3 # Copyright (C) 2013,2014 Cloudwatt <libre.licensing@cloudwatt.com>
4 # Copyright (C) 2014,2015 Red Hat <contact@redhat.com>
5 # Copyright (C) 2014 Federico Gimenez <fgimenez@coit.es>
7 # Author: Loic Dachary <loic@dachary.org>
8 # Author: Federico Gimenez <fgimenez@coit.es>
10 # This program is free software; you can redistribute it and/or modify
11 # it under the terms of the GNU Library Public License as published by
12 # the Free Software Foundation; either version 2, or (at your option)
15 # This program is distributed in the hope that it will be useful,
16 # but WITHOUT ANY WARRANTY; without even the implied warranty of
17 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18 # GNU Library Public License for more details.
21 WAIT_FOR_CLEAN_TIMEOUT
=90
24 TMPDIR
=${TMPDIR:-/tmp}
25 CEPH_BUILD_VIRTUALENV
=${TMPDIR}
26 TESTDIR
=${TESTDIR:-${TMPDIR}}
28 if type xmlstarlet
> /dev
/null
2>&1; then
30 elif type xml
> /dev
/null
2>&1; then
33 echo "Missing xmlstarlet binary!"
37 if [ `uname` = FreeBSD
]; then
41 KERNCORE
="kern.corefile"
45 termwidth
=$
(stty
-a |
head -1 |
sed -e 's/.*columns \([0-9]*\).*/\1/')
46 if [ -n "$termwidth" -a "$termwidth" != "0" ]; then
47 termwidth
="-W ${termwidth}"
49 DIFFCOLOPTS
="-y $termwidth"
50 KERNCORE
="kernel.core_pattern"
55 #! @file ceph-helpers.sh
56 # @brief Toolbox to manage Ceph cluster dedicated to testing
60 # ~~~~~~~~~~~~~~~~{.sh}
61 # source ceph-helpers.sh
64 # # cleanup leftovers and reset mydir
66 # # create a cluster with one monitor and three osds
71 # # put and get an object
72 # rados --pool rbd put GROUP /etc/group
73 # rados --pool rbd get GROUP /tmp/GROUP
74 # # stop the cluster and cleanup the directory
79 # The focus is on simplicity and efficiency, in the context of
80 # functional tests. The output is intentionally very verbose
81 # and functions return as soon as an error is found. The caller
82 # is also expected to abort on the first error so that debugging
83 # can be done by looking at the end of the output.
85 # Each function is documented, implemented and tested independently.
86 # When modifying a helper, the test and the documentation are
87 # expected to be updated and it is easier of they are collocated. A
88 # test for a given function can be run with
90 # ~~~~~~~~~~~~~~~~{.sh}
91 # ceph-helpers.sh TESTS test_get_osds
94 # and all the tests (i.e. all functions matching test_*) are run
97 # ~~~~~~~~~~~~~~~~{.sh}
98 # ceph-helpers.sh TESTS
101 # A test function takes a single argument : the directory dedicated
102 # to the tests. It is expected to not create any file outside of this
103 # directory and remove it entirely when it completes successfully.
107 function get_asok_dir
() {
108 if [ -n "$CEPH_ASOK_DIR" ]; then
109 echo "$CEPH_ASOK_DIR"
111 echo ${TMPDIR:-/tmp}/ceph-asok.$$
115 function get_asok_path
() {
117 if [ -n "$name" ]; then
118 echo $
(get_asok_dir
)/ceph-
$name.asok
120 echo $
(get_asok_dir
)/\
$cluster-\
$name.asok
124 # Cleanup any leftovers found in **dir** via **teardown**
125 # and reset **dir** as an empty environment.
127 # @param dir path name of the environment
128 # @return 0 on success, 1 on error
132 teardown
$dir ||
return 1
134 mkdir
-p $
(get_asok_dir
)
135 if [ $
(ulimit -n) -le 1024 ]; then
136 ulimit -n 4096 ||
return 1
138 if [ -z "$LOCALRUN" ]; then
139 trap "teardown $dir 1" TERM HUP INT
143 function test_setup
() {
145 setup
$dir ||
return 1
146 test -d $dir ||
return 1
147 setup
$dir ||
return 1
148 test -d $dir ||
return 1
152 #######################################################################
155 # Kill all daemons for which a .pid file exists in **dir** and remove
156 # **dir**. If the file system in which **dir** is btrfs, delete all
157 # subvolumes that relate to it.
159 # @param dir path name of the environment
160 # @param dumplogs pass "1" to dump logs otherwise it will only if cores found
161 # @return 0 on success, 1 on error
163 function teardown
() {
166 kill_daemons
$dir KILL
167 if [ `uname` != FreeBSD
] \
168 && [ $
(stat
-f -c '%T' .
) == "btrfs" ]; then
169 __teardown_btrfs
$dir
172 local pattern
="$(sysctl -n $KERNCORE)"
173 # See if we have apport core handling
174 if [ "${pattern:0:1}" = "|" ]; then
175 # TODO: Where can we get the dumps?
176 # Not sure where the dumps really are so this will look in the CWD
179 # Local we start with core and teuthology ends with core
180 if ls $
(dirname "$pattern") |
grep -q '^core\|core$' ; then
182 if [ -n "$LOCALRUN" ]; then
183 mkdir
/tmp
/cores.$$
2> /dev
/null || true
184 for i
in $
(ls $
(dirname $
(sysctl
-n $KERNCORE)) |
grep '^core\|core$'); do
189 if [ "$cores" = "yes" -o "$dumplogs" = "1" ]; then
190 if [ -n "$LOCALRUN" ]; then
193 # Move logs to where Teuthology will archive it
194 mkdir
-p $TESTDIR/archive
/log
195 mv $dir/*.log
$TESTDIR/archive
/log
199 rm -rf $
(get_asok_dir
)
200 if [ "$cores" = "yes" ]; then
201 echo "ERROR: Failure due to cores found"
202 if [ -n "$LOCALRUN" ]; then
203 echo "Find saved core files in /tmp/cores.$$"
210 function __teardown_btrfs
() {
211 local btrfs_base_dir
=$1
212 local btrfs_root
=$
(df
-P . |
tail -1 |
$AWK '{print $NF}')
213 local btrfs_dirs
=$
(cd $btrfs_base_dir; sudo btrfs subvolume list
-t . |
$AWK '/^[0-9]/ {print $4}' |
grep "$btrfs_base_dir/$btrfs_dir")
214 for subvolume
in $btrfs_dirs; do
215 sudo btrfs subvolume delete
$btrfs_root/$subvolume
219 function test_teardown
() {
221 setup
$dir ||
return 1
222 teardown
$dir ||
return 1
223 ! test -d $dir ||
return 1
226 #######################################################################
229 # Sends a signal to a single daemon.
230 # This is a helper function for kill_daemons
232 # After the daemon is sent **signal**, its actual termination
233 # will be verified by sending it signal 0. If the daemon is
234 # still alive, kill_daemon will pause for a few seconds and
235 # try again. This will repeat for a fixed number of times
236 # before kill_daemon returns on failure. The list of
237 # sleep intervals can be specified as **delays** and defaults
240 # 0.1 0.2 1 1 1 2 3 5 5 5 10 10 20 60 60 60 120
242 # This sequence is designed to run first a very short sleep time (0.1)
243 # if the machine is fast enough and the daemon terminates in a fraction of a
244 # second. The increasing sleep numbers should give plenty of time for
245 # the daemon to die even on the slowest running machine. If a daemon
246 # takes more than a few minutes to stop (the sum of all sleep times),
247 # there probably is no point in waiting more and a number of things
248 # are likely to go wrong anyway: better give up and return on error.
