9265 // ** DISABLE THIS FOR NOW **
9267 } else if (prefix == "osd down" ||
9268 prefix == "osd out" ||
9269 prefix == "osd in" ||
9270 prefix == "osd rm") {
9274 bool verbose = true;
9276 vector<string> idvec;
9277 cmd_getval(g_ceph_context, cmdmap, "ids", idvec);
9278 for (unsigned j = 0; j < idvec.size() && !stop; j++) {
9283 (idvec[0] == "any" || idvec[0] == "all" || idvec[0] == "*")) {
9284 if (prefix == "osd in") {
9285 // touch out osds only
9286 osdmap.get_out_osds(osds);
9288 osdmap.get_all_osds(osds);
9291 verbose = false; // so the output is less noisy.
9293 long osd = parse_osd_id(idvec[j].c_str(), &ss);
9295 ss << "invalid osd id" << osd;
9298 } else if (!osdmap.exists(osd)) {
9299 ss << "osd." << osd << " does not exist. ";
9306 for (auto &osd : osds) {
9307 if (prefix == "osd down") {
9308 if (osdmap.is_down(osd)) {
9310 ss << "osd." << osd << " is already down. ";
9312 pending_inc.pending_osd_state_set(osd, CEPH_OSD_UP);
9313 ss << "marked down osd." << osd << ". ";
9316 } else if (prefix == "osd out") {
9317 if (osdmap.is_out(osd)) {
9319 ss << "osd." << osd << " is already out. ";
9321 pending_inc.new_weight[osd] = CEPH_OSD_OUT;
9322 if (osdmap.osd_weight[osd]) {
9323 if (pending_inc.new_xinfo.count(osd) == 0) {
9324 pending_inc.new_xinfo[osd] = osdmap.osd_xinfo[osd];
9326 pending_inc.new_xinfo[osd].old_weight = osdmap.osd_weight[osd];
9328 ss << "marked out osd." << osd << ". ";
9329 std::ostringstream msg;
9330 msg << "Client " << op->get_session()->entity_name
9331 << " marked osd." << osd << " out";
9332 if (osdmap.is_up(osd)) {
9333 msg << ", while it was still marked up";
9335 auto period = ceph_clock_now() - down_pending_out[osd];
9336 msg << ", after it was down for " << int(period.sec())
9340 mon->clog->info() << msg.str();
9343 } else if (prefix == "osd in") {
9344 if (osdmap.is_in(osd)) {
9346 ss << "osd." << osd << " is already in. ";
9348 if (osdmap.osd_xinfo[osd].old_weight > 0) {
9349 pending_inc.new_weight[osd] = osdmap.osd_xinfo[osd].old_weight;
9350 if (pending_inc.new_xinfo.count(osd) == 0) {
9351 pending_inc.new_xinfo[osd] = osdmap.osd_xinfo[osd];
9353 pending_inc.new_xinfo[osd].old_weight = 0;
9355 pending_inc.new_weight[osd] = CEPH_OSD_IN;
9357 ss << "marked in osd." << osd << ". ";
9360 } else if (prefix == "osd rm") {
9361 err = prepare_command_osd_remove(osd);
9363 if (err == -EBUSY) {
9366 ss << "osd." << osd << " is still up; must be down before removal. ";
9370 ss << ", osd." << osd;
9372 ss << "removed osd." << osd;
9381 wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, err, rs,
9382 get_last_committed() + 1));
9385 } else if (prefix == "osd add-noup" ||
9386 prefix == "osd add-nodown" ||
9387 prefix == "osd add-noin" ||
9388 prefix == "osd add-noout") {
9397 if (prefix == "osd add-noup") {
9399 } else if (prefix == "osd add-nodown") {
9401 } else if (prefix == "osd add-noin") {
9410 vector<string> idvec;
9411 cmd_getval(g_ceph_context, cmdmap, "ids", idvec);
9412 for (unsigned j = 0; j < idvec.size() && !stop; j++) {
9418 (idvec[0] == "any" || idvec[0] == "all" || idvec[0] == "*")) {
9419 osdmap.get_all_osds(osds);
9422 // try traditional single osd way
9424 long osd = parse_osd_id(idvec[j].c_str(), &ss);
9426 // ss has reason for failure
9427 ss << ", unable to parse osd id:\"" << idvec[j] << "\". ";
9435 for (auto &osd : osds) {
9437 if (!osdmap.exists(osd)) {
9438 ss << "osd." << osd << " does not exist. ";
9444 if (osdmap.is_up(osd)) {
9445 ss << "osd." << osd << " is already up. ";
9449 if (osdmap.is_noup(osd)) {
9450 if (pending_inc.pending_osd_state_clear(osd, CEPH_OSD_NOUP))
9453 pending_inc.pending_osd_state_set(osd, CEPH_OSD_NOUP);
9460 if (osdmap.is_down(osd)) {
9461 ss << "osd." << osd << " is already down. ";
9465 if (osdmap.is_nodown(osd)) {
9466 if (pending_inc.pending_osd_state_clear(osd, CEPH_OSD_NODOWN))
9469 pending_inc.pending_osd_state_set(osd, CEPH_OSD_NODOWN);
9476 if (osdmap.is_in(osd)) {
9477 ss << "osd." << osd << " is already in. ";
9481 if (osdmap.is_noin(osd)) {
9482 if (pending_inc.pending_osd_state_clear(osd, CEPH_OSD_NOIN))
9485 pending_inc.pending_osd_state_set(osd, CEPH_OSD_NOIN);
9492 if (osdmap.is_out(osd)) {
9493 ss << "osd." << osd << " is already out. ";
9497 if (osdmap.is_noout(osd)) {
9498 if (pending_inc.pending_osd_state_clear(osd, CEPH_OSD_NOOUT))
9501 pending_inc.pending_osd_state_set(osd, CEPH_OSD_NOOUT);
9508 assert(0 == "invalid option");
9515 wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, err, rs,
9516 get_last_committed() + 1));
9519 } else if (prefix == "osd rm-noup" ||
9520 prefix == "osd rm-nodown" ||
9521 prefix == "osd rm-noin" ||
9522 prefix == "osd rm-noout") {
9531 if (prefix == "osd rm-noup") {
9533 } else if (prefix == "osd rm-nodown") {
9535 } else if (prefix == "osd rm-noin") {
9544 vector<string> idvec;
9545 cmd_getval(g_ceph_context, cmdmap, "ids", idvec);
9547 for (unsigned j = 0; j < idvec.size() && !stop; j++) {
9553 (idvec[0] == "any" || idvec[0] == "all" || idvec[0] == "*")) {
9555 // touch previous noup/nodown/noin/noout osds only
9558 osdmap.get_noup_osds(&osds);
9561 osdmap.get_nodown_osds(&osds);
9564 osdmap.get_noin_osds(&osds);
9567 osdmap.get_noout_osds(&osds);
9570 assert(0 == "invalid option");
9573 // cancel any pending noup/nodown/noin/noout requests too
9574 vector<int> pending_state_osds;
9575 (void) pending_inc.get_pending_state_osds(&pending_state_osds);
9576 for (auto &p : pending_state_osds) {
9580 if (!osdmap.is_noup(p) &&
9581 pending_inc.pending_osd_state_clear(p, CEPH_OSD_NOUP)) {
9587 if (!osdmap.is_nodown(p) &&
9588 pending_inc.pending_osd_state_clear(p, CEPH_OSD_NODOWN)) {
9594 if (!osdmap.is_noin(p) &&
9595 pending_inc.pending_osd_state_clear(p, CEPH_OSD_NOIN)) {
9601 if (!osdmap.is_noout(p) &&
9602 pending_inc.pending_osd_state_clear(p, CEPH_OSD_NOOUT)) {
9608 assert(0 == "invalid option");
9614 // try traditional single osd way
9616 long osd = parse_osd_id(idvec[j].c_str(), &ss);
9618 // ss has reason for failure
9619 ss << ", unable to parse osd id:\"" << idvec[j] << "\". ";
9624 osds.push_back(osd);
9627 for (auto &osd : osds) {
9629 if (!osdmap.exists(osd)) {
9630 ss << "osd." << osd << " does not exist. ";
9636 if (osdmap.is_noup(osd)) {
9637 pending_inc.pending_osd_state_set(osd, CEPH_OSD_NOUP);
9639 } else if (pending_inc.pending_osd_state_clear(
9640 osd, CEPH_OSD_NOUP)) {
9646 if (osdmap.is_nodown(osd)) {
9647 pending_inc.pending_osd_state_set(osd, CEPH_OSD_NODOWN);
9649 } else if (pending_inc.pending_osd_state_clear(
9650 osd, CEPH_OSD_NODOWN)) {
9656 if (osdmap.is_noin(osd)) {
9657 pending_inc.pending_osd_state_set(osd, CEPH_OSD_NOIN);
9659 } else if (pending_inc.pending_osd_state_clear(
9660 osd, CEPH_OSD_NOIN)) {
9666 if (osdmap.is_noout(osd)) {
9667 pending_inc.pending_osd_state_set(osd, CEPH_OSD_NOOUT);
9669 } else if (pending_inc.pending_osd_state_clear(
9670 osd, CEPH_OSD_NOOUT)) {
9676 assert(0 == "invalid option");
9683 wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, err, rs,
9684 get_last_committed() + 1));
9687 } else if (prefix == "osd pg-temp") {
9689 if (!cmd_getval(g_ceph_context, cmdmap, "pgid", pgidstr)) {
9690 ss << "unable to parse 'pgid' value '"
9691 << cmd_vartype_stringify(cmdmap["pgid"]) << "'";
9696 if (!pgid.parse(pgidstr.c_str())) {
9697 ss << "invalid pgid '" << pgidstr << "'";
9701 if (!osdmap.pg_exists(pgid)) {
9702 ss << "pg " << pgid << " does not exist";
9706 if (pending_inc.new_pg_temp.count(pgid)) {
9707 dout(10) << __func__ << " waiting for pending update on " << pgid << dendl;
