--- /dev/null
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+/*
+ * Ceph - scalable distributed file system
+ *
+ * Copyright (C) 2012 New Dream Network
+ *
+ * This is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License version 2.1, as published by the Free Software
+ * Foundation. See file COPYING.
+ */
+#include <stdio.h>
+#include <string.h>
+#include <iostream>
+#include <assert.h>
+#include <time.h>
+#include <stdlib.h>
+#include <signal.h>
+#include <cctype>
+#include <errno.h>
+#include <sys/time.h>
+#include "os/ObjectStore.h"
+#include "common/ceph_argparse.h"
+#include "global/global_init.h"
+#include "common/debug.h"
+#include <boost/scoped_ptr.hpp>
+#include <boost/lexical_cast.hpp>
+#include "workload_generator.h"
+#include "include/assert.h"
+
+#include "TestObjectStoreState.h"
+
+#define dout_context g_ceph_context
+
+static const char *our_name = NULL;
+void usage();
+
+boost::scoped_ptr<WorkloadGenerator> wrkldgen;
+
+#define dout_subsys ceph_subsys_
+
+
+WorkloadGenerator::WorkloadGenerator(vector<const char*> args)
+ : TestObjectStoreState(NULL),
+ m_max_in_flight(def_max_in_flight),
+ m_num_ops(-1),
+ m_destroy_coll_every_nr_runs(def_destroy_coll_every_nr_runs),
+ m_num_colls(def_num_colls),
+ m_write_data_bytes(0), m_write_xattr_obj_bytes(0),
+ m_write_xattr_coll_bytes(0), m_write_pglog_bytes(0),
+ m_suppress_write_data(false), m_suppress_write_xattr_obj(false),
+ m_suppress_write_xattr_coll(false), m_suppress_write_log(false),
+ m_do_stats(false),
+ m_stats_finished_txs(0),
+ m_stats_lock("WorldloadGenerator::m_stats_lock"),
+ m_stats_show_secs(5),
+ m_stats_total_written(0),
+ m_stats_begin()
+{
+ int err = 0;
+
+ m_nr_runs = 0;
+
+ init_args(args);
+ dout(0) << "data = " << g_conf->osd_data << dendl;
+ dout(0) << "journal = " << g_conf->osd_journal << dendl;
+ dout(0) << "journal size = " << g_conf->osd_journal_size << dendl;
+
+ err = ::mkdir(g_conf->osd_data.c_str(), 0755);
+ ceph_assert(err == 0 || (err < 0 && errno == EEXIST));
+ ObjectStore *store_ptr = ObjectStore::create(g_ceph_context,
+ g_conf->osd_objectstore,
+ g_conf->osd_data,
+ g_conf->osd_journal);
+ m_store.reset(store_ptr);
+ err = m_store->mkfs();
+ ceph_assert(err == 0);
+ err = m_store->mount();
+ ceph_assert(err == 0);
+
+ set_max_in_flight(m_max_in_flight);
+ set_num_objs_per_coll(def_num_obj_per_coll);
+
+ init(m_num_colls, 0);
+
+ dout(0) << "#colls = " << m_num_colls << dendl;
+ dout(0) << "#objs per coll = " << m_num_objs_per_coll << dendl;
+ dout(0) << "#txs per destr = " << m_destroy_coll_every_nr_runs << dendl;
+
+}
+
+size_t WorkloadGenerator::_parse_size_or_die(std::string& val)
+{
+ size_t s = 0;
+ int multiplier = 0;
+ size_t i = 0;
+
+ if (val.empty()) // this should never happen, but catch it anyway.