250 # @param pid the process id to send a signal
251 # @param send_signal the signal to send
252 # @param delays sequence of sleep times before failure
254 function kill_daemon
() {
257 local delays
=${3:-0.1 0.2 1 1 1 2 3 5 5 5 10 10 20 60 60 60 120}
259 # In order to try after the last large sleep add 0 at the end so we check
260 # one last time before dropping out of the loop
261 for try
in $delays 0 ; do
262 if kill -$send_signal $pid 2> /dev
/null
; then
274 function test_kill_daemon
() {
276 setup
$dir ||
return 1
277 run_mon
$dir a
--osd_pool_default_size=1 ||
return 1
278 run_mgr
$dir x ||
return 1
279 run_osd
$dir 0 ||
return 1
282 for pidfile
in $
(find $dir 2>/dev
/null |
grep $name_prefix'[^/]*\.pid') ; do
284 # sending signal 0 won't kill the daemon
285 # waiting just for one second instead of the default schedule
286 # allows us to quickly verify what happens when kill fails
287 # to stop the daemon (i.e. it must return false)
289 ! kill_daemon
$pidfile 0 1 ||
return 1
291 # killing just the osd and verify the mon still is responsive
293 kill_daemon
$pidfile TERM ||
return 1
297 for pidfile
in $
(find $dir 2>/dev
/null |
grep $name_prefix'[^/]*\.pid') ; do
301 kill_daemon
$pidfile TERM ||
return 1
305 for pidfile
in $
(find $dir 2>/dev
/null |
grep $name_prefix'[^/]*\.pid') ; do
307 # kill the mon and verify it cannot be reached
309 kill_daemon
$pidfile TERM ||
return 1
310 ! timeout
5 ceph status ||
return 1
313 teardown
$dir ||
return 1
317 # Kill all daemons for which a .pid file exists in **dir**. Each
318 # daemon is sent a **signal** and kill_daemons waits for it to exit
319 # during a few minutes. By default all daemons are killed. If a
320 # **name_prefix** is provided, only the daemons for which a pid
321 # file is found matching the prefix are killed. See run_osd and
322 # run_mon for more information about the name conventions for
325 # Send TERM to all daemons : kill_daemons $dir
326 # Send KILL to all daemons : kill_daemons $dir KILL
327 # Send KILL to all osds : kill_daemons $dir KILL osd
328 # Send KILL to osd 1 : kill_daemons $dir KILL osd.1
330 # If a daemon is sent the TERM signal and does not terminate
331 # within a few minutes, it will still be running even after
332 # kill_daemons returns.
334 # If all daemons are kill successfully the function returns 0
335 # if at least one daemon remains, this is treated as an
336 # error and the function return 1.
338 # @param dir path name of the environment
339 # @param signal name of the first signal (defaults to TERM)
340 # @param name_prefix only kill match daemons (defaults to all)
341 # @param delays sequence of sleep times before failure
342 # @return 0 on success, 1 on error
344 function kill_daemons
() {
345 local trace
=$
(shopt -q -o xtrace
&& echo true ||
echo false
)
346 $trace && shopt -u -o xtrace
348 local signal
=${2:-TERM}
349 local name_prefix
=$3 # optional, osd, mon, osd.1
350 local delays
=$4 #optional timing
354 for pidfile
in $
(find $dir 2>/dev
/null |
grep $name_prefix'[^/]*\.pid') ; do
355 run_in_background pids kill_daemon
$pidfile $signal $delays
361 $trace && shopt -s -o xtrace
365 function test_kill_daemons
() {
367 setup
$dir ||
return 1
368 run_mon
$dir a
--osd_pool_default_size=1 ||
return 1
369 run_mgr
$dir x ||
return 1
370 run_osd
$dir 0 ||
return 1
372 # sending signal 0 won't kill the daemon
373 # waiting just for one second instead of the default schedule
374 # allows us to quickly verify what happens when kill fails
375 # to stop the daemon (i.e. it must return false)
377 ! kill_daemons
$dir 0 osd
1 ||
return 1
379 # killing just the osd and verify the mon still is responsive
381 kill_daemons
$dir TERM osd ||
return 1
385 kill_daemons
$dir TERM mgr ||
return 1
387 # kill the mon and verify it cannot be reached
389 kill_daemons
$dir TERM ||
return 1
390 ! timeout
5 ceph status ||
return 1
391 teardown
$dir ||
return 1
395 # return a random TCP port which is not used yet
397 # please note, there could be racing if we use this function for
398 # a free port, and then try to bind on this port.
400 function get_unused_port
() {
402 python3
-c "import socket; s=socket.socket(); s.bind(('$ip', 0)); print(s.getsockname()[1]); s.close()"
405 #######################################################################
408 # Run a monitor by the name mon.**id** with data in **dir**/**id**.
409 # The logs can be found in **dir**/mon.**id**.log and the pid file
410 # is **dir**/mon.**id**.pid and the admin socket is
411 # **dir**/**id**/ceph-mon.**id**.asok.
413 # The remaining arguments are passed verbatim to ceph-mon --mkfs
414 # and the ceph-mon daemon.
416 # Two mandatory arguments must be provided: --fsid and --mon-host
417 # Instead of adding them to every call to run_mon, they can be
418 # set in the CEPH_ARGS environment variable to be read implicitly
419 # by every ceph command.
421 # The CEPH_CONF variable is expected to be set to /dev/null to
422 # only rely on arguments for configuration.
426 # CEPH_ARGS="--fsid=$(uuidgen) "
427 # CEPH_ARGS+="--mon-host=127.0.0.1:7018 "
428 # run_mon $dir a # spawn a mon and bind port 7018
429 # run_mon $dir a --debug-filestore=20 # spawn with filestore debugging
431 # If mon_initial_members is not set, the default rbd pool is deleted
432 # and replaced with a replicated pool with less placement groups to
433 # speed up initialization. If mon_initial_members is set, no attempt
434 # is made to recreate the rbd pool because it would hang forever,
435 # waiting for other mons to join.
437 # A **dir**/ceph.conf file is created but not meant to be used by any
438 # function. It is convenient for debugging a failure with:
440 # ceph --conf **dir**/ceph.conf -s
442 # @param dir path name of the environment
443 # @param id mon identifier
444 # @param ... can be any option valid for ceph-mon
445 # @return 0 on success, 1 on error
463 --osd-failsafe-full-ratio=.99 \
464 --mon-osd-full-ratio=.99 \
465 --mon-data-avail-crit=1 \
466 --mon-data-avail-warn=5 \
467 --paxos-propose-interval=0.1 \
468 --osd-crush-chooseleaf-type=0 \
475 --log-file=$dir/\
$name.log \
476 --admin-socket=$
(get_asok_path
) \
477 --mon-cluster-log-file=$dir/log \
479 --pid-file=$dir/\
$name.pid \
480 --mon-allow-pool-delete \
481 --osd-pool-default-pg-autoscale-mode off \
482 --mon-osd-backfillfull-ratio .99 \
485 cat > $dir/ceph.conf
<<EOF
487 fsid = $(get_config mon $id fsid)
488 mon host = $(get_config mon $id mon_host)
492 function test_run_mon
() {
495 setup
$dir ||
return 1
497 run_mon
$dir a
--mon-initial-members=a ||
return 1
498 ceph mon dump |
grep "mon.a" ||
return 1
499 kill_daemons
$dir ||
return 1
501 run_mon
$dir a
--osd_pool_default_size=3 ||
return 1
502 run_osd
$dir 0 ||
return 1
503 run_osd
$dir 1 ||
return 1
504 run_osd
$dir 2 ||
return 1
505 create_rbd_pool ||
return 1
506 ceph osd dump |
grep "pool 1 'rbd'" ||
return 1
507 local size
=$
(CEPH_ARGS
='' ceph
--format=json daemon $
(get_asok_path mon.a
) \
508 config get osd_pool_default_size
)
509 test "$size" = '{"osd_pool_default_size":"3"}' ||
return 1
511 ! CEPH_ARGS
='' ceph status ||
return 1
512 CEPH_ARGS
='' ceph
--conf $dir/ceph.conf status ||
return 1
514 kill_daemons
$dir ||
return 1
516 run_mon
$dir a
--osd_pool_default_size=1 ||
return 1
517 local size
=$
(CEPH_ARGS
='' ceph
--format=json daemon $
(get_asok_path mon.a
) \
518 config get osd_pool_default_size
)
519 test "$size" = '{"osd_pool_default_size":"1"}' ||
return 1
520 kill_daemons
$dir ||
return 1
522 CEPH_ARGS
="$CEPH_ARGS --osd_pool_default_size=2" \
523 run_mon
$dir a ||
return 1
524 local size
=$
(CEPH_ARGS
='' ceph
--format=json daemon $
(get_asok_path mon.a
) \
525 config get osd_pool_default_size
)
526 test "$size" = '{"osd_pool_default_size":"2"}' ||
return 1
527 kill_daemons
$dir ||
return 1
529 teardown
$dir ||
return 1
532 function create_rbd_pool
() {
533 ceph osd pool delete rbd rbd
--yes-i-really-really-mean-it ||
return 1
534 create_pool rbd
$PG_NUM ||
return 1
538 function create_pool
() {
539 ceph osd pool create
"$@"
543 function delete_pool
() {
545 ceph osd pool delete
$poolname $poolname --yes-i-really-really-mean-it
548 #######################################################################
557 ceph config
set mgr mgr
/devicehealth
/enable_monitoring off
--force
561 --osd-failsafe-full-ratio=.99 \
563 --debug-objecter 20 \
568 --log-file=$dir/\
$name.log \
569 --admin-socket=$
(get_asok_path
) \
571 --pid-file=$dir/\
$name.pid \
572 --mgr-module-path=$
(realpath
${CEPH_ROOT}/src
/pybind
/mgr
) \
587 --debug-objecter 20 \
591 --log-file=$dir/\
$name.log \
592 --admin-socket=$
(get_asok_path
) \
594 --pid-file=$dir/\
$name.pid \
598 #######################################################################
601 # Create (prepare) and run (activate) an osd by the name osd.**id**
602 # with data in **dir**/**id**. The logs can be found in
603 # **dir**/osd.**id**.log, the pid file is **dir**/osd.**id**.pid and
604 # the admin socket is **dir**/**id**/ceph-osd.**id**.asok.