9708 wait_for_finished_proposal(op, new C_RetryMessage(this, op));
9712 vector<int64_t> id_vec;
9713 vector<int32_t> new_pg_temp;
9714 if (!cmd_getval(g_ceph_context, cmdmap, "id", id_vec)) {
9715 ss << "unable to parse 'id' value(s) '"
9716 << cmd_vartype_stringify(cmdmap["id"]) << "'";
9720 for (auto osd : id_vec) {
9721 if (!osdmap.exists(osd)) {
9722 ss << "osd." << osd << " does not exist";
9726 new_pg_temp.push_back(osd);
9729 int pool_min_size = osdmap.get_pg_pool_min_size(pgid);
9730 if ((int)new_pg_temp.size() < pool_min_size) {
9731 ss << "num of osds (" << new_pg_temp.size() <<") < pool min size ("
9732 << pool_min_size << ")";
9737 int pool_size = osdmap.get_pg_pool_size(pgid);
9738 if ((int)new_pg_temp.size() > pool_size) {
9739 ss << "num of osds (" << new_pg_temp.size() <<") > pool size ("
9740 << pool_size << ")";
9745 pending_inc.new_pg_temp[pgid] = mempool::osdmap::vector<int>(
9746 new_pg_temp.begin(), new_pg_temp.end());
9747 ss << "set " << pgid << " pg_temp mapping to " << new_pg_temp;
9749 } else if (prefix == "osd primary-temp") {
9751 if (!cmd_getval(g_ceph_context, cmdmap, "pgid", pgidstr)) {
9752 ss << "unable to parse 'pgid' value '"
9753 << cmd_vartype_stringify(cmdmap["pgid"]) << "'";
9758 if (!pgid.parse(pgidstr.c_str())) {
9759 ss << "invalid pgid '" << pgidstr << "'";
9763 if (!osdmap.pg_exists(pgid)) {
9764 ss << "pg " << pgid << " does not exist";
9770 if (!cmd_getval(g_ceph_context, cmdmap, "id", osd)) {
9771 ss << "unable to parse 'id' value '"
9772 << cmd_vartype_stringify(cmdmap["id"]) << "'";
9776 if (osd != -1 && !osdmap.exists(osd)) {
9777 ss << "osd." << osd << " does not exist";
9782 if (osdmap.require_min_compat_client > 0 &&
9783 osdmap.require_min_compat_client < CEPH_RELEASE_FIREFLY) {
9784 ss << "require_min_compat_client "
9785 << ceph_release_name(osdmap.require_min_compat_client)
9786 << " < firefly, which is required for primary-temp";
9789 } else if (!g_conf->mon_osd_allow_primary_temp) {
9790 ss << "you must enable 'mon osd allow primary temp = true' on the mons before you can set primary_temp mappings. note that this is for developers only: older clients/OSDs will break and there is no feature bit infrastructure in place.";
9795 pending_inc.new_primary_temp[pgid] = osd;
9796 ss << "set " << pgid << " primary_temp mapping to " << osd;
9798 } else if (prefix == "osd pg-upmap" ||
9799 prefix == "osd rm-pg-upmap" ||
9800 prefix == "osd pg-upmap-items" ||
9801 prefix == "osd rm-pg-upmap-items") {
9802 if (osdmap.require_osd_release < CEPH_RELEASE_LUMINOUS) {
9803 ss << "you must complete the upgrade and 'ceph osd require-osd-release "
9804 << "luminous' before using the new interface";
9808 if (osdmap.require_min_compat_client < CEPH_RELEASE_LUMINOUS) {
9809 ss << "min_compat_client "
9810 << ceph_release_name(osdmap.require_min_compat_client)
9811 << " < luminous, which is required for pg-upmap. "
9812 << "Try 'ceph osd set-require-min-compat-client luminous' "
9813 << "before using the new interface";
9817 err = check_cluster_features(CEPH_FEATUREMASK_OSDMAP_PG_UPMAP, ss);
9823 if (!cmd_getval(g_ceph_context, cmdmap, "pgid", pgidstr)) {
9824 ss << "unable to parse 'pgid' value '"
9825 << cmd_vartype_stringify(cmdmap["pgid"]) << "'";
9830 if (!pgid.parse(pgidstr.c_str())) {
9831 ss << "invalid pgid '" << pgidstr << "'";
9835 if (!osdmap.pg_exists(pgid)) {
9836 ss << "pg " << pgid << " does not exist";
9845 OP_RM_PG_UPMAP_ITEMS,
9848 if (prefix == "osd pg-upmap") {
9849 option = OP_PG_UPMAP;
9850 } else if (prefix == "osd rm-pg-upmap") {
9851 option = OP_RM_PG_UPMAP;
9852 } else if (prefix == "osd pg-upmap-items") {
9853 option = OP_PG_UPMAP_ITEMS;
9855 option = OP_RM_PG_UPMAP_ITEMS;
9858 // check pending upmap changes
9860 case OP_PG_UPMAP: // fall through
9861 case OP_RM_PG_UPMAP:
9862 if (pending_inc.new_pg_upmap.count(pgid) ||
9863 pending_inc.old_pg_upmap.count(pgid)) {
9864 dout(10) << __func__ << " waiting for pending update on "
9866 wait_for_finished_proposal(op, new C_RetryMessage(this, op));
9871 case OP_PG_UPMAP_ITEMS: // fall through
9872 case OP_RM_PG_UPMAP_ITEMS:
9873 if (pending_inc.new_pg_upmap_items.count(pgid) ||
9874 pending_inc.old_pg_upmap_items.count(pgid)) {
9875 dout(10) << __func__ << " waiting for pending update on "
9877 wait_for_finished_proposal(op, new C_RetryMessage(this, op));
9883 assert(0 == "invalid option");
9889 vector<int64_t> id_vec;
9890 if (!cmd_getval(g_ceph_context, cmdmap, "id", id_vec)) {
9891 ss << "unable to parse 'id' value(s) '"
9892 << cmd_vartype_stringify(cmdmap["id"]) << "'";
9897 int pool_min_size = osdmap.get_pg_pool_min_size(pgid);
9898 if ((int)id_vec.size() < pool_min_size) {
9899 ss << "num of osds (" << id_vec.size() <<") < pool min size ("
9900 << pool_min_size << ")";
9905 int pool_size = osdmap.get_pg_pool_size(pgid);
9906 if ((int)id_vec.size() > pool_size) {
9907 ss << "num of osds (" << id_vec.size() <<") > pool size ("
9908 << pool_size << ")";
9913 vector<int32_t> new_pg_upmap;
9914 for (auto osd : id_vec) {
9915 if (osd != CRUSH_ITEM_NONE && !osdmap.exists(osd)) {
9916 ss << "osd." << osd << " does not exist";
9920 auto it = std::find(new_pg_upmap.begin(), new_pg_upmap.end(), osd);
9921 if (it != new_pg_upmap.end()) {
9922 ss << "osd." << osd << " already exists, ";
9925 new_pg_upmap.push_back(osd);
9928 if (new_pg_upmap.empty()) {
9929 ss << "no valid upmap items(pairs) is specified";
9934 pending_inc.new_pg_upmap[pgid] = mempool::osdmap::vector<int32_t>(
9935 new_pg_upmap.begin(), new_pg_upmap.end());
9936 ss << "set " << pgid << " pg_upmap mapping to " << new_pg_upmap;
9940 case OP_RM_PG_UPMAP:
9942 pending_inc.old_pg_upmap.insert(pgid);
9943 ss << "clear " << pgid << " pg_upmap mapping";
9947 case OP_PG_UPMAP_ITEMS:
9949 vector<int64_t> id_vec;
9950 if (!cmd_getval(g_ceph_context, cmdmap, "id", id_vec)) {
9951 ss << "unable to parse 'id' value(s) '"
9952 << cmd_vartype_stringify(cmdmap["id"]) << "'";
9957 if (id_vec.size() % 2) {
9958 ss << "you must specify pairs of osd ids to be remapped";
9963 int pool_size = osdmap.get_pg_pool_size(pgid);
9964 if ((int)(id_vec.size() / 2) > pool_size) {
9965 ss << "num of osd pairs (" << id_vec.size() / 2 <<") > pool size ("
9966 << pool_size << ")";
9971 vector<pair<int32_t,int32_t>> new_pg_upmap_items;
9972 ostringstream items;
9974 for (auto p = id_vec.begin(); p != id_vec.end(); ++p) {
9978 ss << "from osd." << from << " == to osd." << to << ", ";
9981 if (!osdmap.exists(from)) {
9982 ss << "osd." << from << " does not exist";
9986 if (to != CRUSH_ITEM_NONE && !osdmap.exists(to)) {
9987 ss << "osd." << to << " does not exist";
9991 pair<int32_t,int32_t> entry = make_pair(from, to);
9992 auto it = std::find(new_pg_upmap_items.begin(),
9993 new_pg_upmap_items.end(), entry);
9994 if (it != new_pg_upmap_items.end()) {
9995 ss << "osd." << from << " -> osd." << to << " already exists, ";
9998 new_pg_upmap_items.push_back(entry);
9999 items << from << "->" << to << ",";
10001 string out(items.str());
10002 out.resize(out.size() - 1); // drop last ','
10005 if (new_pg_upmap_items.empty()) {
10006 ss << "no valid upmap items(pairs) is specified";
10011 pending_inc.new_pg_upmap_items[pgid] =
10012 mempool::osdmap::vector<pair<int32_t,int32_t>>(
10013 new_pg_upmap_items.begin(), new_pg_upmap_items.end());
10014 ss << "set " << pgid << " pg_upmap_items mapping to " << out;
10018 case OP_RM_PG_UPMAP_ITEMS:
10020 pending_inc.old_pg_upmap_items.insert(pgid);
10021 ss << "clear " << pgid << " pg_upmap_items mapping";
10026 assert(0 == "invalid option");