+ goto die;
+
+
+ for (i = 0; i < val.length(); i++) {
+ if (!isdigit(val[i])) {
+ if (isalpha(val[i])) {
+ val[i] = tolower(val[i]);
+ switch (val[i]) {
+ case 'b': break;
+ case 'k': multiplier = 10; break;
+ case 'm': multiplier = 20; break;
+ case 'g': multiplier = 30; break;
+ default:
+ goto die;
+ }
+ val[i] = '\0';
+ break;
+ } else {
+ goto die;
+ }
+ }
+ }
+
+ s = strtoll(val.c_str(), NULL, 10) * (1 << multiplier);
+ return s;
+
+die:
+ usage();
+ exit(1);
+}
+
+void WorkloadGenerator::_suppress_ops_or_die(std::string& val)
+{
+ for (size_t i = 0; i < val.length(); i++) {
+ switch (val[i]) {
+ case 'c': m_suppress_write_xattr_coll = true; break;
+ case 'o': m_suppress_write_xattr_obj = true; break;
+ case 'l': m_suppress_write_log = true; break;
+ case 'd': m_suppress_write_data = true; break;
+ default:
+ usage();
+ exit(1);
+ }
+ }
+}
+
+void WorkloadGenerator::init_args(vector<const char*> args)
+{
+ for (std::vector<const char*>::iterator i = args.begin(); i != args.end();) {
+ string val;
+
+ if (ceph_argparse_double_dash(args, i)) {
+ break;
+ } else if (ceph_argparse_witharg(args, i, &val,
+ "--test-num-colls", (char*) NULL)) {
+ m_num_colls = strtoll(val.c_str(), NULL, 10);
+ } else if (ceph_argparse_witharg(args, i, &val,
+ "--test-objs-per-coll", (char*) NULL)) {
+ m_num_objs_per_coll = strtoll(val.c_str(), NULL, 10);
+ } else if (ceph_argparse_witharg(args, i, &val,
+ "--test-destroy-coll-per-N-trans", (char*) NULL)) {
+ m_destroy_coll_every_nr_runs = strtoll(val.c_str(), NULL, 10);
+ } else if (ceph_argparse_witharg(args, i, &val,
+ "--test-num-ops", (char*) NULL)) {
+ m_num_ops = strtoll(val.c_str(), NULL, 10);
+ } else if (ceph_argparse_witharg(args, i, &val,
+ "--test-max-in-flight", (char*) NULL)) {
+ m_max_in_flight = strtoll(val.c_str(), NULL, 10);
+ } else if (ceph_argparse_witharg(args, i, &val,
+ "--test-write-data-size", (char*) NULL)) {
+ m_write_data_bytes = _parse_size_or_die(val);
+ } else if (ceph_argparse_witharg(args, i, &val,
+ "--test-write-xattr-obj-size", (char*) NULL)) {
+ m_write_xattr_obj_bytes = _parse_size_or_die(val);
+ } else if (ceph_argparse_witharg(args, i, &val,
+ "--test-write-xattr-coll-size", (char*) NULL)) {
+ m_write_xattr_coll_bytes = _parse_size_or_die(val);
+ } else if (ceph_argparse_witharg(args, i, &val,
+ "--test-write-pglog-size", (char*) NULL)) {
+ m_write_pglog_bytes = _parse_size_or_die(val);
+ } else if (ceph_argparse_witharg(args, i, &val,
+ "--test-suppress-ops", (char*) NULL)) {
+ _suppress_ops_or_die(val);
+ } else if (ceph_argparse_witharg(args, i, &val,
+ "--test-show-stats-period", (char*) NULL)) {
+ m_stats_show_secs = strtoll(val.c_str(), NULL, 10);
+ } else if (ceph_argparse_flag(args, i, "--test-show-stats", (char*) NULL)) {
+ m_do_stats = true;
+ } else if (ceph_argparse_flag(args, i, "--help", (char*) NULL)) {
+ usage();
+ exit(0);
+ }
+ }
+}
+
+int WorkloadGenerator::get_uniform_random_value(int min, int max)
+{
+ boost::uniform_int<> value(min, max);
+ return value(m_rng);
+}
+
+TestObjectStoreState::coll_entry_t *WorkloadGenerator::get_rnd_coll_entry(bool erase = false)
+{
+ int index = get_uniform_random_value(0, m_collections_ids.size()-1);
+ coll_entry_t *entry = get_coll_at(index, erase);
+ return entry;
+}
+
+hobject_t *WorkloadGenerator::get_rnd_obj(coll_entry_t *entry)
+{
+ assert(entry != NULL);
+
+ bool create =
+ (get_uniform_random_value(0,100) < 50 || !entry->m_objects.size());
+
+ if (create && ((int) entry->m_objects.size() < m_num_objs_per_coll)) {
+ return (entry->touch_obj(entry->m_next_object_id++));
+ }
+
+ int idx = get_uniform_random_value(0, entry->m_objects.size()-1);
+ return entry->get_obj_at(idx);
+}
+
+/**
+ * We'll generate a random amount of bytes, ranging from a single byte up to
+ * a couple of MB.