606 # The remaining arguments are passed verbatim to ceph-osd.
608 # Two mandatory arguments must be provided: --fsid and --mon-host
609 # Instead of adding them to every call to run_osd, they can be
610 # set in the CEPH_ARGS environment variable to be read implicitly
611 # by every ceph command.
613 # The CEPH_CONF variable is expected to be set to /dev/null to
614 # only rely on arguments for configuration.
616 # The run_osd function creates the OSD data directory on the **dir**/**id**
617 # directory and relies on the activate_osd function to run the daemon.
621 # CEPH_ARGS="--fsid=$(uuidgen) "
622 # CEPH_ARGS+="--mon-host=127.0.0.1:7018 "
623 # run_osd $dir 0 # prepare and activate an osd using the monitor listening on 7018
625 # @param dir path name of the environment
626 # @param id osd identifier
627 # @param ... can be any option valid for ceph-osd
628 # @return 0 on success, 1 on error
635 local osd_data
=$dir/$id
637 local ceph_args
="$CEPH_ARGS"
638 ceph_args
+=" --osd-failsafe-full-ratio=.99"
639 ceph_args
+=" --osd-journal-size=100"
640 ceph_args
+=" --osd-scrub-load-threshold=2000"
641 ceph_args
+=" --osd-data=$osd_data"
642 ceph_args
+=" --osd-journal=${osd_data}/journal"
643 ceph_args
+=" --chdir="
644 ceph_args
+=$EXTRA_OPTS
645 ceph_args
+=" --run-dir=$dir"
646 ceph_args
+=" --admin-socket=$(get_asok_path)"
647 ceph_args
+=" --debug-osd=20"
648 ceph_args
+=" --debug-ms=1"
649 ceph_args
+=" --debug-monc=20"
650 ceph_args
+=" --log-file=$dir/\$name.log"
651 ceph_args
+=" --pid-file=$dir/\$name.pid"
652 ceph_args
+=" --osd-max-object-name-len=460"
653 ceph_args
+=" --osd-max-object-namespace-len=64"
654 ceph_args
+=" --enable-experimental-unrecoverable-data-corrupting-features=*"
660 echo "add osd$id $uuid"
661 OSD_SECRET
=$
(ceph-authtool
--gen-print-key)
662 echo "{\"cephx_secret\": \"$OSD_SECRET\"}" > $osd_data/new.json
663 ceph osd new
$uuid -i $osd_data/new.json
664 rm $osd_data/new.json
665 ceph-osd
-i $id $ceph_args --mkfs --key $OSD_SECRET --osd-uuid $uuid
667 local key_fn
=$osd_data/keyring
672 echo adding osd
$id key to auth repository
673 ceph
-i "$key_fn" auth add osd.
$id osd
"allow *" mon
"allow profile osd" mgr
"allow profile osd"
675 ceph-osd
-i $id $ceph_args &
677 # If noup is set, then can't wait for this osd
678 if ceph osd dump
--format=json | jq
'.flags_set[]' |
grep -q '"noup"' ; then
681 wait_for_osd up
$id ||
return 1
685 function run_osd_filestore
() {
690 local osd_data
=$dir/$id
692 local ceph_args
="$CEPH_ARGS"
693 ceph_args
+=" --osd-failsafe-full-ratio=.99"
694 ceph_args
+=" --osd-journal-size=100"
695 ceph_args
+=" --osd-scrub-load-threshold=2000"
696 ceph_args
+=" --osd-data=$osd_data"
697 ceph_args
+=" --osd-journal=${osd_data}/journal"
698 ceph_args
+=" --chdir="
699 ceph_args
+=$EXTRA_OPTS
700 ceph_args
+=" --run-dir=$dir"
701 ceph_args
+=" --admin-socket=$(get_asok_path)"
702 ceph_args
+=" --debug-osd=20"
703 ceph_args
+=" --debug-ms=1"
704 ceph_args
+=" --debug-monc=20"
705 ceph_args
+=" --log-file=$dir/\$name.log"
706 ceph_args
+=" --pid-file=$dir/\$name.pid"
707 ceph_args
+=" --osd-max-object-name-len=460"
708 ceph_args
+=" --osd-max-object-namespace-len=64"
709 ceph_args
+=" --enable-experimental-unrecoverable-data-corrupting-features=*"
715 echo "add osd$osd $uuid"
716 OSD_SECRET
=$
(ceph-authtool
--gen-print-key)
717 echo "{\"cephx_secret\": \"$OSD_SECRET\"}" > $osd_data/new.json
718 ceph osd new
$uuid -i $osd_data/new.json
719 rm $osd_data/new.json
720 ceph-osd
-i $id $ceph_args --mkfs --key $OSD_SECRET --osd-uuid $uuid --osd-objectstore=filestore
722 local key_fn
=$osd_data/keyring
727 echo adding osd
$id key to auth repository
728 ceph
-i "$key_fn" auth add osd.
$id osd
"allow *" mon
"allow profile osd" mgr
"allow profile osd"
730 ceph-osd
-i $id $ceph_args &
732 # If noup is set, then can't wait for this osd
733 if ceph osd dump
--format=json | jq
'.flags_set[]' |
grep -q '"noup"' ; then
736 wait_for_osd up
$id ||
return 1
741 function test_run_osd
() {
744 setup
$dir ||
return 1
746 run_mon
$dir a ||
return 1
747 run_mgr
$dir x ||
return 1
749 run_osd
$dir 0 ||
return 1
750 local backfills
=$
(CEPH_ARGS
='' ceph
--format=json daemon $
(get_asok_path osd
.0) \
751 config get osd_max_backfills
)
752 echo "$backfills" |
grep --quiet 'osd_max_backfills' ||
return 1
754 run_osd
$dir 1 --osd-max-backfills 20 ||
return 1
755 local backfills
=$
(CEPH_ARGS
='' ceph
--format=json daemon $
(get_asok_path osd
.1) \
756 config get osd_max_backfills
)
757 test "$backfills" = '{"osd_max_backfills":"20"}' ||
return 1
759 CEPH_ARGS
="$CEPH_ARGS --osd-max-backfills 30" run_osd
$dir 2 ||
return 1
760 local backfills
=$
(CEPH_ARGS
='' ceph
--format=json daemon $
(get_asok_path osd
.2) \
761 config get osd_max_backfills
)
762 test "$backfills" = '{"osd_max_backfills":"30"}' ||
return 1
764 teardown
$dir ||
return 1
767 #######################################################################
770 # Shutdown and remove all traces of the osd by the name osd.**id**.
