X-Git-Url: https://gerrit.opnfv.org/gerrit/gitweb?a=blobdiff_plain;f=src%2Fceph%2Fqa%2Fworkunits%2Fcephtool%2Ftest.sh;fp=src%2Fceph%2Fqa%2Fworkunits%2Fcephtool%2Ftest.sh;h=15344172a966cd972e851308566361c1b33bd00d;hb=812ff6ca9fcd3e629e49d4328905f33eee8ca3f5;hp=0000000000000000000000000000000000000000;hpb=15280273faafb77777eab341909a3f495cf248d9;p=stor4nfv.git diff --git a/src/ceph/qa/workunits/cephtool/test.sh b/src/ceph/qa/workunits/cephtool/test.sh new file mode 100755 index 0000000..1534417 --- /dev/null +++ b/src/ceph/qa/workunits/cephtool/test.sh @@ -0,0 +1,2621 @@ +#!/bin/bash -x +# -*- mode:shell-script; tab-width:8; sh-basic-offset:2; indent-tabs-mode:t -*- +# vim: ts=8 sw=8 ft=bash smarttab + +source $(dirname $0)/../../standalone/ceph-helpers.sh + +set -e +set -o functrace +PS4='${BASH_SOURCE[0]}:$LINENO: ${FUNCNAME[0]}: ' +SUDO=${SUDO:-sudo} +export CEPH_DEV=1 + +function get_admin_socket() +{ + local client=$1 + + if test -n "$CEPH_ASOK_DIR"; + then + echo $(get_asok_dir)/$client.asok + else + local cluster=$(echo $CEPH_ARGS | sed -r 's/.*--cluster[[:blank:]]*([[:alnum:]]*).*/\1/') + echo "/var/run/ceph/$cluster-$client.asok" + fi +} + +function check_no_osd_down() +{ + ! ceph osd dump | grep ' down ' +} + +function wait_no_osd_down() +{ + max_run=300 + for i in $(seq 1 $max_run) ; do + if ! check_no_osd_down ; then + echo "waiting for osd(s) to come back up ($i/$max_run)" + sleep 1 + else + break + fi + done + check_no_osd_down +} + +function expect_false() +{ + set -x + if "$@"; then return 1; else return 0; fi +} + + +TEMP_DIR=$(mktemp -d ${TMPDIR-/tmp}/cephtool.XXX) +trap "rm -fr $TEMP_DIR" 0 + +TMPFILE=$(mktemp $TEMP_DIR/test_invalid.XXX) + +# +# retry_eagain max cmd args ... +# +# retry cmd args ... if it exits on error and its output contains the +# string EAGAIN, at most $max times +# +function retry_eagain() +{ + local max=$1 + shift + local status + local tmpfile=$TEMP_DIR/retry_eagain.$$ + local count + for count in $(seq 1 $max) ; do + status=0 + "$@" > $tmpfile 2>&1 || status=$? + if test $status = 0 || + ! grep --quiet EAGAIN $tmpfile ; then + break + fi + sleep 1 + done + if test $count = $max ; then + echo retried with non zero exit status, $max times: "$@" >&2 + fi + cat $tmpfile + rm $tmpfile + return $status +} + +# +# map_enxio_to_eagain cmd arg ... +# +# add EAGAIN to the output of cmd arg ... if the output contains +# ENXIO. +# +function map_enxio_to_eagain() +{ + local status=0 + local tmpfile=$TEMP_DIR/map_enxio_to_eagain.$$ + + "$@" > $tmpfile 2>&1 || status=$? + if test $status != 0 && + grep --quiet ENXIO $tmpfile ; then + echo "EAGAIN added by $0::map_enxio_to_eagain" >> $tmpfile + fi + cat $tmpfile + rm $tmpfile + return $status +} + +function check_response() +{ + expected_string=$1 + retcode=$2 + expected_retcode=$3 + if [ "$expected_retcode" -a $retcode != $expected_retcode ] ; then + echo "return code invalid: got $retcode, expected $expected_retcode" >&2 + exit 1 + fi + + if ! grep --quiet -- "$expected_string" $TMPFILE ; then + echo "Didn't find $expected_string in output" >&2 + cat $TMPFILE >&2 + exit 1 + fi +} + +function get_config_value_or_die() +{ + local target config_opt raw val + + target=$1 + config_opt=$2 + + raw="`$SUDO ceph daemon $target config get $config_opt 2>/dev/null`" + if [[ $? -ne 0 ]]; then + echo "error obtaining config opt '$config_opt' from '$target': $raw" + exit 1 + fi + + raw=`echo $raw | sed -e 's/[{} "]//g'` + val=`echo $raw | cut -f2 -d:` + + echo "$val" + return 0 +} + +function expect_config_value() +{ + local target config_opt expected_val val + target=$1 + config_opt=$2 + expected_val=$3 + + val=$(get_config_value_or_die $target $config_opt) + + if [[ "$val" != "$expected_val" ]]; then + echo "expected '$expected_val', got '$val'" + exit 1 + fi +} + +function ceph_watch_start() +{ + local whatch_opt=--watch + + if [ -n "$1" ]; then + whatch_opt=--watch-$1 + if [ -n "$2" ]; then + whatch_opt+=" --watch-channel $2" + fi + fi + + CEPH_WATCH_FILE=${TEMP_DIR}/CEPH_WATCH_$$ + ceph $whatch_opt > $CEPH_WATCH_FILE & + CEPH_WATCH_PID=$! + + # wait until the "ceph" client is connected and receiving + # log messages from monitor + for i in `seq 3`; do + grep -q "cluster" $CEPH_WATCH_FILE && break + sleep 1 + done +} + +function ceph_watch_wait() +{ + local regexp=$1 + local timeout=30 + + if [ -n "$2" ]; then + timeout=$2 + fi + + for i in `seq ${timeout}`; do + grep -q "$regexp" $CEPH_WATCH_FILE && break + sleep 1 + done + + kill $CEPH_WATCH_PID + + if ! grep "$regexp" $CEPH_WATCH_FILE; then + echo "pattern ${regexp} not found in watch file. Full watch file content:" >&2 + cat $CEPH_WATCH_FILE >&2 + return 1 + fi +} + +function test_mon_injectargs() +{ + CEPH_ARGS='--mon_debug_dump_location the.dump' ceph tell osd.0 injectargs --no-osd_enable_op_tracker >& $TMPFILE || return 1 + check_response "osd_enable_op_tracker = 'false'" + ! grep "the.dump" $TMPFILE || return 1 + ceph tell osd.0 injectargs '--osd_enable_op_tracker --osd_op_history_duration 500' >& $TMPFILE || return 1 + check_response "osd_enable_op_tracker = 'true' osd_op_history_duration = '500'" + ceph tell osd.0 injectargs --no-osd_enable_op_tracker >& $TMPFILE || return 1 + check_response "osd_enable_op_tracker = 'false'" + ceph tell osd.0 injectargs -- --osd_enable_op_tracker >& $TMPFILE || return 1 + check_response "osd_enable_op_tracker = 'true'" + ceph tell osd.0 injectargs -- '--osd_enable_op_tracker --osd_op_history_duration 600' >& $TMPFILE || return 1 + check_response "osd_enable_op_tracker = 'true' osd_op_history_duration = '600'" + expect_failure $TEMP_DIR "Option --osd_op_history_duration requires an argument" \ + ceph tell osd.0 injectargs -- '--osd_op_history_duration' + + ceph tell osd.0 injectargs -- '--osd_deep_scrub_interval 2419200' >& $TMPFILE || return 1 + check_response "osd_deep_scrub_interval = '2419200.000000' (not observed, change may require restart)" + + ceph tell osd.0 injectargs -- '--mon_probe_timeout 2' >& $TMPFILE || return 1 + check_response "mon_probe_timeout = '2.000000' (not observed, change may require restart)" + + ceph tell osd.0 injectargs -- '--mon-lease 6' >& $TMPFILE || return 1 + check_response "mon_lease = '6.000000' (not observed, change may require restart)" + + # osd-scrub-auto-repair-num-errors is an OPT_U32, so -1 is not a valid setting + expect_false ceph tell osd.0 injectargs --osd-scrub-auto-repair-num-errors -1 >& $TMPFILE || return 1 + check_response "Error EINVAL: Parse error setting osd_scrub_auto_repair_num_errors to '-1' using injectargs" +} + +function test_mon_injectargs_SI() +{ + # Test SI units during injectargs and 'config set' + # We only aim at testing the units are parsed accordingly + # and don't intend to test whether the options being set + # actually expect SI units to be passed. + # Keep in mind that all integer based options (i.e., INT, + # LONG, U32, U64) will accept SI unit modifiers. + initial_value=$(get_config_value_or_die "mon.a" "mon_pg_warn_min_objects") + $SUDO ceph daemon mon.a config set mon_pg_warn_min_objects 10 + expect_config_value "mon.a" "mon_pg_warn_min_objects" 10 + $SUDO ceph daemon mon.a config set mon_pg_warn_min_objects 10K + expect_config_value "mon.a" "mon_pg_warn_min_objects" 10240 + $SUDO ceph daemon mon.a config set mon_pg_warn_min_objects 1G + expect_config_value "mon.a" "mon_pg_warn_min_objects" 1073741824 + $SUDO ceph daemon mon.a config set mon_pg_warn_min_objects 10F > $TMPFILE || true + check_response "'10F': (22) Invalid argument" + # now test with injectargs + ceph tell mon.a injectargs '--mon_pg_warn_min_objects 10' + expect_config_value "mon.a" "mon_pg_warn_min_objects" 10 + ceph tell mon.a injectargs '--mon_pg_warn_min_objects 10K' + expect_config_value "mon.a" "mon_pg_warn_min_objects" 10240 + ceph tell mon.a injectargs '--mon_pg_warn_min_objects 1G' + expect_config_value "mon.a" "mon_pg_warn_min_objects" 1073741824 + expect_false ceph tell mon.a injectargs '--mon_pg_warn_min_objects 10F' + expect_false ceph tell mon.a injectargs '--mon_globalid_prealloc -1' + $SUDO ceph daemon mon.a config set mon_pg_warn_min_objects $initial_value +} + +function test_tiering_agent() +{ + local slow=slow_eviction + local fast=fast_eviction + ceph osd pool create $slow 1 1 + ceph osd pool application enable $slow rados + ceph osd pool create $fast 1 1 + ceph osd tier add $slow $fast + ceph osd tier cache-mode $fast writeback + ceph osd tier set-overlay $slow $fast + ceph osd pool set $fast hit_set_type bloom + rados -p $slow put obj1 /etc/group + ceph osd pool set $fast target_max_objects 1 + ceph osd pool set $fast hit_set_count 1 + ceph osd pool set $fast hit_set_period 5 + # wait for the object to be evicted from the cache + local evicted + evicted=false + for i in `seq 1 300` ; do + if ! rados -p $fast ls | grep obj1 ; then + evicted=true + break + fi + sleep 1 + done + $evicted # assert + # the object is proxy read and promoted to the cache + rados -p $slow get obj1 - >/dev/null + # wait for the promoted object to be evicted again + evicted=false + for i in `seq 1 300` ; do + if ! rados -p $fast ls | grep obj1 ; then + evicted=true + break + fi + sleep 1 + done + $evicted # assert + ceph osd tier remove-overlay $slow + ceph osd tier remove $slow $fast + ceph osd pool delete $fast $fast --yes-i-really-really-mean-it + ceph osd pool delete $slow $slow --yes-i-really-really-mean-it +} + +function test_tiering_1() +{ + # tiering + ceph osd pool create slow 2 + ceph osd pool application enable slow rados + ceph osd pool create slow2 2 + ceph osd pool application enable slow2 rados + ceph osd pool create cache 2 + ceph osd pool create cache2 2 + ceph osd tier add slow cache + ceph osd tier add slow cache2 + expect_false ceph osd tier add slow2 cache + # test some state transitions + ceph osd tier cache-mode cache writeback + expect_false ceph osd tier cache-mode cache forward + ceph osd tier cache-mode cache forward --yes-i-really-mean-it + expect_false ceph osd tier cache-mode cache readonly + ceph osd tier cache-mode cache readonly --yes-i-really-mean-it + expect_false ceph osd tier cache-mode cache forward + ceph osd tier cache-mode cache forward --yes-i-really-mean-it + ceph osd tier cache-mode cache none + ceph osd tier cache-mode cache writeback + ceph osd tier cache-mode cache proxy + ceph osd tier cache-mode cache writeback + expect_false ceph osd tier cache-mode cache none + expect_false ceph osd tier cache-mode cache readonly --yes-i-really-mean-it + # test with dirty objects in the tier pool + # tier pool currently set to 'writeback' + rados -p cache put /etc/passwd /etc/passwd + flush_pg_stats + # 1 dirty object in pool 'cache' + ceph osd tier cache-mode cache proxy + expect_false ceph osd tier cache-mode cache none + expect_false ceph osd tier cache-mode cache readonly --yes-i-really-mean-it + ceph osd tier cache-mode cache writeback + # remove object from tier