10030 } else if (prefix == "osd primary-affinity") {
10032 if (!cmd_getval(g_ceph_context, cmdmap, "id", id)) {
10033 ss << "invalid osd id value '"
10034 << cmd_vartype_stringify(cmdmap["id"]) << "'";
10039 if (!cmd_getval(g_ceph_context, cmdmap, "weight", w)) {
10040 ss << "unable to parse 'weight' value '"
10041 << cmd_vartype_stringify(cmdmap["weight"]) << "'";
10045 long ww = (int)((double)CEPH_OSD_MAX_PRIMARY_AFFINITY*w);
10047 ss << "weight must be >= 0";
10051 if (osdmap.require_min_compat_client > 0 &&
10052 osdmap.require_min_compat_client < CEPH_RELEASE_FIREFLY) {
10053 ss << "require_min_compat_client "
10054 << ceph_release_name(osdmap.require_min_compat_client)
10055 << " < firefly, which is required for primary-affinity";
10058 } else if (!g_conf->mon_osd_allow_primary_affinity) {
10059 ss << "you must enable 'mon osd allow primary affinity = true' on the mons before you can adjust primary-affinity. note that older clients will no longer be able to communicate with the cluster.";
10063 err = check_cluster_features(CEPH_FEATURE_OSD_PRIMARY_AFFINITY, ss);
10064 if (err == -EAGAIN)
10068 if (osdmap.exists(id)) {
10069 pending_inc.new_primary_affinity[id] = ww;
10070 ss << "set osd." << id << " primary-affinity to " << w << " (" << ios::hex << ww << ios::dec << ")";
10072 wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs,
10073 get_last_committed() + 1));
10076 ss << "osd." << id << " does not exist";
10080 } else if (prefix == "osd reweight") {
10082 if (!cmd_getval(g_ceph_context, cmdmap, "id", id)) {
10083 ss << "unable to parse osd id value '"
10084 << cmd_vartype_stringify(cmdmap["id"]) << "'";
10089 if (!cmd_getval(g_ceph_context, cmdmap, "weight", w)) {
10090 ss << "unable to parse weight value '"
10091 << cmd_vartype_stringify(cmdmap["weight"]) << "'";
10095 long ww = (int)((double)CEPH_OSD_IN*w);
10097 ss << "weight must be >= 0";
10101 if (osdmap.exists(id)) {
10102 pending_inc.new_weight[id] = ww;
10103 ss << "reweighted osd." << id << " to " << w << " (" << std::hex << ww << std::dec << ")";
10105 wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs,
10106 get_last_committed() + 1));
10109 ss << "osd." << id << " does not exist";
10113 } else if (prefix == "osd reweightn") {
10114 map<int32_t, uint32_t> weights;
10115 err = parse_reweights(g_ceph_context, cmdmap, osdmap, &weights);
10117 ss << "unable to parse 'weights' value '"
10118 << cmd_vartype_stringify(cmdmap["weights"]) << "'";
10121 pending_inc.new_weight.insert(weights.begin(), weights.end());
10122 wait_for_finished_proposal(
10124 new Monitor::C_Command(mon, op, 0, rs, rdata, get_last_committed() + 1));
10126 } else if (prefix == "osd lost") {
10128 if (!cmd_getval(g_ceph_context, cmdmap, "id", id)) {
10129 ss << "unable to parse osd id value '"
10130 << cmd_vartype_stringify(cmdmap["id"]) << "'";
10135 if (!cmd_getval(g_ceph_context, cmdmap, "sure", sure) || sure != "--yes-i-really-mean-it") {
10136 ss << "are you SURE? this might mean real, permanent data loss. pass "
10137 "--yes-i-really-mean-it if you really do.";
10140 } else if (!osdmap.exists(id)) {
10141 ss << "osd." << id << " does not exist";
10144 } else if (!osdmap.is_down(id)) {
10145 ss << "osd." << id << " is not down";
10149 epoch_t e = osdmap.get_info(id).down_at;
10150 pending_inc.new_lost[id] = e;
10151 ss << "marked osd lost in epoch " << e;
10153 wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs,
10154 get_last_committed() + 1));
10158 } else if (prefix == "osd destroy" || prefix == "osd purge") {
10159 /* Destroying an OSD means that we don't expect to further make use of
10160 * the OSDs data (which may even become unreadable after this operation),
10161 * and that we are okay with scrubbing all its cephx keys and config-key
10162 * data (which may include lockbox keys, thus rendering the osd's data
10165 * The OSD will not be removed. Instead, we will mark it as destroyed,
10166 * such that a subsequent call to `create` will not reuse the osd id.
10167 * This will play into being able to recreate the OSD, at the same
10168 * crush location, with minimal data movement.
10171 // make sure authmon is writeable.
10172 if (!mon->authmon()->is_writeable()) {
10173 dout(10) << __func__ << " waiting for auth mon to be writeable for "
10174 << "osd destroy" << dendl;
10175 mon->authmon()->wait_for_writeable(op, new C_RetryMessage(this, op));
10180 if (!cmd_getval(g_ceph_context, cmdmap, "id", id)) {
10181 ss << "unable to parse osd id value '"
10182 << cmd_vartype_stringify(cmdmap["id"]) << "";
10187 bool is_destroy = (prefix == "osd destroy");
10189 assert("osd purge" == prefix);
10193 if (!cmd_getval(g_ceph_context, cmdmap, "sure", sure) ||
10194 sure != "--yes-i-really-mean-it") {
10195 ss << "Are you SURE? This will mean real, permanent data loss, as well "
10196 << "as cephx and lockbox keys. Pass --yes-i-really-mean-it if you "
10200 } else if (!osdmap.exists(id)) {
10201 ss << "osd." << id << " does not exist";
10202 err = 0; // idempotent
10204 } else if (osdmap.is_up(id)) {
10205 ss << "osd." << id << " is not `down`.";
10208 } else if (is_destroy && osdmap.is_destroyed(id)) {
10209 ss << "destroyed osd." << id;
10214 bool goto_reply = false;
10218 err = prepare_command_osd_destroy(id, ss);
10219 // we checked above that it should exist.
10220 assert(err != -ENOENT);
10222 err = prepare_command_osd_purge(id, ss);
10223 if (err == -ENOENT) {
10225 ss << "osd." << id << " does not exist.";
10231 if (err < 0 || goto_reply) {
10236 ss << "destroyed osd." << id;
10238 ss << "purged osd." << id;
10242 wait_for_finished_proposal(op,
10243 new Monitor::C_Command(mon, op, 0, rs, get_last_committed() + 1));
10244 force_immediate_propose();
10247 } else if (prefix == "osd new") {
10249 // make sure authmon is writeable.
10250 if (!mon->authmon()->is_writeable()) {
10251 dout(10) << __func__ << " waiting for auth mon to be writeable for "
10252 << "osd new" << dendl;
10253 mon->authmon()->wait_for_writeable(op, new C_RetryMessage(this, op));
10257 map<string,string> secrets_map;
10259 bufferlist bl = m->get_data();
10260 string secrets_json = bl.to_str();
10261 dout(20) << __func__ << " osd new json = " << secrets_json << dendl;
10263 err = get_json_str_map(secrets_json, ss, &secrets_map);
10267 dout(20) << __func__ << " osd new secrets " << secrets_map << dendl;
10270 err = prepare_command_osd_new(op, cmdmap, secrets_map, ss, f.get());
10283 if (err == EEXIST) {
10284 // idempotent operation
10289 wait_for_finished_proposal(op,
10290 new Monitor::C_Command(mon, op, 0, rs, rdata,
10291 get_last_committed() + 1));
10292 force_immediate_propose();
10295 } else if (prefix == "osd create") {
10297 // optional id provided?
10298 int64_t id = -1, cmd_id = -1;
10299 if (cmd_getval(g_ceph_context, cmdmap, "id", cmd_id)) {
10301 ss << "invalid osd id value '" << cmd_id << "'";
10305 dout(10) << " osd create got id " << cmd_id << dendl;
10310 if (cmd_getval(g_ceph_context, cmdmap, "uuid", uuidstr)) {
10311 if (!uuid.parse(uuidstr.c_str())) {
10312 ss << "invalid uuid value '" << uuidstr << "'";
10316 // we only care about the id if we also have the uuid, to
10317 // ensure the operation's idempotency.
10321 int32_t new_id = -1;
10322 err = prepare_command_osd_create(id, uuid, &new_id, ss);
10324 if (err == -EAGAIN) {
10325 wait_for_finished_proposal(op, new C_RetryMessage(this, op));
10328 // a check has failed; reply to the user.
10331 } else if (err == EEXIST) {
10332 // this is an idempotent operation; we can go ahead and reply.