772 # The OSD is shutdown with the TERM signal. It is then removed from
773 # the auth list, crush map, osd map etc and the files associated with
774 # it are also removed.
776 # @param dir path name of the environment
777 # @param id osd identifier
778 # @return 0 on success, 1 on error
780 function destroy_osd
() {
784 ceph osd out osd.
$id ||
return 1
785 kill_daemons
$dir TERM osd.
$id ||
return 1
786 ceph osd down osd.
$id ||
return 1
787 ceph osd purge osd.
$id --yes-i-really-mean-it ||
return 1
788 teardown
$dir/$id ||
return 1
792 function test_destroy_osd
() {
795 setup
$dir ||
return 1
796 run_mon
$dir a ||
return 1
797 run_mgr
$dir x ||
return 1
798 run_osd
$dir 0 ||
return 1
799 destroy_osd
$dir 0 ||
return 1
800 ! ceph osd dump |
grep "osd.$id " ||
return 1
801 teardown
$dir ||
return 1
804 #######################################################################
807 # Run (activate) an osd by the name osd.**id** with data in
808 # **dir**/**id**. The logs can be found in **dir**/osd.**id**.log,
809 # the pid file is **dir**/osd.**id**.pid and the admin socket is
810 # **dir**/**id**/ceph-osd.**id**.asok.
812 # The remaining arguments are passed verbatim to ceph-osd.
814 # Two mandatory arguments must be provided: --fsid and --mon-host
815 # Instead of adding them to every call to activate_osd, they can be
816 # set in the CEPH_ARGS environment variable to be read implicitly
817 # by every ceph command.
819 # The CEPH_CONF variable is expected to be set to /dev/null to
820 # only rely on arguments for configuration.
822 # The activate_osd function expects a valid OSD data directory
823 # in **dir**/**id**, either just created via run_osd or re-using
824 # one left by a previous run of ceph-osd. The ceph-osd daemon is
825 # run directly on the foreground
827 # The activate_osd function blocks until the monitor reports the osd
828 # up. If it fails to do so within $TIMEOUT seconds, activate_osd
833 # CEPH_ARGS="--fsid=$(uuidgen) "
834 # CEPH_ARGS+="--mon-host=127.0.0.1:7018 "
835 # activate_osd $dir 0 # activate an osd using the monitor listening on 7018
837 # @param dir path name of the environment
838 # @param id osd identifier
839 # @param ... can be any option valid for ceph-osd
840 # @return 0 on success, 1 on error
842 function activate_osd
() {
847 local osd_data
=$dir/$id
849 local ceph_args
="$CEPH_ARGS"
850 ceph_args
+=" --osd-failsafe-full-ratio=.99"
851 ceph_args
+=" --osd-journal-size=100"
852 ceph_args
+=" --osd-scrub-load-threshold=2000"
853 ceph_args
+=" --osd-data=$osd_data"
854 ceph_args
+=" --osd-journal=${osd_data}/journal"
855 ceph_args
+=" --chdir="
856 ceph_args
+=$EXTRA_OPTS
857 ceph_args
+=" --run-dir=$dir"
858 ceph_args
+=" --admin-socket=$(get_asok_path)"
859 ceph_args
+=" --debug-osd=20"
860 ceph_args
+=" --log-file=$dir/\$name.log"
861 ceph_args
+=" --pid-file=$dir/\$name.pid"
862 ceph_args
+=" --osd-max-object-name-len=460"
863 ceph_args
+=" --osd-max-object-namespace-len=64"
864 ceph_args
+=" --enable-experimental-unrecoverable-data-corrupting-features=*"
870 ceph-osd
-i $id $ceph_args &
872 [ "$id" = "$(cat $osd_data/whoami)" ] ||
return 1
874 # If noup is set, then can't wait for this osd
875 if ceph osd dump
--format=json | jq
'.flags_set[]' |
grep -q '"noup"' ; then
878 wait_for_osd up
$id ||
return 1
881 function test_activate_osd
() {
884 setup
$dir ||
return 1
886 run_mon
$dir a ||
return 1
887 run_mgr
$dir x ||
return 1
889 run_osd
$dir 0 ||
return 1
890 local backfills
=$
(CEPH_ARGS
='' ceph
--format=json daemon $
(get_asok_path osd
.0) \
891 config get osd_max_backfills
)
892 echo "$backfills" |
grep --quiet 'osd_max_backfills' ||
return 1
894 kill_daemons
$dir TERM osd ||
return 1
896 activate_osd
$dir 0 --osd-max-backfills 20 ||
return 1
897 local backfills
=$
(CEPH_ARGS
='' ceph
--format=json daemon $
(get_asok_path osd
.0) \
898 config get osd_max_backfills
)
899 test "$backfills" = '{"osd_max_backfills":"20"}' ||
return 1
901 teardown
$dir ||
return 1
904 #######################################################################
907 # Wait until the OSD **id** is either up or down, as specified by
908 # **state**. It fails after $TIMEOUT seconds.
910 # @param state either up or down
911 # @param id osd identifier
912 # @return 0 on success, 1 on error
914 function wait_for_osd
() {
919 for ((i
=0; i
< $TIMEOUT; i
++)); do
921 if ! ceph osd dump |
grep "osd.$id $state"; then
931 function test_wait_for_osd
() {
933 setup
$dir ||
return 1
934 run_mon
$dir a
--osd_pool_default_size=1 ||
return 1
935 run_mgr
$dir x ||
return 1
936 run_osd
$dir 0 ||
return 1
937 run_osd
$dir 1 ||
return 1
938 wait_for_osd up
0 ||
return 1
939 wait_for_osd up
1 ||
return 1
940 kill_daemons
$dir TERM osd
.0 ||
return 1
941 wait_for_osd down
0 ||
return 1
942 ( TIMEOUT
=1 ; ! wait_for_osd up
0 ) ||
return 1
943 teardown
$dir ||
return 1
946 #######################################################################
949 # Display the list of OSD ids supporting the **objectname** stored in
950 # **poolname**, as reported by ceph osd map.
952 # @param poolname an existing pool
953 # @param objectname an objectname (may or may not exist)
954 # @param STDOUT white space separated list of OSD ids
955 # @return 0 on success, 1 on error
957 function get_osds
() {
961 local osds
=$
(ceph
--format json osd map
$poolname $objectname 2>/dev
/null | \
963 # get rid of the trailing space
967 function test_get_osds
() {
970 setup
$dir ||
return 1
971 run_mon
$dir a
--osd_pool_default_size=2 ||
return 1
972 run_mgr
$dir x ||
return 1
973 run_osd
$dir 0 ||
return 1
974 run_osd
$dir 1 ||
return 1
975 create_rbd_pool ||
return 1
976 wait_for_clean ||
return 1
977 create_rbd_pool ||
return 1
978 get_osds rbd GROUP |
grep --quiet '^[0-1] [0-1]$' ||
return 1
979 teardown
$dir ||
return 1
982 #######################################################################
985 # Wait for the monitor to form quorum (optionally, of size N)
987 # @param timeout duration (lower-bound) to wait for quorum to be formed
988 # @param quorumsize size of quorum to wait for
989 # @return 0 on success, 1 on error
991 function wait_for_quorum
() {
995 if [[ -z "$timeout" ]]; then
999 if [[ -z "$quorumsize" ]]; then
1000 timeout
$timeout ceph quorum_status
--format=json
>&/dev
/null ||
return 1
1005 wait_until
=$
((`date +%s` + $timeout))
1006 while [[ $
(date +%s
) -lt $wait_until ]]; do
1007 jqfilter
='.quorum | length == '$quorumsize
1008 jqinput
="$(timeout $timeout ceph quorum_status --format=json 2>/dev/null)"
1009 res
=$
(echo $jqinput | jq
"$jqfilter")
1010 if [[ "$res" == "true" ]]; then
1018 #######################################################################
1021 # Return the PG of supporting the **objectname** stored in
1022 # **poolname**, as reported by ceph osd map.