pool + rados -p cache rm /etc/passwd + rados -p cache cache-flush-evict-all + flush_pg_stats + # no dirty objects in pool 'cache' + ceph osd tier cache-mode cache proxy + ceph osd tier cache-mode cache none + ceph osd tier cache-mode cache readonly --yes-i-really-mean-it + TRIES=0 + while ! ceph osd pool set cache pg_num 3 --yes-i-really-mean-it 2>$TMPFILE + do + grep 'currently creating pgs' $TMPFILE + TRIES=$(( $TRIES + 1 )) + test $TRIES -ne 60 + sleep 3 + done + expect_false ceph osd pool set cache pg_num 4 + ceph osd tier cache-mode cache none + ceph osd tier set-overlay slow cache + expect_false ceph osd tier set-overlay slow cache2 + expect_false ceph osd tier remove slow cache + ceph osd tier remove-overlay slow + ceph osd tier set-overlay slow cache2 + ceph osd tier remove-overlay slow + ceph osd tier remove slow cache + ceph osd tier add slow2 cache + expect_false ceph osd tier set-overlay slow cache + ceph osd tier set-overlay slow2 cache + ceph osd tier remove-overlay slow2 + ceph osd tier remove slow2 cache + ceph osd tier remove slow cache2 + + # make sure a non-empty pool fails + rados -p cache2 put /etc/passwd /etc/passwd + while ! ceph df | grep cache2 | grep ' 1 ' ; do + echo waiting for pg stats to flush + sleep 2 + done + expect_false ceph osd tier add slow cache2 + ceph osd tier add slow cache2 --force-nonempty + ceph osd tier remove slow cache2 + + ceph osd pool ls | grep cache2 + ceph osd pool ls -f json-pretty | grep cache2 + ceph osd pool ls detail | grep cache2 + ceph osd pool ls detail -f json-pretty | grep cache2 + + ceph osd pool delete slow slow --yes-i-really-really-mean-it + ceph osd pool delete slow2 slow2 --yes-i-really-really-mean-it + ceph osd pool delete cache cache --yes-i-really-really-mean-it + ceph osd pool delete cache2 cache2 --yes-i-really-really-mean-it +} + +function test_tiering_2() +{ + # make sure we can't clobber snapshot state + ceph osd pool create snap_base 2 + ceph osd pool application enable snap_base rados + ceph osd pool create snap_cache 2 + ceph osd pool mksnap snap_cache snapname + expect_false ceph osd tier add snap_base snap_cache + ceph osd pool delete snap_base snap_base --yes-i-really-really-mean-it + ceph osd pool delete snap_cache snap_cache --yes-i-really-really-mean-it +} + +function test_tiering_3() +{ + # make sure we can't create snapshot on tier + ceph osd pool create basex 2 + ceph osd pool application enable basex rados + ceph osd pool create cachex 2 + ceph osd tier add basex cachex + expect_false ceph osd pool mksnap cache snapname + ceph osd tier remove basex cachex + ceph osd pool delete basex basex --yes-i-really-really-mean-it + ceph osd pool delete cachex cachex --yes-i-really-really-mean-it +} + +function test_tiering_4() +{ + # make sure we can't create an ec pool tier + ceph osd pool create eccache 2 2 erasure + expect_false ceph osd set-require-min-compat-client bobtail + ceph osd pool create repbase 2 + ceph osd pool application enable repbase rados + expect_false ceph osd tier add repbase eccache + ceph osd pool delete repbase repbase --yes-i-really-really-mean-it + ceph osd pool delete eccache eccache --yes-i-really-really-mean-it +} + +function test_tiering_5() +{ + # convenient add-cache command + ceph osd pool create slow 2 + ceph osd pool application enable slow rados + ceph osd pool create cache3 2 + ceph osd tier add-cache slow cache3 1024000 + ceph osd dump | grep cache3 | grep bloom | grep 'false_positive_probability: 0.05' | grep 'target_bytes 1024000' | grep '1200s x4' + ceph osd tier remove slow cache3 2> $TMPFILE || true + check_response "EBUSY: tier pool 'cache3' is the overlay for 'slow'; please remove-overlay first" + ceph osd tier remove-overlay slow + ceph osd tier remove slow cache3 + ceph osd pool ls | grep cache3 + ceph osd pool delete cache3 cache3 --yes-i-really-really-mean-it + ! ceph osd pool ls | grep cache3 || exit 1 + ceph osd pool delete slow slow --yes-i-really-really-mean-it +} + +function test_tiering_6() +{ + # check add-cache whether work + ceph osd pool create datapool 2 + ceph osd pool application enable datapool rados + ceph osd pool create cachepool 2 + ceph osd tier add-cache datapool cachepool 1024000 + ceph osd tier cache-mode cachepool writeback + rados -p datapool put object /etc/passwd + rados -p cachepool stat object + rados -p cachepool cache-flush object + rados -p datapool stat object + ceph osd tier remove-overlay datapool + ceph osd tier remove datapool cachepool + ceph osd pool delete cachepool cachepool --yes-i-really-really-mean-it + ceph osd pool delete datapool datapool --yes-i-really-really-mean-it +} + +function test_tiering_7() +{ + # protection against pool removal when used as tiers + ceph osd pool create datapool 2 + ceph osd pool application enable datapool rados + ceph osd pool create cachepool 2 + ceph osd tier add-cache datapool cachepool 1024000 + ceph osd pool delete cachepool cachepool --yes-i-really-really-mean-it 2> $TMPFILE || true + check_response "EBUSY: pool 'cachepool' is a tier of 'datapool'" + ceph osd pool delete datapool datapool --yes-i-really-really-mean-it 2> $TMPFILE || true + check_response "EBUSY: pool 'datapool' has tiers cachepool" + ceph osd tier remove-overlay datapool + ceph osd tier remove datapool cachepool + ceph osd pool delete cachepool cachepool --yes-i-really-really-mean-it + ceph osd pool delete datapool datapool --yes-i-really-really-mean-it +} + +function test_tiering_8() +{ + ## check health check + ceph osd set notieragent + ceph osd pool create datapool 2 + ceph osd pool application enable datapool rados + ceph osd pool create cache4 2 + ceph osd tier add-cache datapool cache4 1024000 + ceph osd tier cache-mode cache4 writeback + tmpfile=$(mktemp|grep tmp) + dd if=/dev/zero of=$tmpfile bs=4K count=1 + ceph osd pool set cache4 target_max_objects 200 + ceph osd pool set cache4 target_max_bytes 1000000 + rados -p cache4 put foo1 $tmpfile + rados -p cache4 put foo2 $tmpfile + rm -f $tmpfile + flush_pg_stats + ceph df | grep datapool | grep ' 2 ' + ceph osd tier remove-overlay datapool + ceph osd tier remove datapool cache4 + ceph osd pool delete cache4 cache4 --yes-i-really-really-mean-it + ceph osd pool delete datapool datapool --yes-i-really-really-mean-it + ceph osd unset notieragent +} + +function test_tiering_9() +{ + # make sure 'tier remove' behaves as we expect + # i.e., removing a tier from a pool that's not its base pool only + # results in a 'pool foo is now (or already was) not a tier of bar' + # + ceph osd pool create basepoolA 2 + ceph osd pool application enable basepoolA rados + ceph osd pool create basepoolB 2 + ceph osd pool application enable basepoolB rados + poolA_id=$(ceph osd dump | grep 'pool.*basepoolA' | awk '{print $2;}') + poolB_id=$(ceph osd dump | grep 'pool.*basepoolB' | awk '{print $2;}') + + ceph osd pool create cache5 2 + ceph osd pool create cache6 2 + ceph osd tier add basepoolA cache5 + ceph osd tier add basepoolB cache6 + ceph osd tier remove basepoolB cache5 2>&1 | grep 'not a tier of' + ceph osd dump | grep "pool.*'cache5'" 2>&1 | grep "tier_of[ \t]\+$poolA_id" + ceph osd tier remove basepoolA cache6 2>&1 | grep 'not a tier of' + ceph osd dump | grep "pool.*'cache6'" 2>&1 | grep "tier_of[ \t]\+$poolB_id" + + ceph osd tier remove basepoolA cache5 2>&1 | grep 'not a tier of' + ! ceph osd dump | grep "pool.*'cache5'" 2>&1 | grep "tier_of" || exit 1 + ceph osd tier remove basepoolB cache6 2>&1 | grep 'not a tier of' + ! ceph osd dump | grep "pool.*'cache6'" 2>&1 | grep "tier_of" || exit 1 + + ! ceph osd dump | grep "pool.*'basepoolA'" 2>&1 | grep "tiers" || exit 1 + ! ceph osd dump | grep "pool.*'basepoolB'" 2>&1 | grep "tiers" || exit 1 + + ceph osd pool delete cache6 cache6 --yes-i-really-really-mean-it + ceph osd pool delete cache5 cache5 --yes-i-really-really-mean-it + ceph osd pool delete basepoolB basepoolB --yes-i-really-really-mean-it + ceph osd pool delete basepoolA basepoolA --yes-i-really-really-mean-it +} + +function test_auth() +{ + ceph auth add client.xx mon allow osd "allow *" + ceph auth export client.xx >client.xx.keyring + ceph auth add client.xx -i client.xx.keyring + rm -f client.xx.keyring + ceph auth list | grep client.xx + ceph auth ls | grep client.xx + ceph auth get client.xx | grep caps | grep mon + ceph auth get client.xx | grep caps | grep osd + ceph auth get-key client.xx + ceph auth print-key client.xx + ceph auth print_key client.xx + ceph auth caps client.xx osd "allow rw" + expect_false sh <<< "ceph auth get client.xx | grep caps | grep mon" + ceph auth get client.xx | grep osd | grep "allow rw" + ceph auth export | grep client.xx + ceph auth export -o authfile + ceph auth import -i authfile + ceph auth export -o authfile2 + diff authfile authfile2 + rm authfile authfile2 + ceph auth del client.xx + expect_false ceph auth get client.xx + + # (almost) interactive mode + echo -e 'auth add client.xx mon allow osd "allow *"\n' | ceph + ceph auth get client.xx + # script mode + echo 'auth del client.xx' | ceph + expect_false ceph auth get client.xx + + # + # get / set auid + # + local auid=444 + ceph-authtool --create-keyring --name client.TEST --gen-key --set-uid $auid TEST-keyring + expect_false ceph auth import --in-file TEST-keyring + rm TEST-keyring + ceph-authtool --create-keyring --name client.TEST --gen-key --cap mon "allow r" --set-uid $auid TEST-keyring + ceph auth import --in-file TEST-keyring + rm TEST-keyring + ceph auth get client.TEST > $TMPFILE + check_response "auid = $auid" + ceph --format json-pretty auth get client.TEST > $TMPFILE + check_response '"auid": '$auid + ceph auth ls > $TMPFILE + check_response "auid: $auid" + ceph --format json-pretty auth ls > $TMPFILE + check_response '"auid": '$auid + ceph auth del client.TEST +} + +function test_auth_profiles() +{ + ceph auth add client.xx-profile-ro mon 'allow profile read-only' \ + mgr 'allow profile read-only' + ceph auth add client.xx-profile-rw mon 'allow profile read-write' \ + mgr 'allow profile read-write' + ceph auth add client.xx-profile-rd mon 'allow profile role-definer' + + ceph auth export > client.xx.keyring + + # read-only is allowed all read-only commands (auth excluded) + ceph -n client.xx-profile-ro -k client.xx.keyring status + ceph -n client.xx-profile-ro -k client.xx.keyring osd dump + ceph -n client.xx-profile-ro -k client.xx.keyring pg dump + ceph -n client.xx-profile-ro -k client.xx.keyring mon dump + ceph -n client.xx-profile-ro -k client.xx.keyring mds dump + # read-only gets access denied for rw commands or auth commands + ceph -n client.xx-profile-ro -k client.xx.keyring log foo >& $TMPFILE || true + check_response "EACCES: access denied" + ceph -n client.xx-profile-ro -k client.xx.keyring osd set noout >& $TMPFILE || true + check_response "EACCES: access denied" + ceph -n client.xx-profile-ro -k client.xx.keyring auth ls >& $TMPFILE || true + check_response "EACCES: access denied" + + # read-write is allowed for all read-write commands (except auth) + ceph -n client.xx-profile-rw -k client.xx.keyring status + ceph -n client.xx-profile-rw -k client.xx.keyring osd dump + ceph -n client.xx-profile-rw -k client.xx.keyring pg dump + ceph -n client.xx-profile-rw -k client.xx.keyring mon dump + ceph -n client.xx-profile-rw -k client.xx.keyring mds dump + ceph -n client.xx-profile-rw -k client.xx.keyring log foo + ceph -n client.xx-profile-rw -k client.xx.keyring osd set noout + ceph -n client.xx-profile-rw -k client.xx.keyring osd unset noout + # read-write gets access denied for auth commands + ceph -n client.xx-profile-rw -k client.xx.keyring auth ls >& $TMPFILE || true + check_response "EACCES: access denied" + + # role-definer is allowed RWX 'auth' commands and read-only 'mon' commands + ceph -n client.xx-profile-rd -k client.xx.keyring auth ls + ceph -n client.xx-profile-rd -k client.xx.keyring auth export + ceph -n client.xx-profile-rd -k client.xx.keyring auth add client.xx-profile-foo + ceph -n client.xx-profile-rd -k client.xx.keyring status + ceph -n client.xx-profile-rd -k client.xx.keyring osd dump >& $TMPFILE || true + check_response "EACCES: access denied" + ceph -n client.xx-profile-rd -k client.xx.keyring pg dump >& $TMPFILE || true + check_response "EACCES: access denied" + # read-only 'mon' subsystem commands are allowed + ceph -n client.xx-profile-rd -k client.xx.keyring mon dump + # but read-write 'mon' commands are not + ceph -n client.xx-profile-rd -k client.xx.keyring mon add foo 1.1.1.1 >& $TMPFILE || true + check_response "EACCES: access denied" + ceph -n client.xx-profile-rd -k client.xx.keyring mds dump >& $TMPFILE || true + check_response "EACCES: access denied" + ceph -n client.xx-profile-rd -k client.xx.keyring log foo >& $TMPFILE || true + check_response "EACCES: access denied" + ceph -n client.xx-profile-rd -k client.xx.keyring osd set noout >& $TMPFILE || true + check_response "EACCES: access denied" + + ceph -n client.xx-profile-rd -k client.xx.keyring auth del client.xx-profile-ro + ceph -n client.xx-profile-rd -k client.xx.keyring auth del client.xx-profile-rw + + # add a new role-definer with the existing role-definer + ceph -n client.xx-profile-rd -k client.xx.keyring \ + auth add client.xx-profile-rd2 mon 'allow profile role-definer' + ceph -n client.xx-profile-rd -k client.xx.keyring \ + auth export > client.xx.keyring.2 + # remove old role-definer using the new role-definer + ceph -n client.xx-profile-rd2 -k client.xx.keyring.2 \ + auth del client.xx-profile-rd + # remove the remaining role-definer with admin + ceph auth del client.xx-profile-rd2 + rm -f client.xx.keyring client.xx.keyring.2 +} + +function test_mon_caps() +{ + ceph-authtool --create-keyring $TEMP_DIR/ceph.client.bug.keyring + chmod +r $TEMP_DIR/ceph.client.bug.keyring + ceph-authtool $TEMP_DIR/ceph.client.bug.keyring -n client.bug --gen-key + ceph auth add client.bug -i $TEMP_DIR/ceph.client.bug.keyring + + rados lspools --keyring $TEMP_DIR/ceph.client.bug.keyring -n client.bug >& $TMPFILE || true + check_response "Permission denied" + + rm -rf $TEMP_DIR/ceph.client.bug.keyring + ceph auth del client.bug + ceph-authtool --create-keyring $TEMP_DIR/ceph.client.bug.keyring + chmod +r $TEMP_DIR/ceph.client.bug.keyring + ceph-authtool $TEMP_DIR/ceph.client.bug.keyring -n client.bug --gen-key + ceph-authtool -n client.bug --cap mon '' $TEMP_DIR/ceph.client.bug.keyring + ceph auth add client.bug -i $TEMP_DIR/ceph.client.bug.keyring + rados lspools --keyring $TEMP_DIR/ceph.client.bug.keyring -n client.bug >& $TMPFILE || true + check_response "Permission denied" +} + +function test_mon_misc() +{ + # with and without verbosity + ceph osd dump | grep '^epoch' + ceph --concise osd dump | grep '^epoch' + + ceph osd df | grep 'MIN/MAX VAR' + + # df + ceph df > $TMPFILE + grep GLOBAL $TMPFILE + grep -v DIRTY $TMPFILE + ceph df detail > $TMPFILE + grep DIRTY $TMPFILE + ceph df --format json > $TMPFILE + grep 'total_bytes' $TMPFILE + grep -v 'dirty' $TMPFILE + ceph df detail --format json > $TMPFILE + grep 'rd_bytes' $TMPFILE + grep 'dirty' $TMPFILE + ceph df --format xml | grep '' + ceph df detail --format xml | grep '' + + ceph fsid + ceph health + ceph health detail + ceph health --format json-pretty + ceph health detail --format xml-pretty + + ceph time-sync-status + + ceph node ls + for t in mon osd mds ; do + ceph node ls $t + done + + ceph_watch_start + mymsg="this is a test log message $$.$(date)" + ceph log "$mymsg" + ceph log last | grep "$mymsg" + ceph log last 100 | grep "$mymsg" + ceph_watch_wait "$mymsg" + + ceph mgr dump + ceph mgr module ls + ceph mgr module enable restful + expect_false ceph mgr module enable foodne + ceph mgr module enable foodne --force + ceph mgr module disable foodne + ceph mgr module disable foodnebizbangbash + + ceph mon metadata a + ceph mon metadata + ceph mon count-metadata ceph_version + ceph mon versions + + ceph mgr metadata + ceph mgr versions + ceph mgr count-metadata ceph_version + + ceph versions + + ceph node ls +} + +function check_mds_active() +{ + fs_name=$1 + ceph fs get $fs_name | grep active +} + +function wait_mds_active() +{ + fs_name=$1 + max_run=300 + for i in $(seq 1 $max_run) ; do + if ! check_mds_active $fs_name ; then + echo "waiting for an active MDS daemon ($i/$max_run)" + sleep 5 + else + break + fi + done + check_mds_active $fs_name +} + +function get_mds_gids() +{ + fs_name=$1 + ceph fs get $fs_name --format=json | python -c "import json; import sys; print ' '.join([m['gid'].__str__() for m in json.load(sys.stdin)['mdsmap']['info'].values()])" +} + +function fail_all_mds() +{ + fs_name=$1 + ceph fs set $fs_name cluster_down true + mds_gids=$(get_mds_gids $fs_name) + for mds_gid in $mds_gids ; do + ceph mds fail $mds_gid + done + if check_mds_active $fs_name ; then + echo "An active MDS remains, something went wrong" + ceph fs get $fs_name + exit -1 + fi + +} + +function remove_all_fs() +{ + existing_fs=$(ceph fs ls --format=json | python -c "import json; import sys; print ' '.join([fs['name'] for fs in json.load(sys.stdin)])") + for fs_name in $existing_fs ; do + echo "Removing fs ${fs_name}..." + fail_all_mds $fs_name + echo "Removing existing filesystem '${fs_name}'..." + ceph fs rm $fs_name --yes-i-really-mean-it + echo "Removed '${fs_name}'." + done +} + +# So that tests requiring MDS can skip if one is not configured +# in the cluster at all +function mds_exists() +{ + ceph auth ls | grep "^mds" +} + +# some of the commands are just not idempotent. +function without_test_dup_command() +{ + if [ -z ${CEPH_CLI_TEST_DUP_COMMAND+x} ]; then + $@ + else + local saved=${CEPH_CLI_TEST_DUP_COMMAND} + unset CEPH_CLI_TEST_DUP_COMMAND + $@ + CEPH_CLI_TEST_DUP_COMMAND=saved + fi +} + +function test_mds_tell() +{ + local FS_NAME=cephfs + if ! mds_exists ; then + echo "Skipping test, no MDS found" + return + fi + + remove_all_fs + ceph osd pool create fs_data 10 + ceph osd pool create fs_metadata 10 + ceph fs new $FS_NAME fs_metadata fs_data + wait_mds_active $FS_NAME + + # Test injectargs by GID + old_mds_gids=$(get_mds_gids $FS_NAME) + echo Old GIDs: $old_mds_gids + + for mds_gid in $old_mds_gids ; do + ceph tell mds.$mds_gid injectargs "--debug-mds 20" + done + expect_false ceph tell mds.a injectargs mds_max_file_recover -1 + + # Test respawn by rank + without_test_dup_command ceph tell mds.0 respawn + new_mds_gids=$old_mds_gids + while [ $new_mds_gids -eq $old_mds_gids ] ; do + sleep 5 + new_mds_gids=$(get_mds_gids $FS_NAME) + done + echo New GIDs: $new_mds_gids + + # Test respawn by ID + without_test_dup_command ceph tell mds.a respawn + new_mds_gids=$old_mds_gids + while [ $new_mds_gids -eq $old_mds_gids ] ; do + sleep 5 + new_mds_gids=$(get_mds_gids $FS_NAME) + done + echo New GIDs: $new_mds_gids + + remove_all_fs + ceph osd pool delete fs_data fs_data --yes-i-really-really-mean-it + ceph osd pool delete fs_metadata fs_metadata --yes-i-really-really-mean-it +} + +function test_mon_mds() +{ + local FS_NAME=cephfs + remove_all_fs + + ceph osd pool create fs_data 10 + ceph osd pool create fs_metadata 10 + ceph fs new $FS_NAME fs_metadata fs_data + + ceph fs set $FS_NAME cluster_down true + ceph fs set $FS_NAME cluster_down false + + # Legacy commands, act on default fs + ceph mds cluster_down + ceph mds cluster_up + + ceph mds compat rm_incompat 4 + ceph mds compat rm_incompat 4 + + # We don't want any MDSs to be up, their activity can interfere with + # the "current_epoch + 1" checking below if they're generating updates + fail_all_mds $FS_NAME + + ceph mds compat show + expect_false ceph mds deactivate 2 + ceph mds dump + ceph fs dump + ceph fs get $FS_NAME + for mds_gid in $(get_mds_gids $FS_NAME) ; do + ceph mds metadata $mds_id + done + ceph mds metadata + ceph mds versions + ceph mds count-metadata os + + # XXX mds fail, but how do you undo it? + mdsmapfile=$TEMP_DIR/mdsmap.$$ + current_epoch=$(ceph mds getmap -o $mdsmapfile --no-log-to-stderr 2>&1 | grep epoch | sed 's/.*epoch //') + [ -s $mdsmapfile ] + rm $mdsmapfile + + ceph osd pool create data2 10 + ceph osd pool create data3 10 + data2_pool=$(ceph osd dump | grep "pool.*'data2'" | awk '{print $2;}') + data3_pool=$(ceph osd dump | grep "pool.*'data3'" | awk '{print $2;}') + ceph mds add_data_pool $data2_pool + ceph mds add_data_pool $data3_pool + ceph mds add_data_pool 100 >& $TMPFILE || true + check_response "Error ENOENT" + ceph mds add_data_pool foobarbaz >& $TMPFILE || true + check_response "Error ENOENT" + ceph mds remove_data_pool $data2_pool + ceph mds remove_data_pool $data3_pool + ceph osd pool delete data2 data2 --yes-i-really-really-mean-it + ceph osd pool delete data3 data3 --yes-i-really-really-mean-it + ceph mds set allow_multimds false + expect_false ceph mds set_max_mds 4 + ceph mds set allow_multimds true + ceph mds set_max_mds 4 + ceph mds set_max_mds 3 + ceph mds set_max_mds 256 + expect_false ceph mds set_max_mds 257 + ceph mds set max_mds 4 + ceph mds set max_mds 256 + expect_false ceph mds set max_mds 257 + expect_false ceph mds set max_mds asdf + expect_false ceph mds set inline_data true + ceph mds set inline_data true --yes-i-really-mean-it + ceph mds set inline_data yes --yes-i-really-mean-it + ceph mds set inline_data 1 --yes-i-really-mean-it + expect_false ceph mds set inline_data --yes-i-really-mean-it + ceph mds set inline_data false + ceph mds set inline_data no + ceph mds set inline_data 0 + expect_false ceph mds set inline_data asdf + ceph mds set max_file_size 1048576 + expect_false ceph mds set max_file_size 123asdf + + expect_false ceph mds set allow_new_snaps + expect_false ceph mds set allow_new_snaps true + ceph mds set allow_new_snaps true --yes-i-really-mean-it + ceph mds set allow_new_snaps 0 + ceph mds set allow_new_snaps false + ceph mds set allow_new_snaps no + expect_false ceph mds set allow_new_snaps taco + + # we should never be able to add EC pools as data or metadata pools + # create an ec-pool... + ceph osd pool create mds-ec-pool 10 10 erasure + set +e + ceph mds add_data_pool mds-ec-pool 2>$TMPFILE + check_response 'erasure-code' $? 