10334 f->open_object_section("created_osd");
10335 f->dump_int("osdid", new_id);
10336 f->close_section();
10346 do_osd_create(id, uuid, &new_id);
10349 f->open_object_section("created_osd");
10350 f->dump_int("osdid", new_id);
10351 f->close_section();
10357 wait_for_finished_proposal(op,
10358 new Monitor::C_Command(mon, op, 0, rs, rdata,
10359 get_last_committed() + 1));
10362 } else if (prefix == "osd blacklist clear") {
10363 pending_inc.new_blacklist.clear();
10364 std::list<std::pair<entity_addr_t,utime_t > > blacklist;
10365 osdmap.get_blacklist(&blacklist);
10366 for (const auto &entry : blacklist) {
10367 pending_inc.old_blacklist.push_back(entry.first);
10369 ss << " removed all blacklist entries";
10371 wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs,
10372 get_last_committed() + 1));
10374 } else if (prefix == "osd blacklist") {
10376 cmd_getval(g_ceph_context, cmdmap, "addr", addrstr);
10377 entity_addr_t addr;
10378 if (!addr.parse(addrstr.c_str(), 0)) {
10379 ss << "unable to parse address " << addrstr;
10384 string blacklistop;
10385 cmd_getval(g_ceph_context, cmdmap, "blacklistop", blacklistop);
10386 if (blacklistop == "add") {
10387 utime_t expires = ceph_clock_now();
10389 // default one hour
10390 cmd_getval(g_ceph_context, cmdmap, "expire", d,
10391 g_conf->mon_osd_blacklist_default_expire);
10394 pending_inc.new_blacklist[addr] = expires;
10397 // cancel any pending un-blacklisting request too
10398 auto it = std::find(pending_inc.old_blacklist.begin(),
10399 pending_inc.old_blacklist.end(), addr);
10400 if (it != pending_inc.old_blacklist.end()) {
10401 pending_inc.old_blacklist.erase(it);
10405 ss << "blacklisting " << addr << " until " << expires << " (" << d << " sec)";
10407 wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs,
10408 get_last_committed() + 1));
10410 } else if (blacklistop == "rm") {
10411 if (osdmap.is_blacklisted(addr) ||
10412 pending_inc.new_blacklist.count(addr)) {
10413 if (osdmap.is_blacklisted(addr))
10414 pending_inc.old_blacklist.push_back(addr);
10416 pending_inc.new_blacklist.erase(addr);
10417 ss << "un-blacklisting " << addr;
10419 wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs,
10420 get_last_committed() + 1));
10423 ss << addr << " isn't blacklisted";
10428 } else if (prefix == "osd pool mksnap") {
10430 cmd_getval(g_ceph_context, cmdmap, "pool", poolstr);
10431 int64_t pool = osdmap.lookup_pg_pool_name(poolstr.c_str());
10433 ss << "unrecognized pool '" << poolstr << "'";
10438 cmd_getval(g_ceph_context, cmdmap, "snap", snapname);
10439 const pg_pool_t *p = osdmap.get_pg_pool(pool);
10440 if (p->is_unmanaged_snaps_mode()) {
10441 ss << "pool " << poolstr << " is in unmanaged snaps mode";
10444 } else if (p->snap_exists(snapname.c_str())) {
10445 ss << "pool " << poolstr << " snap " << snapname << " already exists";
10448 } else if (p->is_tier()) {
10449 ss << "pool " << poolstr << " is a cache tier";
10454 if (pending_inc.new_pools.count(pool))
10455 pp = &pending_inc.new_pools[pool];
10457 pp = &pending_inc.new_pools[pool];
10460 if (pp->snap_exists(snapname.c_str())) {
10461 ss << "pool " << poolstr << " snap " << snapname << " already exists";
10463 pp->add_snap(snapname.c_str(), ceph_clock_now());
10464 pp->set_snap_epoch(pending_inc.epoch);
10465 ss << "created pool " << poolstr << " snap " << snapname;
10468 wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs,
10469 get_last_committed() + 1));
10471 } else if (prefix == "osd pool rmsnap") {
10473 cmd_getval(g_ceph_context, cmdmap, "pool", poolstr);
10474 int64_t pool = osdmap.lookup_pg_pool_name(poolstr.c_str());
10476 ss << "unrecognized pool '" << poolstr << "'";
10481 cmd_getval(g_ceph_context, cmdmap, "snap", snapname);
10482 const pg_pool_t *p = osdmap.get_pg_pool(pool);
10483 if (p->is_unmanaged_snaps_mode()) {
10484 ss << "pool " << poolstr << " is in unmanaged snaps mode";
10487 } else if (!p->snap_exists(snapname.c_str())) {
10488 ss << "pool " << poolstr << " snap " << snapname << " does not exist";
10493 if (pending_inc.new_pools.count(pool))
10494 pp = &pending_inc.new_pools[pool];
10496 pp = &pending_inc.new_pools[pool];
10499 snapid_t sn = pp->snap_exists(snapname.c_str());
10501 pp->remove_snap(sn);
10502 pp->set_snap_epoch(pending_inc.epoch);
10503 ss << "removed pool " << poolstr << " snap " << snapname;
10505 ss << "already removed pool " << poolstr << " snap " << snapname;
10508 wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs,
10509 get_last_committed() + 1));
10511 } else if (prefix == "osd pool create") {
10514 cmd_getval(g_ceph_context, cmdmap, "pg_num", pg_num, int64_t(0));
10515 cmd_getval(g_ceph_context, cmdmap, "pgp_num", pgp_num, pg_num);
10517 string pool_type_str;
10518 cmd_getval(g_ceph_context, cmdmap, "pool_type", pool_type_str);
10519 if (pool_type_str.empty())
10520 pool_type_str = g_conf->osd_pool_default_type;
10523 cmd_getval(g_ceph_context, cmdmap, "pool", poolstr);
10524 int64_t pool_id = osdmap.lookup_pg_pool_name(poolstr);
10525 if (pool_id >= 0) {
10526 const pg_pool_t *p = osdmap.get_pg_pool(pool_id);
10527 if (pool_type_str != p->get_type_name()) {
10528 ss << "pool '" << poolstr << "' cannot change to type " << pool_type_str;
10531 ss << "pool '" << poolstr << "' already exists";
10538 if (pool_type_str == "replicated") {
10539 pool_type = pg_pool_t::TYPE_REPLICATED;
10540 } else if (pool_type_str == "erasure") {
10541 err = check_cluster_features(CEPH_FEATURE_CRUSH_V2 |
10542 CEPH_FEATURE_OSD_ERASURE_CODES,
10544 if (err == -EAGAIN)
10548 pool_type = pg_pool_t::TYPE_ERASURE;
10550 ss << "unknown pool type '" << pool_type_str << "'";
10555 bool implicit_rule_creation = false;
10557 cmd_getval(g_ceph_context, cmdmap, "rule", rule_name);
10558 string erasure_code_profile;
10559 cmd_getval(g_ceph_context, cmdmap, "erasure_code_profile", erasure_code_profile);
10561 if (pool_type == pg_pool_t::TYPE_ERASURE) {
10562 if (erasure_code_profile == "")
10563 erasure_code_profile = "default";
10564 //handle the erasure code profile
10565 if (erasure_code_profile == "default") {
10566 if (!osdmap.has_erasure_code_profile(erasure_code_profile)) {
10567 if (pending_inc.has_erasure_code_profile(erasure_code_profile)) {
10568 dout(20) << "erasure code profile " << erasure_code_profile << " already pending" << dendl;
10572 map<string,string> profile_map;
10573 err = osdmap.get_erasure_code_profile_default(g_ceph_context,
10578 dout(20) << "erasure code profile " << erasure_code_profile << " set" << dendl;
10579 pending_inc.set_erasure_code_profile(erasure_code_profile, profile_map);
10583 if (rule_name == "") {
10584 implicit_rule_creation = true;
10585 if (erasure_code_profile == "default") {
10586 rule_name = "erasure-code";
10588 dout(1) << "implicitly use rule named after the pool: "
10589 << poolstr << dendl;
10590 rule_name = poolstr;
10594 //NOTE:for replicated pool,cmd_map will put rule_name to erasure_code_profile field
10595 rule_name = erasure_code_profile;
10598 if (!implicit_rule_creation && rule_name != "") {
10600 err = get_crush_rule(rule_name, &rule, &ss);
10601 if (err == -EAGAIN) {
10602 wait_for_finished_proposal(op, new C_RetryMessage(this, op));
10609 int64_t expected_num_objects;
10610 cmd_getval(g_ceph_context, cmdmap, "expected_num_objects", expected_num_objects, int64_t(0));
10611 if (expected_num_objects < 0) {
10612 ss << "'expected_num_objects' must be non-negative";
10617 int64_t fast_read_param;
10618 cmd_getval(g_ceph_context, cmdmap, "fast_read", fast_read_param, int64_t(-1));
10619 FastReadType fast_read = FAST_READ_DEFAULT;
10620 if (fast_read_param == 0)
10621 fast_read = FAST_READ_OFF;
10622 else if (fast_read_param > 0)
10623 fast_read = FAST_READ_ON;
10625 err = prepare_new_pool(poolstr, 0, // auid=0 for admin created pool
10626 -1, // default crush rule
10629 erasure_code_profile, pool_type,
10630 (uint64_t)expected_num_objects,
10636 ss << "pool '" << poolstr << "' already exists";
10639 wait_for_finished_proposal(op, new C_RetryMessage(this, op));
10648 ss << "pool '" << poolstr << "' created";
10651 wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs,
10652 get_last_committed() + 1));
10655 } else if (prefix == "osd pool delete" ||
10656 prefix == "osd pool rm") {
10657 // osd pool delete/rm <poolname> <poolname again> --yes-i-really-really-mean-it
10658 string poolstr, poolstr2, sure;
10659 cmd_getval(g_ceph_context, cmdmap, "pool", poolstr);
10660 cmd_getval(g_ceph_context, cmdmap, "pool2", poolstr2);
10661 cmd_getval(g_ceph_context, cmdmap, "sure", sure);
10662 int64_t pool = osdmap.lookup_pg_pool_name(poolstr.c_str());
10664 ss << "pool '" << poolstr << "' does not exist";
10669 bool force_no_fake = sure == "--yes-i-really-really-mean-it-not-faking";
10670 if (poolstr2 != poolstr ||
10671 (sure != "--yes-i-really-really-mean-it" && !force_no_fake)) {
10672 ss << "WARNING: this will *PERMANENTLY DESTROY* all data stored in pool " << poolstr
10673 << ". If you are *ABSOLUTELY CERTAIN* that is what you want, pass the pool name *twice*, "
10674 << "followed by --yes-i-really-really-mean-it.";
10678 err = _prepare_remove_pool(pool, &ss, force_no_fake);
10679 if (err == -EAGAIN) {
10680 wait_for_finished_proposal(op, new C_RetryMessage(this, op));
10686 } else if (prefix == "osd pool rename") {
10687 string srcpoolstr, destpoolstr;
10688 cmd_getval(g_ceph_context, cmdmap, "srcpool", srcpoolstr);
10689 cmd_getval(g_ceph_context, cmdmap, "destpool", destpoolstr);
10690 int64_t pool_src = osdmap.lookup_pg_pool_name(srcpoolstr.c_str());
10691 int64_t pool_dst = osdmap.lookup_pg_pool_name(destpoolstr.c_str());
10693 if (pool_src < 0) {
10694 if (pool_dst >= 0) {
10695 // src pool doesn't exist, dst pool does exist: to ensure idempotency
10696 // of operations, assume this rename succeeded, as it is not changing
10697 // the current state. Make sure we output something understandable
10698 // for whoever is issuing the command, if they are paying attention,
10699 // in case it was not intentional; or to avoid a "wtf?" and a bug
10700 // report in case it was intentional, while expecting a failure.