1024 # @param poolname an existing pool
1025 # @param objectname an objectname (may or may not exist)
1026 # @param STDOUT a PG
1027 # @return 0 on success, 1 on error
1033 ceph
--format json osd map
$poolname $objectname 2>/dev
/null | jq
-r '.pgid'
1036 function test_get_pg
() {
1039 setup
$dir ||
return 1
1040 run_mon
$dir a
--osd_pool_default_size=1 ||
return 1
1041 run_mgr
$dir x ||
return 1
1042 run_osd
$dir 0 ||
return 1
1043 create_rbd_pool ||
return 1
1044 wait_for_clean ||
return 1
1045 get_pg rbd GROUP |
grep --quiet '^[0-9]\.[0-9a-f][0-9a-f]*$' ||
return 1
1046 teardown
$dir ||
return 1
1049 #######################################################################
1052 # Return the value of the **config**, obtained via the config get command
1053 # of the admin socket of **daemon**.**id**.
1055 # @param daemon mon or osd
1056 # @param id mon or osd ID
1057 # @param config the configuration variable name as found in config_opts.h
1058 # @param STDOUT the config value
1059 # @return 0 on success, 1 on error
1061 function get_config
() {
1067 ceph
--format json daemon $
(get_asok_path
$daemon.
$id) \
1068 config get
$config 2> /dev
/null | \
1072 function test_get_config
() {
1075 # override the default config using command line arg and check it
1076 setup
$dir ||
return 1
1077 run_mon
$dir a
--osd_pool_default_size=1 ||
return 1
1078 test $
(get_config mon a osd_pool_default_size
) = 1 ||
return 1
1079 run_mgr
$dir x ||
return 1
1080 run_osd
$dir 0 --osd_max_scrubs=3 ||
return 1
1081 test $
(get_config osd
0 osd_max_scrubs
) = 3 ||
return 1
1082 teardown
$dir ||
return 1
1085 #######################################################################
1088 # Set the **config** to specified **value**, via the config set command
1089 # of the admin socket of **daemon**.**id**
1091 # @param daemon mon or osd
1092 # @param id mon or osd ID
1093 # @param config the configuration variable name as found in config_opts.h
1094 # @param value the config value
1095 # @return 0 on success, 1 on error
1097 function set_config
() {
1103 test $
(env CEPH_ARGS
='' ceph
--format json daemon $
(get_asok_path
$daemon.
$id) \
1104 config
set $config $value 2> /dev
/null | \
1105 jq
'has("success")') == true
1108 function test_set_config
() {
1111 setup
$dir ||
return 1
1112 run_mon
$dir a
--osd_pool_default_size=1 ||
return 1
1113 test $
(get_config mon a ms_crc_header
) = true ||
return 1
1114 set_config mon a ms_crc_header false ||
return 1
1115 test $
(get_config mon a ms_crc_header
) = false ||
return 1
1116 set_config mon a ms_crc_header true ||
return 1
1117 test $
(get_config mon a ms_crc_header
) = true ||
return 1
1118 teardown
$dir ||
return 1
1121 #######################################################################
1124 # Return the OSD id of the primary OSD supporting the **objectname**
1125 # stored in **poolname**, as reported by ceph osd map.
1127 # @param poolname an existing pool
1128 # @param objectname an objectname (may or may not exist)
1129 # @param STDOUT the primary OSD id
1130 # @return 0 on success, 1 on error
1132 function get_primary
() {
1136 ceph
--format json osd map
$poolname $objectname 2>/dev
/null | \
1137 jq
'.acting_primary'
1140 function test_get_primary
() {
1143 setup
$dir ||
return 1
1144 run_mon
$dir a
--osd_pool_default_size=1 ||
return 1
1146 run_mgr
$dir x ||
return 1
1147 run_osd
$dir $osd ||
return 1
1148 create_rbd_pool ||
return 1
1149 wait_for_clean ||
return 1
1150 test $
(get_primary rbd GROUP
) = $osd ||
return 1
1151 teardown
$dir ||
return 1
1154 #######################################################################
1157 # Return the id of any OSD supporting the **objectname** stored in
1158 # **poolname**, as reported by ceph osd map, except the primary.
1160 # @param poolname an existing pool
1161 # @param objectname an objectname (may or may not exist)
1162 # @param STDOUT the OSD id
1163 # @return 0 on success, 1 on error
1165 function get_not_primary
() {
1169 local primary
=$
(get_primary
$poolname $objectname)
1170 ceph
--format json osd map
$poolname $objectname 2>/dev
/null | \
1171 jq
".acting | map(select (. != $primary)) | .[0]"
1174 function test_get_not_primary
() {
1177 setup
$dir ||
return 1
1178 run_mon
$dir a
--osd_pool_default_size=2 ||
return 1
1179 run_mgr
$dir x ||
return 1
1180 run_osd
$dir 0 ||
return 1
1181 run_osd
$dir 1 ||
return 1
1182 create_rbd_pool ||
return 1
1183 wait_for_clean ||
return 1
1184 local primary
=$
(get_primary rbd GROUP
)
1185 local not_primary
=$
(get_not_primary rbd GROUP
)
1186 test $not_primary != $primary ||
return 1
1187 test $not_primary = 0 -o $not_primary = 1 ||
return 1
1188 teardown
$dir ||
return 1
1191 #######################################################################
1193 function _objectstore_tool_nodown
() {
1198 local osd_data
=$dir/$id
1200 ceph-objectstore-tool \
1201 --data-path $osd_data \
1205 function _objectstore_tool_nowait
() {
1211 kill_daemons
$dir TERM osd.
$id >&2 < /dev
/null ||
return 1
1213 _objectstore_tool_nodown
$dir $id "$@" ||
return 1
1214 activate_osd
$dir $id $ceph_osd_args >&2 ||
return 1
1218 # Run ceph-objectstore-tool against the OSD **id** using the data path
1219 # **dir**. The OSD is killed with TERM prior to running
1220 # ceph-objectstore-tool because access to the data path is
1221 # exclusive. The OSD is restarted after the command completes. The
1222 # objectstore_tool returns after all PG are active+clean again.