22 + set -e + ec_poolnum=$(ceph osd dump | grep "pool.* 'mds-ec-pool" | awk '{print $2;}') + data_poolnum=$(ceph osd dump | grep "pool.* 'fs_data" | awk '{print $2;}') + metadata_poolnum=$(ceph osd dump | grep "pool.* 'fs_metadata" | awk '{print $2;}') + + fail_all_mds $FS_NAME + + set +e + # Check that rmfailed requires confirmation + expect_false ceph mds rmfailed 0 + ceph mds rmfailed 0 --yes-i-really-mean-it + set -e + + # Check that `newfs` is no longer permitted + expect_false ceph mds newfs $metadata_poolnum $data_poolnum --yes-i-really-mean-it 2>$TMPFILE + + # Check that 'fs reset' runs + ceph fs reset $FS_NAME --yes-i-really-mean-it + + # Check that creating a second FS fails by default + ceph osd pool create fs_metadata2 10 + ceph osd pool create fs_data2 10 + set +e + expect_false ceph fs new cephfs2 fs_metadata2 fs_data2 + set -e + + # Check that setting enable_multiple enables creation of second fs + ceph fs flag set enable_multiple true --yes-i-really-mean-it + ceph fs new cephfs2 fs_metadata2 fs_data2 + + # Clean up multi-fs stuff + fail_all_mds cephfs2 + ceph fs rm cephfs2 --yes-i-really-mean-it + ceph osd pool delete fs_metadata2 fs_metadata2 --yes-i-really-really-mean-it + ceph osd pool delete fs_data2 fs_data2 --yes-i-really-really-mean-it + + fail_all_mds $FS_NAME + + # Clean up to enable subsequent fs new tests + ceph fs rm $FS_NAME --yes-i-really-mean-it + + set +e + ceph fs new $FS_NAME fs_metadata mds-ec-pool --force 2>$TMPFILE + check_response 'erasure-code' $? 22 + ceph fs new $FS_NAME mds-ec-pool fs_data 2>$TMPFILE + check_response 'erasure-code' $? 22 + ceph fs new $FS_NAME mds-ec-pool mds-ec-pool 2>$TMPFILE + check_response 'erasure-code' $? 22 + set -e + + # ... new create a cache tier in front of the EC pool... + ceph osd pool create mds-tier 2 + ceph osd tier add mds-ec-pool mds-tier + ceph osd tier set-overlay mds-ec-pool mds-tier + tier_poolnum=$(ceph osd dump | grep "pool.* 'mds-tier" | awk '{print $2;}') + + # Use of a readonly tier should be forbidden + ceph osd tier cache-mode mds-tier readonly --yes-i-really-mean-it + set +e + ceph fs new $FS_NAME fs_metadata mds-ec-pool --force 2>$TMPFILE + check_response 'has a write tier (mds-tier) that is configured to forward' $? 22 + set -e + + # Use of a writeback tier should enable FS creation + ceph osd tier cache-mode mds-tier writeback + ceph fs new $FS_NAME fs_metadata mds-ec-pool --force + + # While a FS exists using the tiered pools, I should not be allowed + # to remove the tier + set +e + ceph osd tier remove-overlay mds-ec-pool 2>$TMPFILE + check_response 'in use by CephFS' $? 16 + ceph osd tier remove mds-ec-pool mds-tier 2>$TMPFILE + check_response 'in use by CephFS' $? 16 + set -e + + fail_all_mds $FS_NAME + ceph fs rm $FS_NAME --yes-i-really-mean-it + + # ... but we should be forbidden from using the cache pool in the FS directly. + set +e + ceph fs new $FS_NAME fs_metadata mds-tier --force 2>$TMPFILE + check_response 'in use as a cache tier' $? 22 + ceph fs new $FS_NAME mds-tier fs_data 2>$TMPFILE + check_response 'in use as a cache tier' $? 22 + ceph fs new $FS_NAME mds-tier mds-tier 2>$TMPFILE + check_response 'in use as a cache tier' $? 22 + set -e + + # Clean up tier + EC pools + ceph osd tier remove-overlay mds-ec-pool + ceph osd tier remove mds-ec-pool mds-tier + + # Create a FS using the 'cache' pool now that it's no longer a tier + ceph fs new $FS_NAME fs_metadata mds-tier --force + + # We should be forbidden from using this pool as a tier now that + # it's in use for CephFS + set +e + ceph osd tier add mds-ec-pool mds-tier 2>$TMPFILE + check_response 'in use by CephFS' $? 16 + set -e + + fail_all_mds $FS_NAME + ceph fs rm $FS_NAME --yes-i-really-mean-it + + # We should be permitted to use an EC pool with overwrites enabled + # as the data pool... + ceph osd pool set mds-ec-pool allow_ec_overwrites true + ceph fs new $FS_NAME fs_metadata mds-ec-pool --force 2>$TMPFILE + fail_all_mds $FS_NAME + ceph fs rm $FS_NAME --yes-i-really-mean-it + + # ...but not as the metadata pool + set +e + ceph fs new $FS_NAME mds-ec-pool fs_data 2>$TMPFILE + check_response 'erasure-code' $? 22 + set -e + + ceph osd pool delete mds-ec-pool mds-ec-pool --yes-i-really-really-mean-it + + # Create a FS and check that we can subsequently add a cache tier to it + ceph fs new $FS_NAME fs_metadata fs_data --force + + # Adding overlay to FS pool should be permitted, RADOS clients handle this. + ceph osd tier add fs_metadata mds-tier + ceph osd tier cache-mode mds-tier writeback + ceph osd tier set-overlay fs_metadata mds-tier + + # Removing tier should be permitted because the underlying pool is + # replicated (#11504 case) + ceph osd tier cache-mode mds-tier proxy + ceph osd tier remove-overlay fs_metadata + ceph osd tier remove fs_metadata mds-tier + ceph osd pool delete mds-tier mds-tier --yes-i-really-really-mean-it + + # Clean up FS + fail_all_mds $FS_NAME + ceph fs rm $FS_NAME --yes-i-really-mean-it + + + + ceph mds stat + # ceph mds tell mds.a getmap + # ceph mds rm + # ceph mds rmfailed + # ceph mds set_state + # ceph mds stop + + ceph osd pool delete fs_data fs_data --yes-i-really-really-mean-it + ceph osd pool delete fs_metadata fs_metadata --yes-i-really-really-mean-it +} + +function test_mon_mds_metadata() +{ + local nmons=$(ceph tell 'mon.*' version | grep -c 'version') + test "$nmons" -gt 0 + + ceph mds dump | + sed -nEe "s/^([0-9]+):.*'([a-z])' mds\\.([0-9]+)\\..*/\\1 \\2 \\3/p" | + while read gid id rank; do + ceph mds metadata ${gid} | grep '"hostname":' + ceph mds metadata ${id} | grep '"hostname":' + ceph mds metadata ${rank} | grep '"hostname":' + + local n=$(ceph tell 'mon.*' mds metadata ${id} | grep -c '"hostname":') + test "$n" -eq "$nmons" + done + + expect_false ceph mds metadata UNKNOWN +} + +function test_mon_mon() +{ + # print help message + ceph --help mon + # no mon add/remove + ceph mon dump + ceph mon getmap -o $TEMP_DIR/monmap.$$ + [ -s $TEMP_DIR/monmap.$$ ] + # ceph mon tell + ceph mon_status + + # test mon features + ceph mon feature ls + ceph mon feature set kraken --yes-i-really-mean-it + expect_false ceph mon feature set abcd + expect_false ceph mon feature set abcd --yes-i-really-mean-it +} + +function gen_secrets_file() +{ + # lets assume we can have the following types + # all - generates both cephx and lockbox, with mock dm-crypt key + # cephx - only cephx + # no_cephx - lockbox and dm-crypt, no cephx + # no_lockbox - dm-crypt and cephx, no lockbox + # empty - empty file + # empty_json - correct json, empty map + # bad_json - bad json :) + # + local t=$1 + if [[ -z "$t" ]]; then + t="all" + fi + + fn=$(mktemp $TEMP_DIR/secret.XXXXXX) + echo $fn + if [[ "$t" == "empty" ]]; then + return 0 + fi + + echo "{" > $fn + if [[ "$t" == "bad_json" ]]; then + echo "asd: ; }" >> $fn + return 0 + elif [[ "$t" == "empty_json" ]]; then + echo "}" >> $fn + return 0 + fi + + cephx_secret="\"cephx_secret\": \"$(ceph-authtool --gen-print-key)\"" + lb_secret="\"cephx_lockbox_secret\": \"$(ceph-authtool --gen-print-key)\"" + dmcrypt_key="\"dmcrypt_key\": \"$(ceph-authtool --gen-print-key)\"" + + if [[ "$t" == "all" ]]; then + echo "$cephx_secret,$lb_secret,$dmcrypt_key" >> $fn + elif [[ "$t" == "cephx" ]]; then + echo "$cephx_secret" >> $fn + elif [[ "$t" == "no_cephx" ]]; then + echo "$lb_secret,$dmcrypt_key" >> $fn + elif [[ "$t" == "no_lockbox" ]]; then + echo "$cephx_secret,$dmcrypt_key" >> $fn + else + echo "unknown gen_secrets_file() type \'$fn\'" + return 1 + fi + echo "}" >> $fn + return 0 +} + +function test_mon_osd_create_destroy() +{ + ceph osd new 2>&1 | grep 'EINVAL' + ceph osd new '' -1 2>&1 | grep 'EINVAL' + ceph osd new '' 10 2>&1 | grep 'EINVAL' + + old_maxosd=$(ceph osd getmaxosd | sed -e 's/max_osd = //' -e 's/ in epoch.*//') + + old_osds=$(ceph osd ls) + num_osds=$(ceph osd ls | wc -l) + + uuid=$(uuidgen) + id=$(ceph osd new $uuid 2>/dev/null) + + for i in $old_osds; do + [[ "$i" != "$id" ]] + done + + ceph osd find $id + + id2=`ceph osd new $uuid 2>/dev/null` + + [[ $id2 == $id ]] + + ceph osd new $uuid $id + + id3=$(ceph osd getmaxosd | sed -e 's/max_osd = //' -e 's/ in epoch.*//') + ceph osd new $uuid $((id3+1)) 2>&1 | grep EEXIST + + uuid2=$(uuidgen) + id2=$(ceph osd new $uuid2) + ceph osd find $id2 + [[ "$id2" != "$id" ]] + + ceph osd new $uuid $id2 2>&1 | grep EEXIST + ceph osd new $uuid2 $id2 + + # test with secrets + empty_secrets=$(gen_secrets_file "empty") + empty_json=$(gen_secrets_file "empty_json") + all_secrets=$(gen_secrets_file "all") + cephx_only=$(gen_secrets_file "cephx") + no_cephx=$(gen_secrets_file "no_cephx") + no_lockbox=$(gen_secrets_file "no_lockbox") + bad_json=$(gen_secrets_file "bad_json") + + # empty secrets should be idempotent + new_id=$(ceph osd new $uuid $id -i $empty_secrets) + [[ "$new_id" == "$id" ]] + + # empty json, thus empty secrets + new_id=$(ceph osd new $uuid $id -i $empty_json) + [[ "$new_id" == "$id" ]] + + ceph osd new $uuid $id -i $all_secrets 2>&1 | grep 'EEXIST' + + ceph osd rm $id + ceph osd rm $id2 + ceph osd setmaxosd $old_maxosd + + ceph osd new $uuid -i $bad_json 2>&1 | grep 'EINVAL' + ceph osd new $uuid -i $no_cephx 2>&1 | grep 'EINVAL' + ceph osd new $uuid -i $no_lockbox 2>&1 | grep 'EINVAL' + + osds=$(ceph osd ls) + id=$(ceph osd new $uuid -i $all_secrets) + for i in $osds; do + [[ "$i" != "$id" ]] + done + + ceph osd find $id + + # validate secrets and dm-crypt are set + k=$(ceph auth get-key osd.$id --format=json-pretty 2>/dev/null | jq '.key') + s=$(cat $all_secrets | jq '.cephx_secret') + [[ $k == $s ]] + k=$(ceph auth get-key client.osd-lockbox.$uuid --format=json-pretty 2>/dev/null | \ + jq '.key') + s=$(cat $all_secrets | jq '.cephx_lockbox_secret') + [[ $k == $s ]] + ceph config-key exists dm-crypt/osd/$uuid/luks + + osds=$(ceph osd ls) + id2=$(ceph osd new $uuid2 -i $cephx_only) + for i in $osds; do + [[ "$i" != "$id2" ]] + done + + ceph osd find $id2 + k=$(ceph auth get-key osd.$id --format=json-pretty 2>/dev/null | jq '.key') + s=$(cat $all_secrets | jq '.cephx_secret') + [[ $k == $s ]] + expect_false ceph auth get-key client.osd-lockbox.$uuid2 + expect_false ceph config-key exists dm-crypt/osd/$uuid2/luks + + ceph osd destroy osd.$id2 --yes-i-really-mean-it + ceph osd destroy $id2 --yes-i-really-mean-it + ceph osd find $id2 + expect_false ceph auth get-key osd.$id2 + ceph osd dump | grep osd.$id2 | grep destroyed + + id3=$id2 + uuid3=$(uuidgen) + ceph osd new $uuid3 $id3 -i $all_secrets + ceph osd dump | grep osd.$id3 | expect_false grep destroyed + ceph auth get-key client.osd-lockbox.$uuid3 + ceph auth get-key osd.$id3 + ceph config-key exists dm-crypt/osd/$uuid3/luks + + ceph osd purge osd.