+ */
+size_t WorkloadGenerator::get_random_byte_amount(size_t min, size_t max)
+{
+ size_t diff = max - min;
+ return (size_t) (min + (rand() % diff));
+}
+
+void WorkloadGenerator::get_filled_byte_array(bufferlist& bl, size_t size)
+{
+ static const char alphanum[] = "0123456789"
+ "ABCDEFGHIJKLMNOPQRSTUVWXYZ"
+ "abcdefghijklmnopqrstuvwxyz";
+
+ bufferptr bp(size);
+ if (false) {
+ for (unsigned int i = 0; i < size - 1; i++) {
+ bp[i] = alphanum[rand() % sizeof(alphanum)];
+ }
+ bp[size - 1] = '\0';
+ } else {
+ bp.zero();
+ }
+ bl.append(bp);
+}
+
+void WorkloadGenerator::do_write_object(ObjectStore::Transaction *t,
+ coll_t coll, hobject_t obj,
+ C_StatState *stat)
+{
+ if (m_suppress_write_data) {
+ dout(5) << __func__ << " suppressed" << dendl;
+ return;
+ }
+
+ size_t size = m_write_data_bytes;
+ if (!size)
+ size = get_random_byte_amount(min_write_bytes, max_write_bytes);
+
+ bufferlist bl;
+ get_filled_byte_array(bl, size);
+
+ dout(2) << __func__ << " " << coll << "/" << obj
+ << " size " << bl.length() << dendl;
+
+ if (m_do_stats && (stat != NULL))
+ stat->written_data += bl.length();
+
+ t->write(coll, ghobject_t(obj), 0, bl.length(), bl);
+}
+
+void WorkloadGenerator::do_setattr_object(ObjectStore::Transaction *t,
+ coll_t coll, hobject_t obj,
+ C_StatState *stat)
+{
+ if (m_suppress_write_xattr_obj) {
+ dout(5) << __func__ << " suppressed" << dendl;
+ return;
+ }
+
+ size_t size = m_write_xattr_obj_bytes;
+ if (!size)
+ size = get_random_byte_amount(min_xattr_obj_bytes, max_xattr_obj_bytes);
+
+ bufferlist bl;
+ get_filled_byte_array(bl, size);
+
+ dout(2) << __func__ << " " << coll << "/" << obj << " size " << size << dendl;
+
+ if (m_do_stats && (stat != NULL))
+ stat->written_data += bl.length();
+
+ t->setattr(coll, ghobject_t(obj), "objxattr", bl);
+}
+
+void WorkloadGenerator::do_pgmeta_omap_set(ObjectStore::Transaction *t, spg_t pgid,
+ coll_t coll, C_StatState *stat)
+{
+ if (m_suppress_write_xattr_coll) {
+ dout(5) << __func__ << " suppressed" << dendl;
+ return;
+ }
+
+ size_t size = m_write_xattr_coll_bytes;
+ if (!size)
+ size = get_random_byte_amount(min_xattr_coll_bytes, max_xattr_coll_bytes);
+
+ bufferlist bl;
+ get_filled_byte_array(bl, size);
+ dout(2) << __func__ << " coll " << coll << " size " << size << dendl;
+
+ if (m_do_stats && (stat != NULL))
+ stat->written_data += bl.length();
+
+ ghobject_t pgmeta(pgid.make_pgmeta_oid());
+ map<string,bufferlist> values;
+ values["_"].claim(bl);
+ t->omap_setkeys(coll, pgmeta, values);
+}
+
+
+void WorkloadGenerator::do_append_log(ObjectStore::Transaction *t,
+ coll_entry_t *entry, C_StatState *stat)
+{
+ if (m_suppress_write_log) {
+ dout(5) << __func__ << " suppressed" << dendl;
+ return;
+ }
+
+ size_t size = (m_write_pglog_bytes ? m_write_pglog_bytes : log_append_bytes);
+
+ bufferlist bl;
+ get_filled_byte_array(bl, size);
+ ghobject_t log_obj = entry->m_meta_obj;
+
+ dout(2) << __func__ << " coll " << entry->m_coll << " "
+ << coll_t::meta() << " /" << log_obj << " (" << bl.length() << ")" << dendl;
+
+ if (m_do_stats && (stat != NULL))
+ stat->written_data += bl.length();
+
+ uint64_t s = pg_log_size[entry->m_coll];
+ t->write(coll_t::meta(), log_obj, s, bl.length(), bl);
+ pg_log_size[entry->m_coll] += bl.length();
+}
+