1224 # @param dir the data path of the OSD
1225 # @param id the OSD id
1226 # @param ... arguments to ceph-objectstore-tool
1227 # @param STDIN the input of ceph-objectstore-tool
1228 # @param STDOUT the output of ceph-objectstore-tool
1229 # @return 0 on success, 1 on error
1231 # The value of $ceph_osd_args will be passed to restarted osds
1233 function objectstore_tool
() {
1239 _objectstore_tool_nowait
$dir $id "$@" ||
return 1
1243 function test_objectstore_tool
() {
1246 setup
$dir ||
return 1
1247 run_mon
$dir a
--osd_pool_default_size=1 ||
return 1
1249 run_mgr
$dir x ||
return 1
1250 run_osd
$dir $osd ||
return 1
1251 create_rbd_pool ||
return 1
1252 wait_for_clean ||
return 1
1253 rados
--pool rbd put GROUP
/etc
/group ||
return 1
1254 objectstore_tool
$dir $osd GROUP get-bytes | \
1256 ! objectstore_tool
$dir $osd NOTEXISTS get-bytes ||
return 1
1257 teardown
$dir ||
return 1
1260 #######################################################################
1263 # Predicate checking if there is an ongoing recovery in the
1264 # cluster. If any of the recovering_{keys,bytes,objects}_per_sec
1265 # counters are reported by ceph status, it means recovery is in
1268 # @return 0 if recovery in progress, 1 otherwise
1270 function get_is_making_recovery_progress
() {
1271 local recovery_progress
1272 recovery_progress
+=".recovering_keys_per_sec + "
1273 recovery_progress
+=".recovering_bytes_per_sec + "
1274 recovery_progress
+=".recovering_objects_per_sec"
1275 local progress
=$
(ceph
--format json status
2>/dev
/null | \
1276 jq
-r ".pgmap | $recovery_progress")
1277 test "$progress" != null
1280 function test_get_is_making_recovery_progress
() {
1283 setup
$dir ||
return 1
1284 run_mon
$dir a ||
return 1
1285 run_mgr
$dir x ||
return 1
1286 ! get_is_making_recovery_progress ||
return 1
1287 teardown
$dir ||
return 1
1290 #######################################################################
1293 # Return the number of active PGs in the cluster. A PG is active if
1294 # ceph pg dump pgs reports it both **active** and **clean** and that
1297 # @param STDOUT the number of active PGs
1298 # @return 0 on success, 1 on error
1300 function get_num_active_clean
() {
1302 expression
+="select(contains(\"active\") and contains(\"clean\")) | "
1303 expression
+="select(contains(\"stale\") | not)"
1304 ceph
--format json pg dump pgs
2>/dev
/null | \
1305 jq
".pg_stats | [.[] | .state | $expression] | length"
1308 function test_get_num_active_clean
() {
1311 setup
$dir ||
return 1
1312 run_mon
$dir a
--osd_pool_default_size=1 ||
return 1
1313 run_mgr
$dir x ||
return 1
1314 run_osd
$dir 0 ||
return 1
1315 create_rbd_pool ||
return 1
1316 wait_for_clean ||
return 1
1317 local num_active_clean
=$
(get_num_active_clean
)
1318 test "$num_active_clean" = $PG_NUM ||
return 1
1319 teardown
$dir ||
return 1
1323 # Return the number of active or peered PGs in the cluster. A PG matches if
1324 # ceph pg dump pgs reports it is either **active** or **peered** and that
1327 # @param STDOUT the number of active PGs
1328 # @return 0 on success, 1 on error
1330 function get_num_active_or_peered
() {
1332 expression
+="select(contains(\"active\") or contains(\"peered\")) | "
1333 expression
+="select(contains(\"stale\") | not)"
1334 ceph
--format json pg dump pgs
2>/dev
/null | \
1335 jq
".pg_stats | [.[] | .state | $expression] | length"
1338 function test_get_num_active_or_peered
() {
1341 setup
$dir ||
return 1
1342 run_mon
$dir a
--osd_pool_default_size=1 ||
return 1
1343 run_mgr
$dir x ||
return 1
1344 run_osd
$dir 0 ||
return 1
1345 create_rbd_pool ||
return 1
1346 wait_for_clean ||
return 1
1347 local num_peered
=$
(get_num_active_or_peered
)
1348 test "$num_peered" = $PG_NUM ||
return 1
1349 teardown
$dir ||
return 1
1352 #######################################################################
1355 # Return the number of PGs in the cluster, according to
1358 # @param STDOUT the number of PGs
1359 # @return 0 on success, 1 on error
1361 function get_num_pgs
() {
1362 ceph
--format json status
2>/dev
/null | jq
'.pgmap.num_pgs'
1365 function test_get_num_pgs
() {
1368 setup
$dir ||
return 1
1369 run_mon
$dir a
--osd_pool_default_size=1 ||
return 1
1370 run_mgr
$dir x ||
return 1
1371 run_osd
$dir 0 ||
return 1
1372 create_rbd_pool ||
return 1
1373 wait_for_clean ||
return 1
1374 local num_pgs
=$
(get_num_pgs
)
1375 test "$num_pgs" -gt 0 ||
return 1
1376 teardown
$dir ||
return 1
1379 #######################################################################
1382 # Return the OSD ids in use by at least one PG in the cluster (either
1383 # in the up or the acting set), according to ceph pg dump pgs. Every
1384 # OSD id shows as many times as they are used in up and acting sets.
1385 # If an OSD id is in both the up and acting set of a given PG, it will
1388 # @param STDOUT a sorted list of OSD ids
1389 # @return 0 on success, 1 on error
1391 function get_osd_id_used_by_pgs
() {
1392 ceph
--format json pg dump pgs
2>/dev
/null | jq
'.pg_stats | .[] | .up[], .acting[]' |
sort
1395 function test_get_osd_id_used_by_pgs
() {
1398 setup
$dir ||
return 1
1399 run_mon
$dir a
--osd_pool_default_size=1 ||
return 1
1400 run_mgr
$dir x ||
return 1
1401 run_osd
$dir 0 ||
return 1
1402 create_rbd_pool ||
return 1
1403 wait_for_clean ||
return 1
1404 local osd_ids
=$
(get_osd_id_used_by_pgs |
uniq)
1405 test "$osd_ids" = "0" ||
return 1
1406 teardown
$dir ||
return 1
1409 #######################################################################
1412 # Wait until the OSD **id** shows **count** times in the
1413 # PGs (see get_osd_id_used_by_pgs for more information about
1414 # how OSD ids are counted).
1416 # @param id the OSD id
1417 # @param count the number of time it must show in the PGs
1418 # @return 0 on success, 1 on error
1420 function wait_osd_id_used_by_pgs
() {
1425 for ((i
=0; i
< $TIMEOUT / 5; i
++)); do
1427 if ! test $
(get_osd_id_used_by_pgs |
grep -c $id) = $count ; then
1437 function test_wait_osd_id_used_by_pgs
() {
1440 setup
$dir ||
return 1
1441 run_mon
$dir a
--osd_pool_default_size=1 ||
return 1
1442 run_mgr
$dir x ||
return 1
1443 run_osd
$dir 0 ||
return 1
1444 create_rbd_pool ||
return 1
1445 wait_for_clean ||
return 1
1446 wait_osd_id_used_by_pgs
0 8 ||
return 1
1447 ! TIMEOUT
=1 wait_osd_id_used_by_pgs
123 5 ||
return 1
1448 teardown
$dir ||
return 1
1451 #######################################################################
1454 # Return the date and time of the last completed scrub for **pgid**,
1455 # as reported by ceph pg dump pgs. Note that a repair also sets this
1458 # @param pgid the id of the PG
1459 # @param STDOUT the date and time of the last scrub
1460 # @return 0 on success, 1 on error
1462 function get_last_scrub_stamp
() {
1464 local sname
=${2:-last_scrub_stamp}
1465 ceph
--format json pg dump pgs
2>/dev
/null | \
1466 jq
-r ".pg_stats | .[] | select(.pgid==\"$pgid\") | .$sname"
1469 function test_get_last_scrub_stamp
() {
1472 setup
$dir ||
return 1
1473 run_mon
$dir a
--osd_pool_default_size=1 ||
return 1
1474 run_mgr
$dir x ||
return 1
1475 run_osd
$dir 0 ||
return 1
1476 create_rbd_pool ||
return 1
1477 wait_for_clean ||
return 1
1478 stamp
=$
(get_last_scrub_stamp
1.0)
1479 test -n "$stamp" ||
return 1
1480 teardown
$dir ||
return 1
1483 #######################################################################
1486 # Predicate checking if the cluster is clean, i.e. all of its PGs are
1487 # in a clean state (see get_num_active_clean for a definition).
1489 # @return 0 if the cluster is clean, 1 otherwise
1491 function is_clean
() {
1492 num_pgs
=$
(get_num_pgs
)
1493 test $num_pgs != 0 ||
return 1
1494 test $
(get_num_active_clean
) = $num_pgs ||
return 1
1497 function test_is_clean
() {
1500 setup
$dir ||
return 1
1501 run_mon
$dir a
--osd_pool_default_size=1 ||
return 1
1502 run_mgr
$dir x ||
return 1
1503 run_osd
$dir 0 ||
return 1
1504 create_rbd_pool ||
return 1
1505 wait_for_clean ||
return 1
1506 is_clean ||
return 1
1507 teardown
$dir ||
return 1
1510 #######################################################################
1512 calc
() { $AWK "BEGIN{print $*}"; }
1515 # Return a list of numbers that are increasingly larger and whose
1516 # total is **timeout** seconds. It can be used to have short sleep
1517 # delay while waiting for an event on a fast machine. But if running
1518 # very slowly the larger delays avoid stressing the machine even
1519 # further or spamming the logs.
1521 # @param timeout sum of all delays, in seconds
1522 # @return a list of sleep delays
1524 function get_timeout_delays
() {
1525 local trace
=$
(shopt -q -o xtrace
&& echo true ||
echo false
)
1526 $trace && shopt -u -o xtrace
1528 local first_step
=${2:-1}
1529 local max_timeout
=${3:-$MAX_TIMEOUT}
1534 while test "$(calc $total + $i \<= $timeout)" = "1"; do
1535 echo -n "$(calc $i) "
1536 total
=$
(calc
$total + $i)
1538 if [ $max_timeout -gt 0 ]; then
1539 # Did we reach max timeout ?