$id3 --yes-i-really-mean-it + expect_false ceph osd find $id2 + expect_false ceph auth get-key osd.$id2 + expect_false ceph auth get-key client.osd-lockbox.$uuid3 + expect_false ceph config-key exists dm-crypt/osd/$uuid3/luks + ceph osd purge osd.$id3 --yes-i-really-mean-it + ceph osd purge osd.$id3 --yes-i-really-mean-it # idempotent + + ceph osd purge osd.$id --yes-i-really-mean-it + ceph osd purge 123456 --yes-i-really-mean-it + expect_false ceph osd find $id + expect_false ceph auth get-key osd.$id + expect_false ceph auth get-key client.osd-lockbox.$uuid + expect_false ceph config-key exists dm-crypt/osd/$uuid/luks + + rm $empty_secrets $empty_json $all_secrets $cephx_only \ + $no_cephx $no_lockbox $bad_json + + for i in $(ceph osd ls); do + [[ "$i" != "$id" ]] + [[ "$i" != "$id2" ]] + [[ "$i" != "$id3" ]] + done + + [[ "$(ceph osd ls | wc -l)" == "$num_osds" ]] + ceph osd setmaxosd $old_maxosd + +} + +function test_mon_config_key() +{ + key=asdfasdfqwerqwreasdfuniquesa123df + ceph config-key list | grep -c $key | grep 0 + ceph config-key get $key | grep -c bar | grep 0 + ceph config-key set $key bar + ceph config-key get $key | grep bar + ceph config-key list | grep -c $key | grep 1 + ceph config-key dump | grep $key | grep bar + ceph config-key rm $key + expect_false ceph config-key get $key + ceph config-key list | grep -c $key | grep 0 + ceph config-key dump | grep -c $key | grep 0 +} + +function test_mon_osd() +{ + # + # osd blacklist + # + bl=192.168.0.1:0/1000 + ceph osd blacklist add $bl + ceph osd blacklist ls | grep $bl + ceph osd blacklist ls --format=json-pretty | sed 's/\\\//\//' | grep $bl + ceph osd dump --format=json-pretty | grep $bl + ceph osd dump | grep "^blacklist $bl" + ceph osd blacklist rm $bl + ceph osd blacklist ls | expect_false grep $bl + + bl=192.168.0.1 + # test without nonce, invalid nonce + ceph osd blacklist add $bl + ceph osd blacklist ls | grep $bl + ceph osd blacklist rm $bl + ceph osd blacklist ls | expect_false grep $expect_false bl + expect_false "ceph osd blacklist $bl/-1" + expect_false "ceph osd blacklist $bl/foo" + + # test with wrong address + expect_false "ceph osd blacklist 1234.56.78.90/100" + + # Test `clear` + ceph osd blacklist add $bl + ceph osd blacklist ls | grep $bl + ceph osd blacklist clear + ceph osd blacklist ls | expect_false grep $bl + + # + # osd crush + # + ceph osd crush reweight-all + ceph osd crush tunables legacy + ceph osd crush show-tunables | grep argonaut + ceph osd crush tunables bobtail + ceph osd crush show-tunables | grep bobtail + ceph osd crush tunables firefly + ceph osd crush show-tunables | grep firefly + + ceph osd crush set-tunable straw_calc_version 0 + ceph osd crush get-tunable straw_calc_version | grep 0 + ceph osd crush set-tunable straw_calc_version 1 + ceph osd crush get-tunable straw_calc_version | grep 1 + + # + # require-min-compat-client + expect_false ceph osd set-require-min-compat-client dumpling # firefly tunables + ceph osd set-require-min-compat-client luminous + ceph osd dump | grep 'require_min_compat_client luminous' + + # + # osd scrub + # + # how do I tell when these are done? + ceph osd scrub 0 + ceph osd deep-scrub 0 + ceph osd repair 0 + + for f in noup nodown noin noout noscrub nodeep-scrub nobackfill norebalance norecover notieragent full + do + ceph osd set $f + ceph osd unset $f + done + expect_false ceph osd unset sortbitwise # cannot be unset + expect_false ceph osd set bogus + expect_false ceph osd unset bogus + ceph osd require-osd-release luminous + # can't lower (or use new command for anything but jewel) + expect_false ceph osd require-osd-release jewel + # these are no-ops but should succeed. + ceph osd set require_jewel_osds + ceph osd set require_kraken_osds + expect_false ceph osd unset require_jewel_osds + + ceph osd set noup + ceph osd down 0 + ceph osd dump | grep 'osd.0 down' + ceph osd unset noup + max_run=1000 + for ((i=0; i < $max_run; i++)); do + if ! ceph osd dump | grep 'osd.0 up'; then + echo "waiting for osd.0 to come back up ($i/$max_run)" + sleep 1 + else + break + fi + done + ceph osd dump | grep 'osd.0 up' + + ceph osd dump | grep 'osd.0 up' + # ceph osd find expects the OsdName, so both ints and osd.n should work. + ceph osd find 1 + ceph osd find osd.1 + expect_false ceph osd find osd.xyz + expect_false ceph osd find xyz + expect_false ceph osd find 0.1 + ceph --format plain osd find 1 # falls back to json-pretty + if [ `uname` == Linux ]; then + ceph osd metadata 1 | grep 'distro' + ceph --format plain osd metadata 1 | grep 'distro' # falls back to json-pretty + fi + ceph osd out 0 + ceph osd dump | grep 'osd.0.*out' + ceph osd in 0 + ceph osd dump | grep 'osd.0.*in' + ceph osd find 0 + + ceph osd add-nodown 0 1 + ceph health detail | grep 'NODOWN' + ceph osd rm-nodown 0 1 + ! ceph health detail | grep 'NODOWN' + + ceph osd out 0 # so we can mark it as noin later + ceph osd add-noin 0 + ceph health detail | grep 'NOIN' + ceph osd rm-noin 0 + ! ceph health detail | grep 'NOIN' + ceph osd in 0 + + ceph osd add-noout 0 + ceph health detail | grep 'NOOUT' + ceph osd rm-noout 0 + ! ceph health detail | grep 'NOOUT' + + # test osd id parse + expect_false ceph osd add-noup 797er + expect_false ceph osd add-nodown u9uwer + expect_false ceph osd add-noin 78~15 + expect_false ceph osd add-noout 0 all 1 + + expect_false ceph osd rm-noup 1234567 + expect_false ceph osd rm-nodown fsadf7 + expect_false ceph osd rm-noin 0 1 any + expect_false ceph osd rm-noout 790-fd + + ids=`ceph osd ls-tree default` + for osd in $ids + do + ceph osd add-nodown $osd + ceph osd add-noout $osd + done + ceph -s | grep 'NODOWN' + ceph -s | grep 'NOOUT' + ceph osd rm-nodown any + ceph osd rm-noout all + ! ceph -s | grep 'NODOWN' + ! ceph -s | grep 'NOOUT' + + # make sure mark out preserves weight + ceph osd reweight osd.0 .5 + ceph osd dump | grep ^osd.0 | grep 'weight 0.5' + ceph osd out 0 + ceph osd in 0 + ceph osd dump | grep ^osd.0 | grep 'weight 0.5' + + ceph osd getmap -o $f + [ -s $f ] + rm $f + save=$(ceph osd getmaxosd | sed -e 's/max_osd = //' -e 's/ in epoch.*//') + [ "$save" -gt 0 ] + ceph osd setmaxosd $((save - 1)) 2>&1 | grep 'EBUSY' + ceph osd setmaxosd 10 + ceph osd getmaxosd | grep 'max_osd = 10' + ceph osd setmaxosd $save + ceph osd getmaxosd | grep "max_osd = $save" + + for id in `ceph osd ls` ; do + retry_eagain 5 map_enxio_to_eagain ceph tell osd.$id version + done + + ceph osd rm 0 2>&1 | grep 'EBUSY' + + local old_osds=$(echo $(ceph osd ls)) + id=`ceph osd create` + ceph osd find $id + ceph osd lost $id --yes-i-really-mean-it + expect_false ceph osd setmaxosd $id + local new_osds=$(echo $(ceph osd ls)) + for id in $(echo $new_osds | sed -e "s/$old_osds//") ; do + ceph osd rm $id + done + + uuid=`uuidgen` + id=`ceph osd create $uuid` + id2=`ceph osd create $uuid` + [ "$id" = "$id2" ] + ceph osd rm $id + + ceph --help osd + + # reset max_osd. + ceph osd setmaxosd $id + ceph osd getmaxosd | grep "max_osd = $save" + local max_osd=$save + + ceph osd create $uuid 0 2>&1 | grep 'EINVAL' + ceph osd create $uuid $((max_osd - 1)) 2>&1 | grep 'EINVAL' + + id=`ceph osd create $uuid $max_osd` + [ "$id" = "$max_osd" ] + ceph osd find $id + max_osd=$((max_osd + 1)) + ceph osd getmaxosd | grep "max_osd = $max_osd" + + ceph osd create $uuid $((id - 1)) 2>&1 | grep 'EEXIST' + ceph osd create $uuid $((id + 1)) 2>&1 | grep 'EEXIST' + id2=`ceph osd create $uuid` + [ "$id" = "$id2" ] + id2=`ceph osd create $uuid $id` + [ "$id" = "$id2" ] + + uuid=`uuidgen` + local gap_start=$max_osd + id=`ceph osd create $uuid $((gap_start + 100))` + [ "$id" = "$((gap_start + 100))" ] + max_osd=$((id + 1)) + ceph osd getmaxosd | grep "max_osd = $max_osd" + + ceph osd create $uuid $gap_start 2>&1 | grep 'EEXIST' + + # + # When CEPH_CLI_TEST_DUP_COMMAND is set, osd create + # is repeated and consumes two osd id, not just one. + # + local next_osd=$gap_start + id=`ceph osd create $(uuidgen)` + [ "$id" = "$next_osd" ] + + next_osd=$((id + 1)) + id=`ceph osd create $(uuidgen) $next_osd` + [ "$id" = "$next_osd" ] + + local new_osds=$(echo $(ceph osd ls)) + for id in $(echo $new_osds | sed -e "s/$old_osds//") ; do + [ $id -ge $save ] + ceph osd rm $id + done + ceph osd setmaxosd $save + + ceph osd ls + ceph osd pool create data 10 + ceph osd pool application enable data rados + ceph osd lspools | grep data + ceph osd map data foo | grep 'pool.*data.*object.*foo.*pg.*up.*acting' + ceph osd map data foo namespace| grep 'pool.*data.*object.*namespace/foo.*pg.*up.*acting' + ceph osd pool delete data data --yes-i-really-really-mean-it + + ceph osd pause + ceph osd dump | grep 'flags.*pauserd,pausewr' + ceph osd unpause + + ceph osd tree + ceph osd tree up + ceph osd tree down + ceph osd tree in + ceph osd tree out + ceph osd tree destroyed + ceph osd tree up in + ceph osd tree up out + ceph osd tree down in + ceph osd tree down out + ceph osd tree out down + expect_false ceph osd tree up down + expect_false ceph osd tree up destroyed + expect_false ceph osd tree down destroyed + expect_false ceph osd tree up down destroyed + expect_false ceph osd tree in out + expect_false ceph osd tree up foo + + ceph osd metadata + ceph osd count-metadata os + ceph osd versions + + ceph osd perf + ceph osd blocked-by + + ceph osd stat | grep up, +} + +function test_mon_crush() +{ + f=$TEMP_DIR/map.$$ + epoch=$(ceph osd getcrushmap -o $f 2>&1 | tail -n1) + [ -s $f ] + [ "$epoch" -gt 1 ] + nextepoch=$(( $epoch + 1 )) + echo epoch $epoch nextepoch $nextepoch + rm -f $f.epoch + expect_false ceph osd setcrushmap $nextepoch -i $f + gotepoch=$(ceph osd setcrushmap $epoch -i $f 2>&1 | tail -n1) + echo gotepoch $gotepoch + [ "$gotepoch" -eq "$nextepoch" ] + # should be idempotent + gotepoch=$(ceph osd setcrushmap $epoch -i $f 2>&1 | tail -n1) + echo epoch $gotepoch + [ "$gotepoch" -eq "$nextepoch" ] + rm $f +} + +function test_mon_osd_pool() +{ + # + # osd pool + # + ceph osd pool create data 10 + ceph osd pool application enable data rados + ceph osd pool mksnap data datasnap + rados -p data lssnap | grep datasnap + ceph osd pool rmsnap data datasnap + expect_false ceph osd pool rmsnap pool_fake snapshot + ceph osd pool delete data data --yes-i-really-really-mean-it + + ceph osd pool create data2 10 + ceph osd pool application enable data2 rados + ceph osd pool rename data2 data3 + ceph osd lspools | grep data3 + ceph osd pool delete data3 data3 --yes-i-really-really-mean-it + + ceph osd pool create replicated 12 12 replicated + ceph osd pool create replicated 12 12 replicated + ceph osd pool create replicated 12 12 # default is replicated + ceph osd pool create replicated 12 # default is replicated, pgp_num = pg_num + ceph osd pool application enable replicated rados + # should fail because the type is not the same + expect_false ceph osd pool create replicated 12 12 erasure + ceph osd lspools | grep replicated + ceph osd pool create ec_test 1 1 erasure + ceph osd pool application enable ec_test rados + set +e + ceph osd count-metadata osd_objectstore | grep 'bluestore' + if [ $? -eq 1 ]; then # enable ec_overwrites on non-bluestore pools should fail + ceph osd pool set ec_test allow_ec_overwrites true >& $TMPFILE + check_response "pool must only be stored on bluestore for scrubbing to work" $? 