10701 ss << "pool '" << srcpoolstr << "' does not exist; pool '"
10702 << destpoolstr << "' does -- assuming successful rename";
10705 ss << "unrecognized pool '" << srcpoolstr << "'";
10709 } else if (pool_dst >= 0) {
10710 // source pool exists and so does the destination pool
10711 ss << "pool '" << destpoolstr << "' already exists";
10716 int ret = _prepare_rename_pool(pool_src, destpoolstr);
10718 ss << "pool '" << srcpoolstr << "' renamed to '" << destpoolstr << "'";
10720 ss << "failed to rename pool '" << srcpoolstr << "' to '" << destpoolstr << "': "
10721 << cpp_strerror(ret);
10724 wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, ret, rs,
10725 get_last_committed() + 1));
10728 } else if (prefix == "osd pool set") {
10729 err = prepare_command_pool_set(cmdmap, ss);
10730 if (err == -EAGAIN)
10736 wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs,
10737 get_last_committed() + 1));
10739 } else if (prefix == "osd tier add") {
10740 err = check_cluster_features(CEPH_FEATURE_OSD_CACHEPOOL, ss);
10741 if (err == -EAGAIN)
10746 cmd_getval(g_ceph_context, cmdmap, "pool", poolstr);
10747 int64_t pool_id = osdmap.lookup_pg_pool_name(poolstr);
10749 ss << "unrecognized pool '" << poolstr << "'";
10753 string tierpoolstr;
10754 cmd_getval(g_ceph_context, cmdmap, "tierpool", tierpoolstr);
10755 int64_t tierpool_id = osdmap.lookup_pg_pool_name(tierpoolstr);
10756 if (tierpool_id < 0) {
10757 ss << "unrecognized pool '" << tierpoolstr << "'";
10761 const pg_pool_t *p = osdmap.get_pg_pool(pool_id);
10763 const pg_pool_t *tp = osdmap.get_pg_pool(tierpool_id);
10766 if (!_check_become_tier(tierpool_id, tp, pool_id, p, &err, &ss)) {
10770 // make sure new tier is empty
10771 string force_nonempty;
10772 cmd_getval(g_ceph_context, cmdmap, "force_nonempty", force_nonempty);
10773 const pool_stat_t *pstats = mon->pgservice->get_pool_stat(tierpool_id);
10774 if (pstats && pstats->stats.sum.num_objects != 0 &&
10775 force_nonempty != "--force-nonempty") {
10776 ss << "tier pool '" << tierpoolstr << "' is not empty; --force-nonempty to force";
10780 if (tp->ec_pool()) {
10781 ss << "tier pool '" << tierpoolstr
10782 << "' is an ec pool, which cannot be a tier";
10786 if ((!tp->removed_snaps.empty() || !tp->snaps.empty()) &&
10787 ((force_nonempty != "--force-nonempty") ||
10788 (!g_conf->mon_debug_unsafe_allow_tier_with_nonempty_snaps))) {
10789 ss << "tier pool '" << tierpoolstr << "' has snapshot state; it cannot be added as a tier without breaking the pool";
10794 pg_pool_t *np = pending_inc.get_new_pool(pool_id, p);
10795 pg_pool_t *ntp = pending_inc.get_new_pool(tierpool_id, tp);
10796 if (np->tiers.count(tierpool_id) || ntp->is_tier()) {
10797 wait_for_finished_proposal(op, new C_RetryMessage(this, op));
10800 np->tiers.insert(tierpool_id);
10801 np->set_snap_epoch(pending_inc.epoch); // tier will update to our snap info
10802 ntp->tier_of = pool_id;
10803 ss << "pool '" << tierpoolstr << "' is now (or already was) a tier of '" << poolstr << "'";
10804 wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, ss.str(),
10805 get_last_committed() + 1));
10807 } else if (prefix == "osd tier remove" ||
10808 prefix == "osd tier rm") {
10810 cmd_getval(g_ceph_context, cmdmap, "pool", poolstr);
10811 int64_t pool_id = osdmap.lookup_pg_pool_name(poolstr);
10813 ss << "unrecognized pool '" << poolstr << "'";
10817 string tierpoolstr;
10818 cmd_getval(g_ceph_context, cmdmap, "tierpool", tierpoolstr);
10819 int64_t tierpool_id = osdmap.lookup_pg_pool_name(tierpoolstr);
10820 if (tierpool_id < 0) {
10821 ss << "unrecognized pool '" << tierpoolstr << "'";
10825 const pg_pool_t *p = osdmap.get_pg_pool(pool_id);
10827 const pg_pool_t *tp = osdmap.get_pg_pool(tierpool_id);
10830 if (!_check_remove_tier(pool_id, p, tp, &err, &ss)) {
10834 if (p->tiers.count(tierpool_id) == 0) {
10835 ss << "pool '" << tierpoolstr << "' is now (or already was) not a tier of '" << poolstr << "'";
10839 if (tp->tier_of != pool_id) {
10840 ss << "tier pool '" << tierpoolstr << "' is a tier of '"
10841 << osdmap.get_pool_name(tp->tier_of) << "': "
10842 // be scary about it; this is an inconsistency and bells must go off
10843 << "THIS SHOULD NOT HAVE HAPPENED AT ALL";
10847 if (p->read_tier == tierpool_id) {
10848 ss << "tier pool '" << tierpoolstr << "' is the overlay for '" << poolstr << "'; please remove-overlay first";
10853 pg_pool_t *np = pending_inc.get_new_pool(pool_id, p);
10854 pg_pool_t *ntp = pending_inc.get_new_pool(tierpool_id, tp);
10855 if (np->tiers.count(tierpool_id) == 0 ||
10856 ntp->tier_of != pool_id ||
10857 np->read_tier == tierpool_id) {
10858 wait_for_finished_proposal(op, new C_RetryMessage(this, op));
10861 np->tiers.erase(tierpool_id);
10863 ss << "pool '" << tierpoolstr << "' is now (or already was) not a tier of '" << poolstr << "'";
10864 wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, ss.str(),
10865 get_last_committed() + 1));
10867 } else if (prefix == "osd tier set-overlay") {
10868 err = check_cluster_features(CEPH_FEATURE_OSD_CACHEPOOL, ss);
10869 if (err == -EAGAIN)
10874 cmd_getval(g_ceph_context, cmdmap, "pool", poolstr);
10875 int64_t pool_id = osdmap.lookup_pg_pool_name(poolstr);
10877 ss << "unrecognized pool '" << poolstr << "'";
10881 string overlaypoolstr;
10882 cmd_getval(g_ceph_context, cmdmap, "overlaypool", overlaypoolstr);
10883 int64_t overlaypool_id = osdmap.lookup_pg_pool_name(overlaypoolstr);
10884 if (overlaypool_id < 0) {
10885 ss << "unrecognized pool '" << overlaypoolstr << "'";
10889 const pg_pool_t *p = osdmap.get_pg_pool(pool_id);
10891 const pg_pool_t *overlay_p = osdmap.get_pg_pool(overlaypool_id);
10893 if (p->tiers.count(overlaypool_id) == 0) {
10894 ss << "tier pool '" << overlaypoolstr << "' is not a tier of '" << poolstr << "'";
10898 if (p->read_tier == overlaypool_id) {
10900 ss << "overlay for '" << poolstr << "' is now (or already was) '" << overlaypoolstr << "'";
10903 if (p->has_read_tier()) {
10904 ss << "pool '" << poolstr << "' has overlay '"
10905 << osdmap.get_pool_name(p->read_tier)
10906 << "'; please remove-overlay first";
10912 pg_pool_t *np = pending_inc.get_new_pool(pool_id, p);
10913 np->read_tier = overlaypool_id;
10914 np->write_tier = overlaypool_id;
10915 np->set_last_force_op_resend(pending_inc.epoch);
10916 pg_pool_t *noverlay_p = pending_inc.get_new_pool(overlaypool_id, overlay_p);
10917 noverlay_p->set_last_force_op_resend(pending_inc.epoch);
10918 ss << "overlay for '" << poolstr << "' is now (or already was) '" << overlaypoolstr << "'";
10919 if (overlay_p->cache_mode == pg_pool_t::CACHEMODE_NONE)
10920 ss <<" (WARNING: overlay pool cache_mode is still NONE)";
10921 wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, ss.str(),
10922 get_last_committed() + 1));
10924 } else if (prefix == "osd tier remove-overlay" ||
10925 prefix == "osd tier rm-overlay") {
10927 cmd_getval(g_ceph_context, cmdmap, "pool", poolstr);
10928 int64_t pool_id = osdmap.lookup_pg_pool_name(poolstr);
10930 ss << "unrecognized pool '" << poolstr << "'";
10934 const pg_pool_t *p = osdmap.get_pg_pool(pool_id);
10936 if (!p->has_read_tier()) {
10938 ss << "there is now (or already was) no overlay for '" << poolstr << "'";
10942 if (!_check_remove_tier(pool_id, p, NULL, &err, &ss)) {
10947 pg_pool_t *np = pending_inc.get_new_pool(pool_id, p);
10948 if (np->has_read_tier()) {
10949 const pg_pool_t *op = osdmap.get_pg_pool(np->read_tier);
10950 pg_pool_t *nop = pending_inc.get_new_pool(np->read_tier,op);
10951 nop->set_last_force_op_resend(pending_inc.epoch);
10953 if (np->has_write_tier()) {
10954 const pg_pool_t *op = osdmap.get_pg_pool(np->write_tier);
10955 pg_pool_t *nop = pending_inc.get_new_pool(np->write_tier, op);
10956 nop->set_last_force_op_resend(pending_inc.epoch);
10958 np->clear_read_tier();
10959 np->clear_write_tier();
10960 np->set_last_force_op_resend(pending_inc.epoch);
10961 ss << "there is now (or already was) no overlay for '" << poolstr << "'";
10962 wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, ss.str(),
10963 get_last_committed() + 1));
10965 } else if (prefix == "osd tier cache-mode") {
10966 err = check_cluster_features(CEPH_FEATURE_OSD_CACHEPOOL, ss);
10967 if (err == -EAGAIN)
10972 cmd_getval(g_ceph_context, cmdmap, "pool", poolstr);
10973 int64_t pool_id = osdmap.lookup_pg_pool_name(poolstr);
10975 ss << "unrecognized pool '" << poolstr << "'";
10979 const pg_pool_t *p = osdmap.get_pg_pool(pool_id);
10981 if (!p->is_tier()) {
10982 ss << "pool '" << poolstr << "' is not a tier";
10987 cmd_getval(g_ceph_context, cmdmap, "mode", modestr);
10988 pg_pool_t::cache_mode_t mode = pg_pool_t::get_cache_mode_from_str(modestr);
10990 ss << "'" << modestr << "' is not a valid cache mode";
10996 cmd_getval(g_ceph_context, cmdmap, "sure", sure);
10997 if ((mode != pg_pool_t::CACHEMODE_WRITEBACK &&
10998 mode != pg_pool_t::CACHEMODE_NONE &&
10999 mode != pg_pool_t::CACHEMODE_PROXY &&
11000 mode != pg_pool_t::CACHEMODE_READPROXY) &&
11001 sure != "--yes-i-really-mean-it") {
11002 ss << "'" << modestr << "' is not a well-supported cache mode and may "
11003 << "corrupt your data. pass --yes-i-really-mean-it to force.";
11008 // pool already has this cache-mode set and there are no pending changes
11009 if (p->cache_mode == mode &&
11010 (pending_inc.new_pools.count(pool_id) == 0 ||
11011 pending_inc.new_pools[pool_id].cache_mode == p->cache_mode)) {
11012 ss << "set cache-mode for pool '" << poolstr << "'"
11013 << " to " << pg_pool_t::get_cache_mode_name(mode);
11018 /* Mode description:
11020 * none: No cache-mode defined
11021 * forward: Forward all reads and writes to base pool
11022 * writeback: Cache writes, promote reads from base pool
11023 * readonly: Forward writes to base pool
11024 * readforward: Writes are in writeback mode, Reads are in forward mode
11025 * proxy: Proxy all reads and writes to base pool
11026 * readproxy: Writes are in writeback mode, Reads are in proxy mode
11028 * Hence, these are the allowed transitions:
11031 * forward -> proxy || readforward || readproxy || writeback || any IF num_objects_dirty == 0
11032 * proxy -> forward || readforward || readproxy || writeback || any IF num_objects_dirty == 0
11033 * readforward -> forward || proxy || readproxy || writeback || any IF num_objects_dirty == 0
11034 * readproxy -> forward || proxy || readforward || writeback || any IF num_objects_dirty == 0
11035 * writeback -> readforward || readproxy || forward || proxy
11039 // We check if the transition is valid against the current pool mode, as
11040 // it is the only committed state thus far. We will blantly squash
11041 // whatever mode is on the pending state.