+void WorkloadGenerator::do_destroy_collection(ObjectStore::Transaction *t,
+ coll_entry_t *entry,
+ C_StatState *stat)
+{
+ m_nr_runs = 0;
+ entry->m_osr.flush();
+ vector<ghobject_t> ls;
+ m_store->collection_list(entry->m_coll, ghobject_t(), ghobject_t::get_max(),
+ INT_MAX, &ls, NULL);
+ dout(2) << __func__ << " coll " << entry->m_coll
+ << " (" << ls.size() << " objects)" << dendl;
+
+ for (vector<ghobject_t>::iterator it = ls.begin(); it < ls.end(); ++it) {
+ t->remove(entry->m_coll, *it);
+ }
+
+ t->remove_collection(entry->m_coll);
+ t->remove(coll_t::meta(), entry->m_meta_obj);
+}
+
+TestObjectStoreState::coll_entry_t
+*WorkloadGenerator::do_create_collection(ObjectStore::Transaction *t,
+ C_StatState *stat)
+{
+ coll_entry_t *entry = coll_create(m_next_coll_nr++);
+ if (!entry) {
+ dout(0) << __func__ << " failed to create coll id "
+ << m_next_coll_nr << dendl;
+ return NULL;
+ }
+ m_collections.insert(make_pair(entry->m_id, entry));
+
+ dout(2) << __func__ << " id " << entry->m_id << " coll " << entry->m_coll << dendl;
+ t->create_collection(entry->m_coll, 32);
+ dout(2) << __func__ << " meta " << coll_t::meta() << "/" << entry->m_meta_obj << dendl;
+ t->touch(coll_t::meta(), entry->m_meta_obj);
+ return entry;
+}
+
+void WorkloadGenerator::do_stats()
+{
+ utime_t now = ceph_clock_now();
+ m_stats_lock.Lock();
+
+ utime_t duration = (now - m_stats_begin);
+
+ // when cast to double, a utime_t behaves properly
+ double throughput = (m_stats_total_written / ((double) duration));
+ double tx_throughput (m_stats_finished_txs / ((double) duration));
+
+ dout(0) << __func__
+ << " written: " << m_stats_total_written
+ << " duration: " << duration << " sec"
+ << " bandwidth: " << prettybyte_t(throughput) << "/s"
+ << " iops: " << tx_throughput << "/s"
+ << dendl;
+
+ m_stats_lock.Unlock();
+}
+
+void WorkloadGenerator::run()
+{
+ bool create_coll = false;
+ int ops_run = 0;
+
+ utime_t stats_interval(m_stats_show_secs, 0);
+ utime_t now = ceph_clock_now();
+ utime_t stats_time = now;
+ m_stats_begin = now;
+
+ do {
+ C_StatState *stat_state = NULL;
+
+ if (m_num_ops && (ops_run == m_num_ops))
+ break;
+
+ if (!create_coll && !m_collections.size()) {
+ dout(0) << "We ran out of collections!" << dendl;
+ break;
+ }
+
+ dout(5) << __func__
+ << " m_finished_lock is-locked: " << m_finished_lock.is_locked()
+ << " in-flight: " << m_in_flight.load()
+ << dendl;
+
+ wait_for_ready();
+
+ ObjectStore::Transaction *t = new ObjectStore::Transaction;
+ Context *c;
+ bool destroy_collection = false;
+ TestObjectStoreState::coll_entry_t *entry = NULL;
+
+
+ if (m_do_stats) {
+ utime_t now = ceph_clock_now();
+ utime_t elapsed = now - stats_time;
+ if (elapsed >= stats_interval) {
+ do_stats();
+ stats_time = now;
+ }
+ stat_state = new C_StatState(this, now);
+ }
+
+ if (create_coll) {
+ create_coll = false;
+
+ entry = do_create_collection(t, stat_state);
+ if (!entry) {
+ dout(0) << __func__ << " something went terribly wrong creating coll" << dendl;
+ break;
+ }
+
+ c = new C_OnReadable(this);
+ goto queue_tx;
+ }
+
+ destroy_collection = should_destroy_collection();
+ entry = get_rnd_coll_entry(destroy_collection);
+ assert(entry != NULL);
+
+ if (destroy_collection) {
+ do_destroy_collection(t, entry, stat_state);
+ c = new C_OnDestroyed(this, entry);