22 + else + ceph osd pool set ec_test allow_ec_overwrites true || return 1 + expect_false ceph osd pool set ec_test allow_ec_overwrites false + fi + set -e + ceph osd pool delete replicated replicated --yes-i-really-really-mean-it + ceph osd pool delete ec_test ec_test --yes-i-really-really-mean-it +} + +function test_mon_osd_pool_quota() +{ + # + # test osd pool set/get quota + # + + # create tmp pool + ceph osd pool create tmp-quota-pool 36 + ceph osd pool application enable tmp-quota-pool rados + # + # set erroneous quotas + # + expect_false ceph osd pool set-quota tmp-quota-pool max_fooness 10 + expect_false ceph osd pool set-quota tmp-quota-pool max_bytes -1 + expect_false ceph osd pool set-quota tmp-quota-pool max_objects aaa + # + # set valid quotas + # + ceph osd pool set-quota tmp-quota-pool max_bytes 10 + ceph osd pool set-quota tmp-quota-pool max_objects 10M + # + # get quotas + # + ceph osd pool get-quota tmp-quota-pool | grep 'max bytes.*10B' + ceph osd pool get-quota tmp-quota-pool | grep 'max objects.*10240k objects' + # + # get quotas in json-pretty format + # + ceph osd pool get-quota tmp-quota-pool --format=json-pretty | \ + grep '"quota_max_objects":.*10485760' + ceph osd pool get-quota tmp-quota-pool --format=json-pretty | \ + grep '"quota_max_bytes":.*10' + # + # reset pool quotas + # + ceph osd pool set-quota tmp-quota-pool max_bytes 0 + ceph osd pool set-quota tmp-quota-pool max_objects 0 + # + # test N/A quotas + # + ceph osd pool get-quota tmp-quota-pool | grep 'max bytes.*N/A' + ceph osd pool get-quota tmp-quota-pool | grep 'max objects.*N/A' + # + # cleanup tmp pool + ceph osd pool delete tmp-quota-pool tmp-quota-pool --yes-i-really-really-mean-it +} + +function test_mon_pg() +{ + # Make sure we start healthy. + wait_for_health_ok + + ceph pg debug unfound_objects_exist + ceph pg debug degraded_pgs_exist + ceph pg deep-scrub 1.0 + ceph pg dump + ceph pg dump pgs_brief --format=json + ceph pg dump pgs --format=json + ceph pg dump pools --format=json + ceph pg dump osds --format=json + ceph pg dump sum --format=json + ceph pg dump all --format=json + ceph pg dump pgs_brief osds --format=json + ceph pg dump pools osds pgs_brief --format=json + ceph pg dump_json + ceph pg dump_pools_json + ceph pg dump_stuck inactive + ceph pg dump_stuck unclean + ceph pg dump_stuck stale + ceph pg dump_stuck undersized + ceph pg dump_stuck degraded + ceph pg ls + ceph pg ls 1 + ceph pg ls stale + expect_false ceph pg ls scrubq + ceph pg ls active stale repair recovering + ceph pg ls 1 active + ceph pg ls 1 active stale + ceph pg ls-by-primary osd.0 + ceph pg ls-by-primary osd.0 1 + ceph pg ls-by-primary osd.0 active + ceph pg ls-by-primary osd.0 active stale + ceph pg ls-by-primary osd.0 1 active stale + ceph pg ls-by-osd osd.0 + ceph pg ls-by-osd osd.0 1 + ceph pg ls-by-osd osd.0 active + ceph pg ls-by-osd osd.0 active stale + ceph pg ls-by-osd osd.0 1 active stale + ceph pg ls-by-pool rbd + ceph pg ls-by-pool rbd active stale + # can't test this... + # ceph pg force_create_pg + ceph pg getmap -o $TEMP_DIR/map.$$ + [ -s $TEMP_DIR/map.$$ ] + ceph pg map 1.0 | grep acting + ceph pg repair 1.0 + ceph pg scrub 1.0 + + ceph osd set-full-ratio .962 + ceph osd dump | grep '^full_ratio 0.962' + ceph osd set-backfillfull-ratio .912 + ceph osd dump | grep '^backfillfull_ratio 0.912' + ceph osd set-nearfull-ratio .892 + ceph osd dump | grep '^nearfull_ratio 0.892' + + # Check health status + ceph osd set-nearfull-ratio .913 + ceph health -f json | grep OSD_OUT_OF_ORDER_FULL + ceph health detail | grep OSD_OUT_OF_ORDER_FULL + ceph osd set-nearfull-ratio .892 + ceph osd set-backfillfull-ratio .963 + ceph health -f json | grep OSD_OUT_OF_ORDER_FULL + ceph health detail | grep OSD_OUT_OF_ORDER_FULL + ceph osd set-backfillfull-ratio .912 + + # Check injected full results + $SUDO ceph --admin-daemon $(get_admin_socket osd.0) injectfull nearfull + wait_for_health "OSD_NEARFULL" + ceph health detail | grep "osd.0 is near full" + $SUDO ceph --admin-daemon $(get_admin_socket osd.0) injectfull none + wait_for_health_ok + + $SUDO ceph --admin-daemon $(get_admin_socket osd.1) injectfull backfillfull + wait_for_health "OSD_BACKFILLFULL" + ceph health detail | grep "osd.1 is backfill full" + $SUDO ceph --admin-daemon $(get_admin_socket osd.1) injectfull none + wait_for_health_ok + + $SUDO ceph --admin-daemon $(get_admin_socket osd.2) injectfull failsafe + # failsafe and full are the same as far as the monitor is concerned + wait_for_health "OSD_FULL" + ceph health detail | grep "osd.2 is full" + $SUDO ceph --admin-daemon $(get_admin_socket osd.2) injectfull none + wait_for_health_ok + + $SUDO ceph --admin-daemon $(get_admin_socket osd.0) injectfull full + wait_for_health "OSD_FULL" + ceph health detail | grep "osd.0 is full" + $SUDO ceph --admin-daemon $(get_admin_socket osd.0) injectfull none + wait_for_health_ok + + ceph pg stat | grep 'pgs:' + ceph pg 1.0 query + ceph tell 1.0 query + ceph quorum enter + ceph quorum_status + ceph report | grep osd_stats + ceph status + ceph -s + + # + # tell osd version + # + ceph tell osd.0 version + expect_false ceph tell osd.9999 version + expect_false ceph tell osd.foo version + + # back to pg stuff + + ceph tell osd.0 dump_pg_recovery_stats | grep Started + + ceph osd reweight 0 0.9 + expect_false ceph osd reweight 0 -1 + ceph osd reweight osd.0 1 + + ceph osd primary-affinity osd.0 .9 + expect_false ceph osd primary-affinity osd.0 -2 + expect_false ceph osd primary-affinity osd.9999 .5 + ceph osd primary-affinity osd.0 1 + + ceph osd pool set rbd size 2 + ceph osd pg-temp 1.0 0 1 + ceph osd pg-temp 1.0 osd.1 osd.0 + expect_false ceph osd pg-temp 1.0 0 1 2 + expect_false ceph osd pg-temp asdf qwer + expect_false ceph osd pg-temp 1.0 asdf + expect_false ceph osd pg-temp 1.0 + + # don't test ceph osd primary-temp for now +} + +function test_mon_osd_pool_set() +{ + TEST_POOL_GETSET=pool_getset + ceph osd pool create $TEST_POOL_GETSET 1 + ceph osd pool application enable $TEST_POOL_GETSET rados + wait_for_clean + ceph osd pool get $TEST_POOL_GETSET all + + for s in pg_num pgp_num size min_size crush_rule; do + ceph osd pool get $TEST_POOL_GETSET $s + done + + old_size=$(ceph osd pool get $TEST_POOL_GETSET size | sed -e 's/size: //') + (( new_size = old_size + 1 )) + ceph osd pool set $TEST_POOL_GETSET size $new_size + ceph osd pool get $TEST_POOL_GETSET size | grep "size: $new_size" + ceph osd pool set $TEST_POOL_GETSET size $old_size + + ceph osd pool create pool_erasure 1 1 erasure + ceph osd pool application enable pool_erasure rados + wait_for_clean + set +e + ceph osd pool set pool_erasure size 4444 2>$TMPFILE + check_response 'not change the size' + set -e + ceph osd pool get pool_erasure erasure_code_profile + + auid=5555 + ceph osd pool set $TEST_POOL_GETSET auid $auid + ceph osd pool get $TEST_POOL_GETSET auid | grep $auid + ceph --format=xml osd pool get $TEST_POOL_GETSET auid | grep $auid + ceph osd pool set $TEST_POOL_GETSET auid 0 + + for flag in nodelete nopgchange nosizechange write_fadvise_dontneed noscrub nodeep-scrub; do + ceph osd pool set $TEST_POOL_GETSET $flag false + ceph osd pool get $TEST_POOL_GETSET $flag | grep "$flag: false" + ceph osd pool set $TEST_POOL_GETSET $flag true + ceph osd pool get $TEST_POOL_GETSET $flag | grep "$flag: true" + ceph osd pool set $TEST_POOL_GETSET $flag 1 + ceph osd pool get $TEST_POOL_GETSET $flag | grep "$flag: true" + ceph osd pool set $TEST_POOL_GETSET $flag 0 + ceph osd pool get $TEST_POOL_GETSET $flag | grep "$flag: false" + expect_false ceph osd pool set $TEST_POOL_GETSET $flag asdf + expect_false ceph osd pool set $TEST_POOL_GETSET $flag 2 + done + + ceph osd pool get $TEST_POOL_GETSET scrub_min_interval | expect_false grep '.' + ceph osd pool set $TEST_POOL_GETSET scrub_min_interval 123456 + ceph osd pool get $TEST_POOL_GETSET scrub_min_interval | grep 'scrub_min_interval: 123456' + ceph osd pool set $TEST_POOL_GETSET scrub_min_interval 0 + ceph osd pool get $TEST_POOL_GETSET scrub_min_interval | expect_false grep '.' + + ceph osd pool get $TEST_POOL_GETSET scrub_max_interval | expect_false grep '.' + ceph osd pool set $TEST_POOL_GETSET scrub_max_interval 123456 + ceph osd pool get $TEST_POOL_GETSET scrub_max_interval | grep 'scrub_max_interval: 123456' + ceph osd pool set $TEST_POOL_GETSET scrub_max_interval 0 + ceph osd pool get $TEST_POOL_GETSET scrub_max_interval | expect_false grep '.' + + ceph osd pool get $TEST_POOL_GETSET deep_scrub_interval | expect_false grep '.' + ceph osd pool set $TEST_POOL_GETSET deep_scrub_interval 123456 + ceph osd pool get $TEST_POOL_GETSET deep_scrub_interval | grep 'deep_scrub_interval: 123456' + ceph osd pool set $TEST_POOL_GETSET deep_scrub_interval 0 + ceph osd pool get $TEST_POOL_GETSET deep_scrub_interval | expect_false grep '.' + + ceph osd pool get $TEST_POOL_GETSET recovery_priority | expect_false grep '.' + ceph osd pool set $TEST_POOL_GETSET recovery_priority 5 + ceph osd pool get $TEST_POOL_GETSET recovery_priority | grep 'recovery_priority: 5' + ceph osd pool set $TEST_POOL_GETSET recovery_priority 0 + ceph osd pool get $TEST_POOL_GETSET recovery_priority | expect_false grep '.' + + ceph osd pool get $TEST_POOL_GETSET recovery_op_priority | expect_false grep '.' + ceph osd pool set $TEST_POOL_GETSET recovery_op_priority 5 + ceph osd pool get $TEST_POOL_GETSET recovery_op_priority | grep 'recovery_op_priority: 5' + ceph osd pool set $TEST_POOL_GETSET recovery_op_priority 0 + ceph osd pool get $TEST_POOL_GETSET recovery_op_priority | expect_false grep '.' + + ceph osd pool get $TEST_POOL_GETSET scrub_priority | expect_false grep '.' + ceph osd pool set $TEST_POOL_GETSET scrub_priority 5 + ceph osd pool get $TEST_POOL_GETSET scrub_priority | grep 'scrub_priority: 5' + ceph osd pool set $TEST_POOL_GETSET scrub_priority 0 + ceph osd pool get $TEST_POOL_GETSET scrub_priority | expect_false grep '.' + + ceph osd pool set $TEST_POOL_GETSET nopgchange 1 + expect_false ceph osd pool set $TEST_POOL_GETSET pg_num 10 + expect_false ceph osd pool set $TEST_POOL_GETSET pgp_num 10 + ceph osd pool set $TEST_POOL_GETSET nopgchange 0 + ceph osd pool set $TEST_POOL_GETSET pg_num 10 + wait_for_clean + ceph osd pool set $TEST_POOL_GETSET pgp_num 10 + + old_pgs=$(ceph osd pool get $TEST_POOL_GETSET pg_num | sed -e 's/pg_num: //') + new_pgs=$(($old_pgs + $(ceph osd stat --format json | jq '.num_osds') * 32)) + ceph osd pool set $TEST_POOL_GETSET pg_num $new_pgs + ceph osd pool set $TEST_POOL_GETSET