11043 if (p->cache_mode == pg_pool_t::CACHEMODE_WRITEBACK &&
11044 (mode != pg_pool_t::CACHEMODE_FORWARD &&
11045 mode != pg_pool_t::CACHEMODE_PROXY &&
11046 mode != pg_pool_t::CACHEMODE_READFORWARD &&
11047 mode != pg_pool_t::CACHEMODE_READPROXY)) {
11048 ss << "unable to set cache-mode '" << pg_pool_t::get_cache_mode_name(mode)
11049 << "' on a '" << pg_pool_t::get_cache_mode_name(p->cache_mode)
11050 << "' pool; only '"
11051 << pg_pool_t::get_cache_mode_name(pg_pool_t::CACHEMODE_FORWARD)
11053 << pg_pool_t::get_cache_mode_name(pg_pool_t::CACHEMODE_PROXY)
11055 << pg_pool_t::get_cache_mode_name(pg_pool_t::CACHEMODE_READFORWARD)
11057 << pg_pool_t::get_cache_mode_name(pg_pool_t::CACHEMODE_READPROXY)
11062 if ((p->cache_mode == pg_pool_t::CACHEMODE_READFORWARD &&
11063 (mode != pg_pool_t::CACHEMODE_WRITEBACK &&
11064 mode != pg_pool_t::CACHEMODE_FORWARD &&
11065 mode != pg_pool_t::CACHEMODE_PROXY &&
11066 mode != pg_pool_t::CACHEMODE_READPROXY)) ||
11068 (p->cache_mode == pg_pool_t::CACHEMODE_READPROXY &&
11069 (mode != pg_pool_t::CACHEMODE_WRITEBACK &&
11070 mode != pg_pool_t::CACHEMODE_FORWARD &&
11071 mode != pg_pool_t::CACHEMODE_READFORWARD &&
11072 mode != pg_pool_t::CACHEMODE_PROXY)) ||
11074 (p->cache_mode == pg_pool_t::CACHEMODE_PROXY &&
11075 (mode != pg_pool_t::CACHEMODE_WRITEBACK &&
11076 mode != pg_pool_t::CACHEMODE_FORWARD &&
11077 mode != pg_pool_t::CACHEMODE_READFORWARD &&
11078 mode != pg_pool_t::CACHEMODE_READPROXY)) ||
11080 (p->cache_mode == pg_pool_t::CACHEMODE_FORWARD &&
11081 (mode != pg_pool_t::CACHEMODE_WRITEBACK &&
11082 mode != pg_pool_t::CACHEMODE_READFORWARD &&
11083 mode != pg_pool_t::CACHEMODE_PROXY &&
11084 mode != pg_pool_t::CACHEMODE_READPROXY))) {
11086 const pool_stat_t* pstats =
11087 mon->pgservice->get_pool_stat(pool_id);
11089 if (pstats && pstats->stats.sum.num_objects_dirty > 0) {
11090 ss << "unable to set cache-mode '"
11091 << pg_pool_t::get_cache_mode_name(mode) << "' on pool '" << poolstr
11092 << "': dirty objects found";
11098 pg_pool_t *np = pending_inc.get_new_pool(pool_id, p);
11099 np->cache_mode = mode;
11100 // set this both when moving to and from cache_mode NONE. this is to
11101 // capture legacy pools that were set up before this flag existed.
11102 np->flags |= pg_pool_t::FLAG_INCOMPLETE_CLONES;
11103 ss << "set cache-mode for pool '" << poolstr
11104 << "' to " << pg_pool_t::get_cache_mode_name(mode);
11105 if (mode == pg_pool_t::CACHEMODE_NONE) {
11106 const pg_pool_t *base_pool = osdmap.get_pg_pool(np->tier_of);
11108 if (base_pool->read_tier == pool_id ||
11109 base_pool->write_tier == pool_id)
11110 ss <<" (WARNING: pool is still configured as read or write tier)";
11112 wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, ss.str(),
11113 get_last_committed() + 1));
11115 } else if (prefix == "osd tier add-cache") {
11116 err = check_cluster_features(CEPH_FEATURE_OSD_CACHEPOOL, ss);
11117 if (err == -EAGAIN)
11122 cmd_getval(g_ceph_context, cmdmap, "pool", poolstr);
11123 int64_t pool_id = osdmap.lookup_pg_pool_name(poolstr);
11125 ss << "unrecognized pool '" << poolstr << "'";
11129 string tierpoolstr;
11130 cmd_getval(g_ceph_context, cmdmap, "tierpool", tierpoolstr);
11131 int64_t tierpool_id = osdmap.lookup_pg_pool_name(tierpoolstr);
11132 if (tierpool_id < 0) {
11133 ss << "unrecognized pool '" << tierpoolstr << "'";
11137 const pg_pool_t *p = osdmap.get_pg_pool(pool_id);
11139 const pg_pool_t *tp = osdmap.get_pg_pool(tierpool_id);
11142 if (!_check_become_tier(tierpool_id, tp, pool_id, p, &err, &ss)) {
11147 if (!cmd_getval(g_ceph_context, cmdmap, "size", size)) {
11148 ss << "unable to parse 'size' value '"
11149 << cmd_vartype_stringify(cmdmap["size"]) << "'";
11153 // make sure new tier is empty
11154 const pool_stat_t *pstats =
11155 mon->pgservice->get_pool_stat(tierpool_id);
11156 if (pstats && pstats->stats.sum.num_objects != 0) {
11157 ss << "tier pool '" << tierpoolstr << "' is not empty";
11161 string modestr = g_conf->osd_tier_default_cache_mode;
11162 pg_pool_t::cache_mode_t mode = pg_pool_t::get_cache_mode_from_str(modestr);
11164 ss << "osd tier cache default mode '" << modestr << "' is not a valid cache mode";
11168 HitSet::Params hsp;
11169 if (g_conf->osd_tier_default_cache_hit_set_type == "bloom") {
11170 BloomHitSet::Params *bsp = new BloomHitSet::Params;
11171 bsp->set_fpp(g_conf->osd_pool_default_hit_set_bloom_fpp);
11172 hsp = HitSet::Params(bsp);
11173 } else if (g_conf->osd_tier_default_cache_hit_set_type == "explicit_hash") {
11174 hsp = HitSet::Params(new ExplicitHashHitSet::Params);
11176 else if (g_conf->osd_tier_default_cache_hit_set_type == "explicit_object") {
11177 hsp = HitSet::Params(new ExplicitObjectHitSet::Params);
11179 ss << "osd tier cache default hit set type '" <<
11180 g_conf->osd_tier_default_cache_hit_set_type << "' is not a known type";
11185 pg_pool_t *np = pending_inc.get_new_pool(pool_id, p);
11186 pg_pool_t *ntp = pending_inc.get_new_pool(tierpool_id, tp);
11187 if (np->tiers.count(tierpool_id) || ntp->is_tier()) {
11188 wait_for_finished_proposal(op, new C_RetryMessage(this, op));
11191 np->tiers.insert(tierpool_id);
11192 np->read_tier = np->write_tier = tierpool_id;
11193 np->set_snap_epoch(pending_inc.epoch); // tier will update to our snap info
11194 np->set_last_force_op_resend(pending_inc.epoch);
11195 ntp->set_last_force_op_resend(pending_inc.epoch);
11196 ntp->tier_of = pool_id;
11197 ntp->cache_mode = mode;
11198 ntp->hit_set_count = g_conf->osd_tier_default_cache_hit_set_count;
11199 ntp->hit_set_period = g_conf->osd_tier_default_cache_hit_set_period;
11200 ntp->min_read_recency_for_promote = g_conf->osd_tier_default_cache_min_read_recency_for_promote;
11201 ntp->min_write_recency_for_promote = g_conf->osd_tier_default_cache_min_write_recency_for_promote;
11202 ntp->hit_set_grade_decay_rate = g_conf->osd_tier_default_cache_hit_set_grade_decay_rate;
11203 ntp->hit_set_search_last_n = g_conf->osd_tier_default_cache_hit_set_search_last_n;
11204 ntp->hit_set_params = hsp;
11205 ntp->target_max_bytes = size;
11206 ss << "pool '" << tierpoolstr << "' is now (or already was) a cache tier of '" << poolstr << "'";
11207 wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, ss.str(),
11208 get_last_committed() + 1));
11210 } else if (prefix == "osd pool set-quota") {
11212 cmd_getval(g_ceph_context, cmdmap, "pool", poolstr);
11213 int64_t pool_id = osdmap.lookup_pg_pool_name(poolstr);
11215 ss << "unrecognized pool '" << poolstr << "'";
11221 cmd_getval(g_ceph_context, cmdmap, "field", field);
11222 if (field != "max_objects" && field != "max_bytes") {
11223 ss << "unrecognized field '" << field << "'; should be 'max_bytes' or 'max_objects'";
11228 // val could contain unit designations, so we treat as a string
11230 cmd_getval(g_ceph_context, cmdmap, "val", val);
11232 int64_t value = unit_to_bytesize(val, &tss);
11234 ss << "error parsing value '" << value << "': " << tss.str();
11239 pg_pool_t *pi = pending_inc.get_new_pool(pool_id, osdmap.get_pg_pool(pool_id));
11240 if (field == "max_objects") {
11241 pi->quota_max_objects = value;
11242 } else if (field == "max_bytes") {
11243 pi->quota_max_bytes = value;
11245 assert(0 == "unrecognized option");