1540 if [ ${i%.*} -eq ${max_timeout%.*} ] && [ ${i#*.} \> ${max_timeout#*.} ] || [ ${i%.*} -gt ${max_timeout%.*} ]; then
1541 # Yes, so let's cap the max wait time to max
1546 if test "$(calc $total \< $timeout)" = "1"; then
1547 echo -n "$(calc $timeout - $total) "
1549 $trace && shopt -s -o xtrace
1552 function test_get_timeout_delays
() {
1553 test "$(get_timeout_delays 1)" = "1 " ||
return 1
1554 test "$(get_timeout_delays 5)" = "1 2 2 " ||
return 1
1555 test "$(get_timeout_delays 6)" = "1 2 3 " ||
return 1
1556 test "$(get_timeout_delays 7)" = "1 2 4 " ||
return 1
1557 test "$(get_timeout_delays 8)" = "1 2 4 1 " ||
return 1
1558 test "$(get_timeout_delays 1 .1)" = "0.1 0.2 0.4 0.3 " ||
return 1
1559 test "$(get_timeout_delays 1.5 .1)" = "0.1 0.2 0.4 0.8 " ||
return 1
1560 test "$(get_timeout_delays 5 .1)" = "0.1 0.2 0.4 0.8 1.6 1.9 " ||
return 1
1561 test "$(get_timeout_delays 6 .1)" = "0.1 0.2 0.4 0.8 1.6 2.9 " ||
return 1
1562 test "$(get_timeout_delays 6.3 .1)" = "0.1 0.2 0.4 0.8 1.6 3.2 " ||
return 1
1563 test "$(get_timeout_delays 20 .1)" = "0.1 0.2 0.4 0.8 1.6 3.2 6.4 7.3 " ||
return 1
1564 test "$(get_timeout_delays 300 .1 0)" = "0.1 0.2 0.4 0.8 1.6 3.2 6.4 12.8 25.6 51.2 102.4 95.3 " ||
return 1
1565 test "$(get_timeout_delays 300 .1 10)" = "0.1 0.2 0.4 0.8 1.6 3.2 6.4 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 7.3 " ||
return 1
1568 #######################################################################
1571 # Wait until the cluster becomes clean or if it does not make progress
1572 # for $WAIT_FOR_CLEAN_TIMEOUT seconds.
1573 # Progress is measured either via the **get_is_making_recovery_progress**
1574 # predicate or if the number of clean PGs changes (as returned by get_num_active_clean)
1576 # @return 0 if the cluster is clean, 1 otherwise
1578 function wait_for_clean
() {
1580 local num_active_clean
=-1
1581 local cur_active_clean
1582 local -a delays
=($
(get_timeout_delays
$WAIT_FOR_CLEAN_TIMEOUT .1))
1585 flush_pg_stats ||
return 1
1586 while test $
(get_num_pgs
) == 0 ; do
1591 # Comparing get_num_active_clean & get_num_pgs is used to determine
1592 # if the cluster is clean. That's almost an inline of is_clean() to
1593 # get more performance by avoiding multiple calls of get_num_active_clean.
1594 cur_active_clean
=$
(get_num_active_clean
)
1595 test $cur_active_clean = $
(get_num_pgs
) && break
1596 if test $cur_active_clean != $num_active_clean ; then
1598 num_active_clean
=$cur_active_clean
1599 elif get_is_making_recovery_progress
; then
1601 elif (( $loop >= ${#delays[*]} )) ; then
1605 # eval is a no-op if cmd is empty
1607 sleep ${delays[$loop]}
1613 function test_wait_for_clean
() {
1616 setup
$dir ||
return 1
1617 run_mon
$dir a
--osd_pool_default_size=2 ||
return 1
1618 run_osd
$dir 0 ||
return 1
1619 run_mgr
$dir x ||
return 1
1620 create_rbd_pool ||
return 1
1621 ! WAIT_FOR_CLEAN_TIMEOUT
=1 wait_for_clean ||
return 1
1622 run_osd
$dir 1 ||
return 1
1623 wait_for_clean ||
return 1
1624 teardown
$dir ||
return 1
1628 # Wait until the cluster becomes peered or if it does not make progress
1629 # for $WAIT_FOR_CLEAN_TIMEOUT seconds.
1630 # Progress is measured either via the **get_is_making_recovery_progress**
1631 # predicate or if the number of peered PGs changes (as returned by get_num_active_or_peered)
1633 # @return 0 if the cluster is clean, 1 otherwise
1635 function wait_for_peered
() {
1639 local -a delays
=($
(get_timeout_delays
$WAIT_FOR_CLEAN_TIMEOUT .1))
1642 flush_pg_stats ||
return 1
1643 while test $
(get_num_pgs
) == 0 ; do
1648 # Comparing get_num_active_clean & get_num_pgs is used to determine
1649 # if the cluster is clean. That's almost an inline of is_clean() to
1650 # get more performance by avoiding multiple calls of get_num_active_clean.
1651 cur_peered
=$
(get_num_active_or_peered
)
1652 test $cur_peered = $
(get_num_pgs
) && break
1653 if test $cur_peered != $num_peered ; then
1655 num_peered
=$cur_peered
1656 elif get_is_making_recovery_progress
; then
1658 elif (( $loop >= ${#delays[*]} )) ; then
1662 # eval is a no-op if cmd is empty
1664 sleep ${delays[$loop]}
1670 function test_wait_for_peered
() {
1673 setup
$dir ||
return 1
1674 run_mon
$dir a
--osd_pool_default_size=2 ||
return 1
1675 run_osd
$dir 0 ||
return 1
1676 run_mgr
$dir x ||
return 1
1677 create_rbd_pool ||
return 1
1678 ! WAIT_FOR_CLEAN_TIMEOUT
=1 wait_for_clean ||
return 1
1679 run_osd
$dir 1 ||
return 1
1680 wait_for_peered ||
return 1
1681 teardown
$dir ||
return 1
1685 #######################################################################
1688 # Wait until the cluster has health condition passed as arg
1689 # again for $TIMEOUT seconds.
1691 # @param string to grep for in health detail
1692 # @return 0 if the cluster health matches request, 1 otherwise
1694 function wait_for_health
() {
1696 local -a delays
=($
(get_timeout_delays
$TIMEOUT .1))
1699 while ! ceph health detail |
grep "$grepstr" ; do
1700 if (( $loop >= ${#delays[*]} )) ; then
1704 sleep ${delays[$loop]}
1710 # Wait until the cluster becomes HEALTH_OK again or if it does not make progress
1711 # for $TIMEOUT seconds.
1713 # @return 0 if the cluster is HEALTHY, 1 otherwise
1715 function wait_for_health_ok
() {
1716 wait_for_health
"HEALTH_OK" ||
return 1
1719 function test_wait_for_health_ok
() {
1722 setup
$dir ||
return 1
1723 run_mon
$dir a
--osd_failsafe_full_ratio=.99 --mon_pg_warn_min_per_osd=0 ||
return 1
1724 run_mgr
$dir x
--mon_pg_warn_min_per_osd=0 ||
return 1
1725 # start osd_pool_default_size OSDs
1726 run_osd
$dir 0 ||
return 1
1727 run_osd
$dir 1 ||
return 1
1728 run_osd
$dir 2 ||
return 1
1729 kill_daemons
$dir TERM osd ||
return 1
1730 ceph osd down
0 ||
return 1
1731 # expect TOO_FEW_OSDS warning
1732 ! TIMEOUT
=1 wait_for_health_ok ||
return 1
1733 # resurrect all OSDs
1734 activate_osd
$dir 0 ||
return 1
1735 activate_osd
$dir 1 ||
return 1
1736 activate_osd
$dir 2 ||
return 1
1737 wait_for_health_ok ||
return 1
1738 teardown
$dir ||
return 1
1742 #######################################################################
1745 # Run repair on **pgid** and wait until it completes. The repair
1746 # function will fail if repair does not complete within $TIMEOUT
1749 # @param pgid the id of the PG
1750 # @return 0 on success, 1 on error
1754 local last_scrub
=$
(get_last_scrub_stamp
$pgid)
1755 ceph pg repair
$pgid
1756 wait_for_scrub
$pgid "$last_scrub"
1759 function test_repair
() {
1762 setup
$dir ||
return 1
1763 run_mon
$dir a
--osd_pool_default_size=1 ||
return 1
1764 run_mgr
$dir x ||
return 1
1765 run_osd
$dir 0 ||
return 1
1766 create_rbd_pool ||
return 1
1767 wait_for_clean ||
return 1
1768 repair
1.0 ||
return 1
1769 kill_daemons
$dir KILL osd ||
return 1
1770 ! TIMEOUT
=1 repair
1.0 ||
return 1
1771 teardown
$dir ||
return 1
1773 #######################################################################
1776 # Run scrub on **pgid** and wait until it completes. The pg_scrub
1777 # function will fail if repair does not complete within $TIMEOUT
1778 # seconds. The pg_scrub is complete whenever the
1779 # **get_last_scrub_stamp** function reports a timestamp different from
1780 # the one stored before starting the scrub.