pgp_num $new_pgs + wait_for_clean + old_pgs=$(ceph osd pool get $TEST_POOL_GETSET pg_num | sed -e 's/pg_num: //') + new_pgs=$(($old_pgs + $(ceph osd stat --format json | jq '.num_osds') * 32 + 1)) + expect_false ceph osd pool set $TEST_POOL_GETSET pg_num $new_pgs + + ceph osd pool set $TEST_POOL_GETSET nosizechange 1 + expect_false ceph osd pool set $TEST_POOL_GETSET size 2 + expect_false ceph osd pool set $TEST_POOL_GETSET min_size 2 + ceph osd pool set $TEST_POOL_GETSET nosizechange 0 + ceph osd pool set $TEST_POOL_GETSET size 2 + wait_for_clean + ceph osd pool set $TEST_POOL_GETSET min_size 2 + + expect_false ceph osd pool set $TEST_POOL_GETSET hashpspool 0 + ceph osd pool set $TEST_POOL_GETSET hashpspool 0 --yes-i-really-mean-it + + expect_false ceph osd pool set $TEST_POOL_GETSET hashpspool 1 + ceph osd pool set $TEST_POOL_GETSET hashpspool 1 --yes-i-really-mean-it + + ceph osd pool get rbd crush_rule | grep 'crush_rule: ' + + ceph osd pool get $TEST_POOL_GETSET compression_mode | expect_false grep '.' + ceph osd pool set $TEST_POOL_GETSET compression_mode aggressive + ceph osd pool get $TEST_POOL_GETSET compression_mode | grep 'aggressive' + ceph osd pool set $TEST_POOL_GETSET compression_mode unset + ceph osd pool get $TEST_POOL_GETSET compression_mode | expect_false grep '.' + + ceph osd pool get $TEST_POOL_GETSET compression_algorithm | expect_false grep '.' + ceph osd pool set $TEST_POOL_GETSET compression_algorithm zlib + ceph osd pool get $TEST_POOL_GETSET compression_algorithm | grep 'zlib' + ceph osd pool set $TEST_POOL_GETSET compression_algorithm unset + ceph osd pool get $TEST_POOL_GETSET compression_algorithm | expect_false grep '.' + + ceph osd pool get $TEST_POOL_GETSET compression_required_ratio | expect_false grep '.' + expect_false ceph osd pool set $TEST_POOL_GETSET compression_required_ratio 1.1 + expect_false ceph osd pool set $TEST_POOL_GETSET compression_required_ratio -.2 + ceph osd pool set $TEST_POOL_GETSET compression_required_ratio .2 + ceph osd pool get $TEST_POOL_GETSET compression_required_ratio | grep '.2' + ceph osd pool set $TEST_POOL_GETSET compression_required_ratio 0 + ceph osd pool get $TEST_POOL_GETSET compression_required_ratio | expect_false grep '.' + + ceph osd pool get $TEST_POOL_GETSET csum_type | expect_false grep '.' + ceph osd pool set $TEST_POOL_GETSET csum_type crc32c + ceph osd pool get $TEST_POOL_GETSET csum_type | grep 'crc32c' + ceph osd pool set $TEST_POOL_GETSET csum_type unset + ceph osd pool get $TEST_POOL_GETSET csum_type | expect_false grep '.' + + for size in compression_max_blob_size compression_min_blob_size csum_max_block csum_min_block; do + ceph osd pool get $TEST_POOL_GETSET $size | expect_false grep '.' + ceph osd pool set $TEST_POOL_GETSET $size 100 + ceph osd pool get $TEST_POOL_GETSET $size | grep '100' + ceph osd pool set $TEST_POOL_GETSET $size 0 + ceph osd pool get $TEST_POOL_GETSET $size | expect_false grep '.' + done + + ceph osd pool set $TEST_POOL_GETSET nodelete 1 + expect_false ceph osd pool delete $TEST_POOL_GETSET $TEST_POOL_GETSET --yes-i-really-really-mean-it + ceph osd pool set $TEST_POOL_GETSET nodelete 0 + ceph osd pool delete $TEST_POOL_GETSET $TEST_POOL_GETSET --yes-i-really-really-mean-it + +} + +function test_mon_osd_tiered_pool_set() +{ + # this is really a tier pool + ceph osd pool create real-tier 2 + ceph osd tier add rbd real-tier + + ceph osd pool set real-tier hit_set_type explicit_hash + ceph osd pool get real-tier hit_set_type | grep "hit_set_type: explicit_hash" + ceph osd pool set real-tier hit_set_type explicit_object + ceph osd pool get real-tier hit_set_type | grep "hit_set_type: explicit_object" + ceph osd pool set real-tier hit_set_type bloom + ceph osd pool get real-tier hit_set_type | grep "hit_set_type: bloom" + expect_false ceph osd pool set real-tier hit_set_type i_dont_exist + ceph osd pool set real-tier hit_set_period 123 + ceph osd pool get real-tier hit_set_period | grep "hit_set_period: 123" + ceph osd pool set real-tier hit_set_count 12 + ceph osd pool get real-tier hit_set_count | grep "hit_set_count: 12" + ceph osd pool set real-tier hit_set_fpp .01 + ceph osd pool get real-tier hit_set_fpp | grep "hit_set_fpp: 0.01" + + ceph osd pool set real-tier target_max_objects 123 + ceph osd pool get real-tier target_max_objects | \ + grep 'target_max_objects:[ \t]\+123' + ceph osd pool set real-tier target_max_bytes 123456 + ceph osd pool get real-tier target_max_bytes | \ + grep 'target_max_bytes:[ \t]\+123456' + ceph osd pool set real-tier cache_target_dirty_ratio .123 + ceph osd pool get real-tier cache_target_dirty_ratio | \ + grep 'cache_target_dirty_ratio:[ \t]\+0.123' + expect_false ceph osd pool set real-tier cache_target_dirty_ratio -.2 + expect_false ceph osd pool set real-tier cache_target_dirty_ratio 1.1 + ceph osd pool set real-tier cache_target_dirty_high_ratio .123 + ceph osd pool get real-tier cache_target_dirty_high_ratio | \ + grep 'cache_target_dirty_high_ratio:[ \t]\+0.123' + expect_false ceph osd pool set real-tier cache_target_dirty_high_ratio -.2 + expect_false ceph osd pool set real-tier cache_target_dirty_high_ratio 1.1 + ceph osd pool set real-tier cache_target_full_ratio .123 + ceph osd pool get real-tier cache_target_full_ratio | \ + grep 'cache_target_full_ratio:[ \t]\+0.123' + ceph osd dump -f json-pretty | grep '"cache_target_full_ratio_micro": 123000' + ceph osd pool set real-tier cache_target_full_ratio 1.0 + ceph osd pool set real-tier cache_target_full_ratio 0 + expect_false ceph osd pool set real-tier cache_target_full_ratio 1.1 + ceph osd pool set real-tier cache_min_flush_age 123 + ceph osd pool get real-tier cache_min_flush_age | \ + grep 'cache_min_flush_age:[ \t]\+123' + ceph osd pool set real-tier cache_min_evict_age 234 + ceph osd pool get real-tier cache_min_evict_age | \ + grep 'cache_min_evict_age:[ \t]\+234' + + # this is not a tier pool + ceph osd pool create fake-tier 2 + ceph osd pool application enable fake-tier rados + wait_for_clean + + expect_false ceph osd pool set fake-tier hit_set_type explicit_hash + expect_false ceph osd pool get fake-tier hit_set_type + expect_false ceph osd pool set fake-tier hit_set_type explicit_object + expect_false ceph osd pool get fake-tier hit_set_type + expect_false ceph osd pool set fake-tier hit_set_type bloom + expect_false ceph osd pool get fake-tier hit_set_type + expect_false ceph osd pool set fake-tier hit_set_type i_dont_exist + expect_false ceph osd pool set fake-tier hit_set_period 123 + expect_false ceph osd pool get fake-tier hit_set_period + expect_false ceph osd pool set fake-tier hit_set_count 12 + expect_false ceph osd pool get fake-tier hit_set_count + expect_false ceph osd pool set fake-tier hit_set_fpp .01 + expect_false ceph osd pool get fake-tier hit_set_fpp + + expect_false ceph osd pool set fake-tier target_max_objects 123 + expect_false ceph osd pool get fake-tier target_max_objects + expect_false ceph osd pool set fake-tier target_max_bytes 123456 + expect_false ceph osd pool get fake-tier target_max_bytes + expect_false ceph osd pool set fake-tier cache_target_dirty_ratio .123 + expect_false ceph osd pool get fake-tier cache_target_dirty_ratio + expect_false ceph osd pool set fake-tier cache_target_dirty_ratio -.2 + expect_false ceph osd pool set fake-tier cache_target_dirty_ratio 1.1 + expect_false ceph osd pool set fake-tier cache_target_dirty_high_ratio .123 + expect_false ceph osd pool get fake-tier cache_target_dirty_high_ratio + expect_false ceph osd pool set fake-tier cache_target_dirty_high_ratio -.2 + expect_false ceph osd pool set fake-tier cache_target_dirty_high_ratio 1.1 + expect_false ceph osd pool set fake-tier cache_target_full_ratio .123 + expect_false ceph osd pool get fake-tier cache_target_full_ratio + expect_false ceph osd pool set fake-tier cache_target_full_ratio 1.0 + expect_false ceph osd pool set fake-tier cache_target_full_ratio 0 + expect_false ceph osd pool set fake-tier cache_target_full_ratio 1.1 + expect_false ceph osd pool set fake-tier cache_min_flush_age 123 + expect_false ceph osd pool get fake-tier cache_min_flush_age + expect_false ceph osd pool set fake-tier cache_min_evict_age 234 + expect_false ceph osd pool get fake-tier cache_min_evict_age + + ceph osd tier remove rbd real-tier + ceph osd pool delete real-tier real-tier --yes-i-really-really-mean-it + ceph osd pool delete fake-tier fake-tier --yes-i-really-really-mean-it +} + +function test_mon_osd_erasure_code() +{ + + ceph osd erasure-code-profile set fooprofile a=b c=d + ceph osd erasure-code-profile set fooprofile a=b c=d + expect_false ceph osd erasure-code-profile set fooprofile a=b c=d e=f + ceph osd erasure-code-profile set fooprofile a=b c=d e=f --force + ceph osd erasure-code-profile set fooprofile a=b c=d e=f + expect_false ceph osd erasure-code-profile set fooprofile a=b c=d e=f g=h + # ruleset-foo will work for luminous only + ceph osd erasure-code-profile set barprofile ruleset-failure-domain=host + ceph osd erasure-code-profile set barprofile crush-failure-domain=host + # clean up + ceph osd erasure-code-profile rm fooprofile + ceph osd erasure-code-profile rm barprofile +} + +function test_mon_osd_misc() +{ + set +e + + # expect error about missing 'pool' argument + ceph osd map 2>$TMPFILE; check_response 'pool' $? 22 + + # expect error about unused argument foo + ceph osd ls foo 2>$TMPFILE; check_response 'unused' $? 22 + + # expect "not in range" for invalid full ratio + ceph pg set_full_ratio 95 2>$TMPFILE; check_response 'not in range' $? 22 + + # expect "not in range" for invalid overload percentage + ceph osd reweight-by-utilization 80 2>$TMPFILE; check_response 'higher than 100' $? 