11247 ss << "set-quota " << field << " = " << value << " for pool " << poolstr;
11249 wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs,
11250 get_last_committed() + 1));
11252 } else if (prefix == "osd pool application enable" ||
11253 prefix == "osd pool application disable" ||
11254 prefix == "osd pool application set" ||
11255 prefix == "osd pool application rm") {
11256 err = prepare_command_pool_application(prefix, cmdmap, ss);
11257 if (err == -EAGAIN)
11263 wait_for_finished_proposal(
11264 op, new Monitor::C_Command(mon, op, 0, rs, get_last_committed() + 1));
11266 } else if (prefix == "osd reweight-by-pg" ||
11267 prefix == "osd reweight-by-utilization" ||
11268 prefix == "osd test-reweight-by-pg" ||
11269 prefix == "osd test-reweight-by-utilization") {
11271 prefix == "osd reweight-by-pg" || prefix == "osd test-reweight-by-pg";
11273 prefix == "osd test-reweight-by-pg" ||
11274 prefix == "osd test-reweight-by-utilization";
11276 cmd_getval(g_ceph_context, cmdmap, "oload", oload, int64_t(120));
11277 set<int64_t> pools;
11278 vector<string> poolnamevec;
11279 cmd_getval(g_ceph_context, cmdmap, "pools", poolnamevec);
11280 for (unsigned j = 0; j < poolnamevec.size(); j++) {
11281 int64_t pool = osdmap.lookup_pg_pool_name(poolnamevec[j]);
11283 ss << "pool '" << poolnamevec[j] << "' does not exist";
11287 pools.insert(pool);
11289 double max_change = g_conf->mon_reweight_max_change;
11290 cmd_getval(g_ceph_context, cmdmap, "max_change", max_change);
11291 if (max_change <= 0.0) {
11292 ss << "max_change " << max_change << " must be positive";
11296 int64_t max_osds = g_conf->mon_reweight_max_osds;
11297 cmd_getval(g_ceph_context, cmdmap, "max_osds", max_osds);
11298 if (max_osds <= 0) {
11299 ss << "max_osds " << max_osds << " must be positive";
11303 string no_increasing;
11304 cmd_getval(g_ceph_context, cmdmap, "no_increasing", no_increasing);
11306 mempool::osdmap::map<int32_t, uint32_t> new_weights;
11307 err = mon->pgservice->reweight_by_utilization(osdmap,
11312 pools.empty() ? NULL : &pools,
11313 no_increasing == "--no-increasing",
11315 &ss, &out_str, f.get());
11317 dout(10) << "reweight::by_utilization: finished with " << out_str << dendl;
11322 rdata.append(out_str);
11324 ss << "FAILED reweight-by-pg";
11325 } else if (err == 0 || dry_run) {
11328 ss << "SUCCESSFUL reweight-by-pg";
11329 pending_inc.new_weight = std::move(new_weights);
11330 wait_for_finished_proposal(
11332 new Monitor::C_Command(mon, op, 0, rs, rdata, get_last_committed() + 1));
11335 } else if (prefix == "osd force-create-pg") {
11338 cmd_getval(g_ceph_context, cmdmap, "pgid", pgidstr);
11339 if (!pgid.parse(pgidstr.c_str())) {
11340 ss << "invalid pgid '" << pgidstr << "'";
11346 std::lock_guard<std::mutex> l(creating_pgs_lock);
11347 auto emplaced = creating_pgs.pgs.emplace(pgid,
11348 make_pair(osdmap.get_epoch(),
11349 ceph_clock_now()));
11350 creating_now = emplaced.second;
11352 if (creating_now) {
11353 ss << "pg " << pgidstr << " now creating, ok";
11357 ss << "pg " << pgid << " already creating";
11367 if (err < 0 && rs.length() == 0)
11368 rs = cpp_strerror(err);
11369 mon->reply_command(op, err, rs, rdata, get_last_committed());
11374 wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs,
11375 get_last_committed() + 1));
11379 wait_for_finished_proposal(op, new C_RetryMessage(this, op));
11383 bool OSDMonitor::preprocess_pool_op(MonOpRequestRef op)
11385 op->mark_osdmon_event(__func__);
11386 MPoolOp *m = static_cast<MPoolOp*>(op->get_req());
11388 if (m->fsid != mon->monmap->fsid) {
11389 dout(0) << __func__ << " drop message on fsid " << m->fsid
11390 << " != " << mon->monmap->fsid << " for " << *m << dendl;
11391 _pool_op_reply(op, -EINVAL, osdmap.get_epoch());
11395 if (m->op == POOL_OP_CREATE)
11396 return preprocess_pool_op_create(op);
11398 if (!osdmap.get_pg_pool(m->pool)) {
11399 dout(10) << "attempt to operate on non-existent pool id " << m->pool << dendl;
11400 _pool_op_reply(op, 0, osdmap.get_epoch());
11404 // check if the snap and snapname exist
11405 bool snap_exists = false;
11406 const pg_pool_t *p = osdmap.get_pg_pool(m->pool);
11407 if (p->snap_exists(m->name.c_str()))
11408 snap_exists = true;
11411 case POOL_OP_CREATE_SNAP:
11412 if (p->is_unmanaged_snaps_mode() || p->is_tier()) {
11413 _pool_op_reply(op, -EINVAL, osdmap.get_epoch());
11417 _pool_op_reply(op, 0, osdmap.get_epoch());
11421 case POOL_OP_CREATE_UNMANAGED_SNAP:
11422 if (p->is_pool_snaps_mode()) {
11423 _pool_op_reply(op, -EINVAL, osdmap.get_epoch());
11427 case POOL_OP_DELETE_SNAP:
11428 if (p->is_unmanaged_snaps_mode()) {
11429 _pool_op_reply(op, -EINVAL, osdmap.get_epoch());
11432 if (!snap_exists) {
11433 _pool_op_reply(op, 0, osdmap.get_epoch());
11437 case POOL_OP_DELETE_UNMANAGED_SNAP:
11438 if (p->is_pool_snaps_mode()) {
11439 _pool_op_reply(op, -EINVAL, osdmap.get_epoch());
11442 if (p->is_removed_snap(m->snapid)) {
11443 _pool_op_reply(op, 0, osdmap.get_epoch());
11447 case POOL_OP_DELETE:
11448 if (osdmap.lookup_pg_pool_name(m->name.c_str()) >= 0) {
11449 _pool_op_reply(op, 0, osdmap.get_epoch());
11453 case POOL_OP_AUID_CHANGE:
11463 bool OSDMonitor::preprocess_pool_op_create(MonOpRequestRef op)
11465 op->mark_osdmon_event(__func__);
11466 MPoolOp *m = static_cast<MPoolOp*>(op->get_req());
11467 MonSession *session = m->get_session();
11469 _pool_op_reply(op, -EPERM, osdmap.get_epoch());
11472 if (!session->is_capable("osd", MON_CAP_W)) {
11473 dout(5) << "attempt to create new pool without sufficient auid privileges!"
11474 << "message: " << *m << std::endl
11475 << "caps: " << session->caps << dendl;
11476 _pool_op_reply(op, -EPERM, osdmap.get_epoch());
11480 int64_t pool = osdmap.lookup_pg_pool_name(m->name.c_str());
11482 _pool_op_reply(op, 0, osdmap.get_epoch());
11489 bool OSDMonitor::prepare_pool_op(MonOpRequestRef op)
11491 op->mark_osdmon_event(__func__);
11492 MPoolOp *m = static_cast<MPoolOp*>(op->get_req());
11493 dout(10) << "prepare_pool_op " << *m << dendl;
11494 if (m->op == POOL_OP_CREATE) {
11495 return prepare_pool_op_create(op);
11496 } else if (m->op == POOL_OP_DELETE) {
11497 return prepare_pool_op_delete(op);
11501 bool changed = false;
11503 if (!osdmap.have_pg_pool(m->pool)) {
11504 _pool_op_reply(op, -ENOENT, osdmap.get_epoch());
11508 const pg_pool_t *pool = osdmap.get_pg_pool(m->pool);
11511 case POOL_OP_CREATE_SNAP:
11512 if (pool->is_tier()) {
11514 _pool_op_reply(op, ret, osdmap.get_epoch());
11516 } // else, fall through
11517 case POOL_OP_DELETE_SNAP:
11518 if (!pool->is_unmanaged_snaps_mode()) {
11519 bool snap_exists = pool->snap_exists(m->name.c_str());
11520 if ((m->op == POOL_OP_CREATE_SNAP && snap_exists)
11521 || (m->op == POOL_OP_DELETE_SNAP && !snap_exists)) {
11529 _pool_op_reply(op, ret, osdmap.get_epoch());
11532 case POOL_OP_DELETE_UNMANAGED_SNAP:
11533 // we won't allow removal of an unmanaged snapshot from a pool
11534 // not in unmanaged snaps mode.
11535 if (!pool->is_unmanaged_snaps_mode()) {
11536 _pool_op_reply(op, -ENOTSUP, osdmap.get_epoch());
11540 case POOL_OP_CREATE_UNMANAGED_SNAP:
11541 // but we will allow creating an unmanaged snapshot on any pool
11542 // as long as it is not in 'pool' snaps mode.