+ if (!m_num_ops)
+ create_coll = true;
+ } else {
+ hobject_t *obj = get_rnd_obj(entry);
+
+ do_write_object(t, entry->m_coll, *obj, stat_state);
+ do_setattr_object(t, entry->m_coll, *obj, stat_state);
+ do_pgmeta_omap_set(t, entry->m_pgid, entry->m_coll, stat_state);
+ do_append_log(t, entry, stat_state);
+
+ c = new C_OnReadable(this);
+ }
+
+queue_tx:
+
+ if (m_do_stats) {
+ Context *tmp = c;
+ c = new C_StatWrapper(stat_state, tmp);
+ }
+
+ m_store->queue_transaction(&(entry->m_osr), std::move(*t), c);
+ delete t;
+
+ inc_in_flight();
+
+ ops_run ++;
+
+ } while (true);
+
+ dout(2) << __func__ << " waiting for "
+ << m_in_flight.load() << " in-flight transactions" << dendl;
+
+ wait_for_done();
+
+ do_stats();
+
+ dout(0) << __func__ << " finishing" << dendl;
+}
+
+void usage()
+{
+ cout << "usage: " << our_name << "[options]" << std::endl;
+
+ cout << "\
+\n\
+Global Options:\n\
+ -c FILE Read configuration from FILE\n\
+ --osd-objectstore TYPE Set OSD ObjectStore type\n\
+ --osd-data PATH Set OSD Data path\n\
+ --osd-journal PATH Set OSD Journal path\n\
+ --osd-journal-size VAL Set Journal size\n\
+ --help This message\n\
+\n\
+Test-specific Options:\n\
+ --test-num-colls VAL Set the number of collections\n\
+ --test-num-objs-per-coll VAL Set the number of objects per collection\n\
+ --test-destroy-coll-per-N-trans VAL Set how many transactions to run before\n\
+ destroying a collection.\n\
+ --test-num-ops VAL Run a certain number of operations\n\
+ (a VAL of 0 runs the test forever)\n\
+ --test-max-in-flight VAL Maximum number of in-flight transactions\n\
+ (default: 50)\n\
+ --test-suppress-ops OPS Suppress ops specified in OPS\n\
+ --test-write-data-size SIZE Specify SIZE for all data writes\n\
+ --test-write-xattr-obj-size SIZE Specify SIZE for all xattrs on objects\n\
+ --test-write-xattr-coll-size SIZE Specify SIZE for all xattrs on colls\n\
+ --test-write-pglog-size SIZE Specify SIZE for all pglog writes\n\
+ --test-show-stats Show stats as we go\n\
+ --test-show-stats-period SECS Show stats every SECS (default: 5)\n\
+\n\
+ SIZE is a numeric value that can be assumed as being bytes, or may be any\n\
+ other unit if specified: B or b, K or k, M or m, G or g.\n\
+ e.g., 1G = 1024M = 1048576k = 1073741824\n\
+\n\
+ OPS can be one or more of the following options:\n\
+ c writes on collection's xattrs\n\
+ o writes on object's xattr\n\
+ l writes on pglog\n\
+ d data writes on objects\n\
+\n\
+" << std::endl;
+}
+
+int main(int argc, const char *argv[])
+{
+ vector<const char*> def_args;
+ vector<const char*> args;
+
+ our_name = argv[0];
+
+ def_args.push_back("--osd-journal-size");
+ def_args.push_back("400");
+// def_args.push_back("--osd-data");
+// def_args.push_back("workload_gen_dir");
+// def_args.push_back("--osd-journal");
+// def_args.push_back("workload_gen_dir/journal");
+ argv_to_vec(argc, argv, args);
+
+ auto cct = global_init(&def_args, args,
+ CEPH_ENTITY_TYPE_CLIENT, CODE_ENVIRONMENT_UTILITY,
+ CINIT_FLAG_NO_DEFAULT_CONFIG_FILE);
+ common_init_finish(g_ceph_context);
+ g_ceph_context->_conf->apply_changes(NULL);
+
+ WorkloadGenerator *wrkldgen_ptr = new WorkloadGenerator(args);
+ wrkldgen.reset(wrkldgen_ptr);
+ wrkldgen->run();
+ return 0;
+}
Code Review / stor4nfv.git / blobdiff