22 + + set -e + + ceph osd reweight-by-utilization 110 + ceph osd reweight-by-utilization 110 .5 + expect_false ceph osd reweight-by-utilization 110 0 + expect_false ceph osd reweight-by-utilization 110 -0.1 + ceph osd test-reweight-by-utilization 110 .5 --no-increasing + ceph osd test-reweight-by-utilization 110 .5 4 --no-increasing + expect_false ceph osd test-reweight-by-utilization 110 .5 0 --no-increasing + expect_false ceph osd test-reweight-by-utilization 110 .5 -10 --no-increasing + ceph osd reweight-by-pg 110 + ceph osd test-reweight-by-pg 110 .5 + ceph osd reweight-by-pg 110 rbd + ceph osd reweight-by-pg 110 .5 rbd + expect_false ceph osd reweight-by-pg 110 boguspoolasdfasdfasdf +} + +function test_mon_heap_profiler() +{ + do_test=1 + set +e + # expect 'heap' commands to be correctly parsed + ceph heap stats 2>$TMPFILE + if [[ $? -eq 22 && `grep 'tcmalloc not enabled' $TMPFILE` ]]; then + echo "tcmalloc not enabled; skip heap profiler test" + do_test=0 + fi + set -e + + [[ $do_test -eq 0 ]] && return 0 + + ceph heap start_profiler + ceph heap dump + ceph heap stop_profiler + ceph heap release +} + +function test_admin_heap_profiler() +{ + do_test=1 + set +e + # expect 'heap' commands to be correctly parsed + ceph heap stats 2>$TMPFILE + if [[ $? -eq 22 && `grep 'tcmalloc not enabled' $TMPFILE` ]]; then + echo "tcmalloc not enabled; skip heap profiler test" + do_test=0 + fi + set -e + + [[ $do_test -eq 0 ]] && return 0 + + local admin_socket=$(get_admin_socket osd.0) + + $SUDO ceph --admin-daemon $admin_socket heap start_profiler + $SUDO ceph --admin-daemon $admin_socket heap dump + $SUDO ceph --admin-daemon $admin_socket heap stop_profiler + $SUDO ceph --admin-daemon $admin_socket heap release +} + +function test_osd_bench() +{ + # test osd bench limits + # As we should not rely on defaults (as they may change over time), + # lets inject some values and perform some simple tests + # max iops: 10 # 100 IOPS + # max throughput: 10485760 # 10MB/s + # max block size: 2097152 # 2MB + # duration: 10 # 10 seconds + + local args="\ + --osd-bench-duration 10 \ + --osd-bench-max-block-size 2097152 \ + --osd-bench-large-size-max-throughput 10485760 \ + --osd-bench-small-size-max-iops 10" + ceph tell osd.0 injectargs ${args## } + + # anything with a bs larger than 2097152 must fail + expect_false ceph tell osd.0 bench 1 2097153 + # but using 'osd_bench_max_bs' must succeed + ceph tell osd.0 bench 1 2097152 + + # we assume 1MB as a large bs; anything lower is a small bs + # for a 4096 bytes bs, for 10 seconds, we are limited by IOPS + # max count: 409600 (bytes) + + # more than max count must not be allowed + expect_false ceph tell osd.0 bench 409601 4096 + # but 409600 must be succeed + ceph tell osd.0 bench 409600 4096 + + # for a large bs, we are limited by throughput. + # for a 2MB block size for 10 seconds, assuming 10MB/s throughput, + # the max count will be (10MB * 10s) = 100MB + # max count: 104857600 (bytes) + + # more than max count must not be allowed + expect_false ceph tell osd.0 bench 104857601 2097152 + # up to max count must be allowed + ceph tell osd.0 bench 104857600 2097152 +} + +function test_osd_negative_filestore_merge_threshold() +{ + $SUDO ceph daemon osd.0 config set filestore_merge_threshold -1 + expect_config_value "osd.0" "filestore_merge_threshold" -1 +} + +function test_mon_tell() +{ + ceph tell mon.a version + ceph tell mon.b version + expect_false ceph tell mon.foo version + + sleep 1 + + ceph_watch_start debug audit + ceph tell mon.a version + ceph_watch_wait 'mon.a \[DBG\] from.*cmd=\[{"prefix": "version"}\]: dispatch' + + ceph_watch_start debug audit + ceph tell mon.b version + ceph_watch_wait 'mon.b \[DBG\] from.*cmd=\[{"prefix": "version"}\]: dispatch' +} + +function test_mon_ping() +{ + ceph ping mon.a + ceph ping mon.b + expect_false ceph ping mon.foo + + ceph ping mon.\* +} + +function test_mon_deprecated_commands() +{ + # current DEPRECATED commands are: + # ceph compact + # ceph scrub + # ceph sync force + # + # Testing should be accomplished by setting + # 'mon_debug_deprecated_as_obsolete = true' and expecting ENOTSUP for + # each one of these commands. + + ceph tell mon.a injectargs '--mon-debug-deprecated-as-obsolete' + expect_false ceph tell mon.a compact 2> $TMPFILE + check_response "\(EOPNOTSUPP\|ENOTSUP\): command is obsolete" + + expect_false ceph tell mon.a scrub 2> $TMPFILE + check_response "\(EOPNOTSUPP\|ENOTSUP\): command is obsolete" + + expect_false ceph tell mon.a sync force 2> $TMPFILE + check_response "\(EOPNOTSUPP\|ENOTSUP\): command is obsolete" + + ceph tell mon.a injectargs '--no-mon-debug-deprecated-as-obsolete' +} + +function test_mon_cephdf_commands() +{ + # ceph df detail: + # pool section: + # RAW USED The near raw used per pool in raw total + + ceph osd pool create cephdf_for_test 32 32 replicated + ceph osd pool application enable cephdf_for_test rados + ceph osd pool set cephdf_for_test size 2 + + dd if=/dev/zero of=./cephdf_for_test bs=4k count=1 + rados put cephdf_for_test cephdf_for_test -p cephdf_for_test + + #wait for update + for i in `seq 1 10`; do + rados -p cephdf_for_test ls - | grep -q cephdf_for_test && break + sleep 1 + done + # "rados ls" goes straight to osd, but "ceph df" is served by mon. so we need + # to sync mon with osd + flush_pg_stats + local jq_filter='.pools | .[] | select(.name == "cephdf_for_test") | .stats' + cal_raw_used_size=`ceph df detail --format=json | jq "$jq_filter.raw_bytes_used"` + raw_used_size=`ceph df detail --format=json | jq "$jq_filter.bytes_used * 2"` + + ceph osd pool delete cephdf_for_test cephdf_for_test --yes-i-really-really-mean-it + rm ./cephdf_for_test + + expect_false test $cal_raw_used_size != $raw_used_size +} + +function test_mon_pool_application() +{ + ceph osd pool create app_for_test 10 + + ceph osd pool application enable app_for_test rbd + expect_false ceph osd pool application enable app_for_test rgw + ceph osd pool application enable app_for_test rgw --yes-i-really-mean-it + ceph osd pool ls detail | grep "application rbd,rgw" + ceph osd pool ls detail --format=json | grep '"application_metadata":{"rbd":{},"rgw":{}}' + + expect_false ceph osd pool application set app_for_test cephfs key value + ceph osd pool application set app_for_test rbd key1 value1 + ceph osd pool application set app_for_test rbd key2 value2 + ceph osd pool application set app_for_test rgw key1 value1 + ceph osd pool application get app_for_test rbd key1 | grep 'value1' + ceph osd pool application get app_for_test rbd key2 | grep 'value2' + ceph osd pool application get app_for_test rgw key1 | grep 'value1' + + ceph osd pool ls detail --format=json | grep '"application_metadata":{"rbd":{"key1":"value1","key2":"value2"},"rgw":{"key1":"value1"}}' + + ceph osd pool application rm app_for_test rgw key1 + ceph osd pool ls detail --format=json | grep '"application_metadata":{"rbd":{"key1":"value1","key2":"value2"},"rgw":{}}' + ceph osd pool application rm app_for_test rbd key2 + ceph osd pool ls detail --format=json | grep '"application_metadata":{"rbd":{"key1":"value1"},"rgw":{}}' + ceph osd pool application rm app_for_test rbd key1 + ceph osd pool ls detail --format=json | grep '"application_metadata":{"rbd":{},"rgw":{}}' + ceph osd pool application rm app_for_test rbd key1 # should be idempotent + + expect_false ceph osd pool application disable app_for_test rgw + ceph osd pool application disable app_for_test rgw --yes-i-really-mean-it + ceph osd pool application disable app_for_test rgw --yes-i-really-mean-it # should be idempotent + ceph osd pool ls detail | grep "application rbd" + ceph osd pool ls detail --format=json | grep '"application_metadata":{"rbd":{}}' + + ceph osd pool application disable app_for_test rgw --yes-i-really-mean-it + ceph osd pool ls detail | grep -v "application " + ceph osd pool ls detail --format=json | grep '"application_metadata":{}' + + ceph osd pool rm app_for_test app_for_test --yes-i-really-really-mean-it +} + +function test_mon_tell_help_command() +{ + ceph tell mon.a help + + # wrong target + expect_false ceph tell mon.zzz help +} + +function test_mon_stdin_stdout() +{ + echo foo | ceph config-key set test_key -i - + ceph config-key get test_key -o - | grep -c foo | grep -q 1 +} + +function test_osd_tell_help_command() +{ + ceph tell osd.1 help + expect_false ceph tell osd.100 help +} + +function test_osd_compact() +{ + ceph tell osd.1 compact + $SUDO ceph daemon osd.1 compact +} + +function test_mds_tell_help_command() +{ + local FS_NAME=cephfs + if ! mds_exists ; then + echo "Skipping test, no MDS found" + return + fi + + remove_all_fs + ceph osd pool create fs_data 10 + ceph osd pool create fs_metadata 10 + ceph fs new $FS_NAME fs_metadata fs_data + wait_mds_active $FS_NAME + + + ceph tell mds.a help + expect_false ceph tell mds.z help + + remove_all_fs + ceph osd pool delete fs_data fs_data --yes-i-really-really-mean-it + ceph osd pool delete fs_metadata fs_metadata --yes-i-really-really-mean-it +} + +function test_mgr_tell() +{ + ceph tell mgr help + #ceph tell mgr fs status # see http://tracker.ceph.com/issues/20761 + ceph tell mgr osd status +} + +# +# New tests should be added to the TESTS array below +# +# Individual tests may be run using the '-t ' argument +# The user can specify '-t ' as many times as she wants +# +# Tests will be run in order presented in the TESTS array, or in +# the order specified by the '-t ' options. +# +# '-l' will list all the available test names +# '-h' will show usage +# +# The test maintains backward compatibility: not specifying arguments +# will run all tests following the order they appear in the TESTS array. +# + +set +x +MON_TESTS+=" mon_injectargs" +MON_TESTS+=" mon_injectargs_SI" +for i in `seq 9`; do + MON_TESTS+=" tiering_$i"; +done +MON_TESTS+=" auth" +MON_TESTS+=" auth_profiles" +MON_TESTS+=" mon_misc" +MON_TESTS+=" mon_mon" +MON_TESTS+=" mon_osd" +MON_TESTS+=" mon_config_key" +MON_TESTS+=" mon_crush" +MON_TESTS+=" mon_osd_create_destroy" +MON_TESTS+=" mon_osd_pool" +MON_TESTS+=" mon_osd_pool_quota" +MON_TESTS+=" mon_pg" +MON_TESTS+=" mon_osd_pool_set" +MON_TESTS+=" mon_osd_tiered_pool_set" +MON_TESTS+=" mon_osd_erasure_code" +MON_TESTS+=" mon_osd_misc" +MON_TESTS+=" mon_heap_profiler" +MON_TESTS+=" mon_tell" +MON_TESTS+=" mon_ping" +MON_TESTS+=" mon_deprecated_commands" +MON_TESTS+=" mon_caps" +MON_TESTS+=" mon_cephdf_commands" +MON_TESTS+=" mon_tell_help_command" +MON_TESTS+=" mon_stdin_stdout" + +OSD_TESTS+=" osd_bench" +OSD_TESTS+=" osd_negative_filestore_merge_threshold" +OSD_TESTS+=" tiering_agent" +OSD_TESTS+=" admin_heap_profiler" +OSD_TESTS+=" osd_tell_help_command" +OSD_TESTS+=" osd_compact" + +MDS_TESTS+=" mds_tell" +MDS_TESTS+=" mon_mds" +MDS_TESTS+=" mon_mds_metadata" +MDS_TESTS+=" mds_tell_help_command" + +MGR_TESTS+=" mgr_tell" + +TESTS+=$MON_TESTS +TESTS+=$OSD_TESTS +TESTS+=$MDS_TESTS +TESTS+=$MGR_TESTS + +# +# "main" follows +# + +function list_tests() +{ + echo "AVAILABLE TESTS" + for i in $TESTS; do + echo " $i" + done +} + +function usage() +{ + echo "usage: $0 [-h|-l|-t [-t ...]]" +} + +tests_to_run=() + +sanity_check=true + +while [[ $# -gt 0 ]]; do + opt=$1 + + case "$opt" in + "-l" ) + do_list=1 + ;; + "--asok-does-not-need-root" ) + SUDO="" + ;; + "--no-sanity-check" ) + sanity_check=false + ;; + "--test-mon" ) + tests_to_run+="$MON_TESTS" + ;; + "--test-osd" ) + tests_to_run+="$OSD_TESTS" + ;; + "--test-mds" ) + tests_to_run+="$MDS_TESTS" + ;; + "--test-mgr" ) + tests_to_run+="$MGR_TESTS" + ;; + "-t" ) + shift + if [[ -z "$1" ]]; then + echo "missing argument to '-t'" + usage ; + exit 1 + fi + tests_to_run+=" $1" + ;; + "-h" ) + usage ; + exit 0 + ;; + esac + shift +done + +if [[ $do_list -eq 1 ]]; then + list_tests ; + exit 0 +fi + +ceph osd pool create rbd 10 + +if test -z "$tests_to_run" ; then + tests_to_run="$TESTS" +fi + +if $sanity_check ; then + wait_no_osd_down +fi +for i in $tests_to_run; do + if $sanity_check ; then + check_no_osd_down + fi + set -x + test_${i} + set +x +done +if $sanity_check ; then + check_no_osd_down +fi + +set -x + +echo OK