11543 if (pool->is_pool_snaps_mode()) {
11544 _pool_op_reply(op, -EINVAL, osdmap.get_epoch());
11549 // projected pool info
11551 if (pending_inc.new_pools.count(m->pool))
11552 pp = pending_inc.new_pools[m->pool];
11554 pp = *osdmap.get_pg_pool(m->pool);
11556 bufferlist reply_data;
11558 // pool snaps vs unmanaged snaps are mutually exclusive
11560 case POOL_OP_CREATE_SNAP:
11561 case POOL_OP_DELETE_SNAP:
11562 if (pp.is_unmanaged_snaps_mode()) {
11568 case POOL_OP_CREATE_UNMANAGED_SNAP:
11569 case POOL_OP_DELETE_UNMANAGED_SNAP:
11570 if (pp.is_pool_snaps_mode()) {
11577 case POOL_OP_CREATE_SNAP:
11578 if (!pp.snap_exists(m->name.c_str())) {
11579 pp.add_snap(m->name.c_str(), ceph_clock_now());
11580 dout(10) << "create snap in pool " << m->pool << " " << m->name << " seq " << pp.get_snap_epoch() << dendl;
11585 case POOL_OP_DELETE_SNAP:
11587 snapid_t s = pp.snap_exists(m->name.c_str());
11595 case POOL_OP_CREATE_UNMANAGED_SNAP:
11598 pp.add_unmanaged_snap(snapid);
11599 ::encode(snapid, reply_data);
11604 case POOL_OP_DELETE_UNMANAGED_SNAP:
11605 if (!pp.is_removed_snap(m->snapid)) {
11606 pp.remove_unmanaged_snap(m->snapid);
11611 case POOL_OP_AUID_CHANGE:
11612 if (pp.auid != m->auid) {
11624 pp.set_snap_epoch(pending_inc.epoch);
11625 pending_inc.new_pools[m->pool] = pp;
11629 wait_for_finished_proposal(op, new OSDMonitor::C_PoolOp(this, op, ret, pending_inc.epoch, &reply_data));
11633 bool OSDMonitor::prepare_pool_op_create(MonOpRequestRef op)
11635 op->mark_osdmon_event(__func__);
11636 int err = prepare_new_pool(op);
11637 wait_for_finished_proposal(op, new OSDMonitor::C_PoolOp(this, op, err, pending_inc.epoch));
11641 int OSDMonitor::_check_remove_pool(int64_t pool_id, const pg_pool_t& pool,
11644 const string& poolstr = osdmap.get_pool_name(pool_id);
11646 // If the Pool is in use by CephFS, refuse to delete it
11647 FSMap const &pending_fsmap = mon->mdsmon()->get_pending();
11648 if (pending_fsmap.pool_in_use(pool_id)) {
11649 *ss << "pool '" << poolstr << "' is in use by CephFS";
11653 if (pool.tier_of >= 0) {
11654 *ss << "pool '" << poolstr << "' is a tier of '"
11655 << osdmap.get_pool_name(pool.tier_of) << "'";
11658 if (!pool.tiers.empty()) {
11659 *ss << "pool '" << poolstr << "' has tiers";
11660 for(auto tier : pool.tiers) {
11661 *ss << " " << osdmap.get_pool_name(tier);
11666 if (!g_conf->mon_allow_pool_delete) {
11667 *ss << "pool deletion is disabled; you must first set the mon_allow_pool_delete config option to true before you can destroy a pool";
11671 if (pool.has_flag(pg_pool_t::FLAG_NODELETE)) {
11672 *ss << "pool deletion is disabled; you must unset nodelete flag for the pool first";
11676 *ss << "pool '" << poolstr << "' removed";
11681 * Check if it is safe to add a tier to a base pool
11684 * True if the operation should proceed, false if we should abort here
11685 * (abort doesn't necessarily mean error, could be idempotency)
11687 bool OSDMonitor::_check_become_tier(
11688 const int64_t tier_pool_id, const pg_pool_t *tier_pool,
11689 const int64_t base_pool_id, const pg_pool_t *base_pool,
11693 const std::string &tier_pool_name = osdmap.get_pool_name(tier_pool_id);
11694 const std::string &base_pool_name = osdmap.get_pool_name(base_pool_id);
11696 const FSMap &pending_fsmap = mon->mdsmon()->get_pending();
11697 if (pending_fsmap.pool_in_use(tier_pool_id)) {
11698 *ss << "pool '" << tier_pool_name << "' is in use by CephFS";
11703 if (base_pool->tiers.count(tier_pool_id)) {
11704 assert(tier_pool->tier_of == base_pool_id);
11706 *ss << "pool '" << tier_pool_name << "' is now (or already was) a tier of '"
11707 << base_pool_name << "'";
11711 if (base_pool->is_tier()) {
11712 *ss << "pool '" << base_pool_name << "' is already a tier of '"
11713 << osdmap.get_pool_name(base_pool->tier_of) << "', "
11714 << "multiple tiers are not yet supported.";
11719 if (tier_pool->has_tiers()) {
11720 *ss << "pool '" << tier_pool_name << "' has following tier(s) already:";
11721 for (set<uint64_t>::iterator it = tier_pool->tiers.begin();
11722 it != tier_pool->tiers.end(); ++it)
11723 *ss << "'" << osdmap.get_pool_name(*it) << "',";
11724 *ss << " multiple tiers are not yet supported.";
11729 if (tier_pool->is_tier()) {
11730 *ss << "tier pool '" << tier_pool_name << "' is already a tier of '"
11731 << osdmap.get_pool_name(tier_pool->tier_of) << "'";
11742 * Check if it is safe to remove a tier from this base pool
11745 * True if the operation should proceed, false if we should abort here
11746 * (abort doesn't necessarily mean error, could be idempotency)
11748 bool OSDMonitor::_check_remove_tier(
11749 const int64_t base_pool_id, const pg_pool_t *base_pool,
11750 const pg_pool_t *tier_pool,
11751 int *err, ostream *ss) const
11753 const std::string &base_pool_name = osdmap.get_pool_name(base_pool_id);
11755 // Apply CephFS-specific checks
11756 const FSMap &pending_fsmap = mon->mdsmon()->get_pending();
11757 if (pending_fsmap.pool_in_use(base_pool_id)) {
11758 if (base_pool->type != pg_pool_t::TYPE_REPLICATED) {
11759 // If the underlying pool is erasure coded, we can't permit the
11760 // removal of the replicated tier that CephFS relies on to access it
11761 *ss << "pool '" << base_pool_name << "' is in use by CephFS via its tier";
11766 if (tier_pool && tier_pool->cache_mode == pg_pool_t::CACHEMODE_WRITEBACK) {
11767 *ss << "pool '" << base_pool_name << "' is in use by CephFS, and this "
11768 "tier is still in use as a writeback cache. Change the cache "
11769 "mode and flush the cache before removing it";
11779 int OSDMonitor::_prepare_remove_pool(
11780 int64_t pool, ostream *ss, bool no_fake)
11782 dout(10) << __func__ << " " << pool << dendl;
11783 const pg_pool_t *p = osdmap.get_pg_pool(pool);
11784 int r = _check_remove_pool(pool, *p, ss);
11788 auto new_pool = pending_inc.new_pools.find(pool);
11789 if (new_pool != pending_inc.new_pools.end()) {
11790 // if there is a problem with the pending info, wait and retry
11792 const auto& p = new_pool->second;
11793 int r = _check_remove_pool(pool, p, ss);
11798 if (pending_inc.old_pools.count(pool)) {
11799 dout(10) << __func__ << " " << pool << " already pending removal"
11804 if (g_conf->mon_fake_pool_delete && !no_fake) {
11805 string old_name = osdmap.get_pool_name(pool);
11806 string new_name = old_name + "." + stringify(pool) + ".DELETED";
11807 dout(1) << __func__ << " faking pool deletion: renaming " << pool << " "
11808 << old_name << " -> " << new_name << dendl;
11809 pending_inc.new_pool_names[pool] = new_name;
11814 pending_inc.old_pools.insert(pool);
11816 // remove any pg_temp mappings for this pool
11817 for (auto p = osdmap.pg_temp->begin();
11818 p != osdmap.pg_temp->end();
11820 if (p->first.pool() == (uint64_t)pool) {
11821 dout(10) << __func__ << " " << pool << " removing obsolete pg_temp "
11822 << p->first << dendl;
11823 pending_inc.new_pg_temp[p->first].clear();
11826 // remove any primary_temp mappings for this pool
11827 for (auto p = osdmap.primary_temp->begin();
11828 p != osdmap.primary_temp->end();
11830 if (p->first.pool() == (uint64_t)pool) {
11831 dout(10) << __func__ << " " << pool
11832 << " removing obsolete primary_temp" << p->first << dendl;
11833 pending_inc.new_primary_temp[p->first] = -1;
11836 // remove any pg_upmap mappings for this pool
11837 for (auto& p : osdmap.pg_upmap) {
11838 if (p.first.pool() == (uint64_t)pool) {
11839 dout(10) << __func__ << " " << pool
11840 << " removing obsolete pg_upmap "
11841 << p.first << dendl;
11842 pending_inc.old_pg_upmap.insert(p.first);
11845 // remove any pg_upmap_items mappings for this pool
11846 for (auto& p : osdmap.pg_upmap_items) {
11847 if (p.first.pool() == (uint64_t)pool) {
11848 dout(10) << __func__ << " " << pool
11849 << " removing obsolete pg_upmap_items " << p.first
11851 pending_inc.old_pg_upmap_items.insert(p.first);
11855 // remove any choose_args for this pool
11856 CrushWrapper newcrush;
11857 _get_pending_crush(newcrush);
11858 if (newcrush.have_choose_args(pool)) {
11859 dout(10) << __func__ << " removing choose_args for pool " << pool << dendl;
11860 newcrush.rm_choose_args(pool);
11861 pending_inc.crush.clear();
11862 newcrush.encode(pending_inc.crush, mon->get_quorum_con_features());
11867 int OSDMonitor::_prepare_rename_pool(int64_t pool, string newname)
11869 dout(10) << "_prepare_rename_pool " << pool << dendl;
11870 if (pending_inc.old_pools.count(pool)) {
11871 dout(10) << "_prepare_rename_pool " << pool << " pending removal" << dendl;
11874 for (map<int64_t,string>::iterator p = pending_inc.new_pool_names.begin();
11875 p != pending_inc.new_pool_names.end();
11877 if (p->second == newname && p->first != pool) {
11882 pending_inc.new_pool_names[pool] = newname;
11886 bool OSDMonitor::prepare_pool_op_delete(MonOpRequestRef op)
11888 op->mark_osdmon_event(__func__);
11889 MPoolOp *m = static_cast<MPoolOp*>(op->get_req());
11891 int ret = _prepare_remove_pool(m->pool, &ss, false);
11892 if (ret == -EAGAIN) {
11893 wait_for_finished_proposal(op, new C_RetryMessage(this, op));
11897 dout(10) << __func__ << " got " << ret << " " << ss.str() << dendl;
11898 wait_for_finished_proposal(op, new OSDMonitor::C_PoolOp(this, op, ret,
11899 pending_inc.epoch));
11903 void OSDMonitor::_pool_op_reply(MonOpRequestRef op,
11904 int ret, epoch_t epoch, bufferlist *blp)
11906 op->mark_osdmon_event(__func__);
11907 MPoolOp *m = static_cast<MPoolOp*>(op->get_req());
11908 dout(20) << "_pool_op_reply " << ret << dendl;
11909 MPoolOpReply *reply = new MPoolOpReply(m->fsid, m->get_tid(),
11910 ret, epoch, get_last_committed(), blp);
11911 mon->send_reply(op, reply);