1782 # @param pgid the id of the PG
1783 # @return 0 on success, 1 on error
1785 function pg_scrub
() {
1787 local last_scrub
=$
(get_last_scrub_stamp
$pgid)
1789 wait_for_scrub
$pgid "$last_scrub"
1792 function pg_deep_scrub
() {
1794 local last_scrub
=$
(get_last_scrub_stamp
$pgid last_deep_scrub_stamp
)
1795 ceph pg deep-scrub
$pgid
1796 wait_for_scrub
$pgid "$last_scrub" last_deep_scrub_stamp
1799 function test_pg_scrub
() {
1802 setup
$dir ||
return 1
1803 run_mon
$dir a
--osd_pool_default_size=1 ||
return 1
1804 run_mgr
$dir x ||
return 1
1805 run_osd
$dir 0 ||
return 1
1806 create_rbd_pool ||
return 1
1807 wait_for_clean ||
return 1
1808 pg_scrub
1.0 ||
return 1
1809 kill_daemons
$dir KILL osd ||
return 1
1810 ! TIMEOUT
=1 pg_scrub
1.0 ||
return 1
1811 teardown
$dir ||
return 1
1814 #######################################################################
1817 # Run the *command* and expect it to fail (i.e. return a non zero status).
1818 # The output (stderr and stdout) is stored in a temporary file in *dir*
1819 # and is expected to contain the string *expected*.
1821 # Return 0 if the command failed and the string was found. Otherwise
1822 # return 1 and cat the full output of the command on stderr for debug.
1824 # @param dir temporary directory to store the output
1825 # @param expected string to look for in the output
1826 # @param command ... the command and its arguments
1827 # @return 0 on success, 1 on error
1830 function expect_failure
() {
1837 if "$@" > $dir/out
2>&1 ; then
1843 if $success ||
! grep --quiet "$expected" $dir/out
; then
1851 function test_expect_failure
() {
1854 setup
$dir ||
return 1
1855 expect_failure
$dir FAIL bash
-c 'echo FAIL ; exit 1' ||
return 1
1856 # the command did not fail
1857 ! expect_failure
$dir FAIL bash
-c 'echo FAIL ; exit 0' > $dir/out ||
return 1
1858 grep --quiet FAIL
$dir/out ||
return 1
1859 # the command failed but the output does not contain the expected string
1860 ! expect_failure
$dir FAIL bash
-c 'echo UNEXPECTED ; exit 1' > $dir/out ||
return 1
1861 ! grep --quiet FAIL
$dir/out ||
return 1
1862 teardown
$dir ||
return 1
1865 #######################################################################
1868 # Given the *last_scrub*, wait for scrub to happen on **pgid**. It
1869 # will fail if scrub does not complete within $TIMEOUT seconds. The
1870 # repair is complete whenever the **get_last_scrub_stamp** function
1871 # reports a timestamp different from the one given in argument.
1873 # @param pgid the id of the PG
1874 # @param last_scrub timestamp of the last scrub for *pgid*
1875 # @return 0 on success, 1 on error
1877 function wait_for_scrub
() {
1879 local last_scrub
="$2"
1880 local sname
=${3:-last_scrub_stamp}
1882 for ((i
=0; i
< $TIMEOUT; i
++)); do
1883 if test "$(get_last_scrub_stamp $pgid $sname)" '>' "$last_scrub" ; then
1891 function test_wait_for_scrub
() {
1894 setup
$dir ||
return 1
1895 run_mon
$dir a
--osd_pool_default_size=1 ||
return 1
1896 run_mgr
$dir x ||
return 1
1897 run_osd
$dir 0 ||
return 1
1898 create_rbd_pool ||
return 1
1899 wait_for_clean ||
return 1
1901 ceph pg repair
$pgid
1902 local last_scrub
=$
(get_last_scrub_stamp
$pgid)
1903 wait_for_scrub
$pgid "$last_scrub" ||
return 1
1904 kill_daemons
$dir KILL osd ||
return 1
1905 last_scrub
=$
(get_last_scrub_stamp
$pgid)
1906 ! TIMEOUT
=1 wait_for_scrub
$pgid "$last_scrub" ||
return 1
1907 teardown
$dir ||
return 1
1910 #######################################################################
1913 # Return 0 if the erasure code *plugin* is available, 1 otherwise.
1915 # @param plugin erasure code plugin
1916 # @return 0 on success, 1 on error
1919 function erasure_code_plugin_exists
() {
1925 FreeBSD
) grepstr
="Cannot open.*$plugin" ;;
1926 *) grepstr
="$plugin.*No such file" ;;
1929 s
=$
(ceph osd erasure-code-profile
set TESTPROFILE plugin
=$plugin 2>&1)
1931 if [ $status -eq 0 ]; then
1932 ceph osd erasure-code-profile
rm TESTPROFILE
1933 elif ! echo $s |
grep --quiet "$grepstr" ; then
1935 # display why the string was rejected.
1941 function test_erasure_code_plugin_exists
() {
1944 setup
$dir ||
return 1
1945 run_mon
$dir a ||
return 1
1946 run_mgr
$dir x ||
return 1
1947 erasure_code_plugin_exists jerasure ||
return 1
1948 ! erasure_code_plugin_exists FAKE ||
return 1
1949 teardown
$dir ||
return 1
1952 #######################################################################
1955 # Display all log files from **dir** on stdout.
1957 # @param dir directory in which all data is stored
1960 function display_logs
() {
1963 find $dir -maxdepth 1 -name '*.log' | \
1964 while read file ; do
1965 echo "======================= $file"
1970 function test_display_logs
() {
1973 setup
$dir ||
return 1
1974 run_mon
$dir a ||
return 1
1975 kill_daemons
$dir ||
return 1
1976 display_logs
$dir > $dir/log.out
1977 grep --quiet mon.a.log
$dir/log.out ||
return 1
1978 teardown
$dir ||
return 1
1981 #######################################################################
1983 # Spawn a command in background and save the pid in the variable name
1984 # passed in argument. To make the output reading easier, the output is
1985 # prepend with the process id.
1989 # run_in_background pids1 bash -c 'sleep 1; exit 1'
1991 # @param pid_variable the variable name (not value) where the pids will be stored
1992 # @param ... the command to execute
1993 # @return only the pid_variable output should be considered and used with **wait_background**
1995 function run_in_background
() {
1996 local pid_variable
=$1
1998 # Execute the command and prepend the output with its pid
1999 # We enforce to return the exit status of the command and not the sed one.
2000 ("$@" |
& sed 's/^/'$$
': /'; return "${PIPESTATUS[0]}") >&2 &
2001 eval "$pid_variable+=\" $!\""
2004 function save_stdout
{
2010 function test_run_in_background
() {
2012 run_in_background pids
sleep 1
2013 run_in_background pids
sleep 1
2014 test $
(echo $pids |
wc -w) = 2 ||
return 1
2015 wait $pids ||
return 1
2018 #######################################################################
2020 # Wait for pids running in background to complete.
2021 # This function is usually used after a **run_in_background** call
2024 # run_in_background pids1 bash -c 'sleep 1; exit 1'
2025 # wait_background pids1
2027 # @param pids The variable name that contains the active PIDS. Set as empty at then end of the function.
2028 # @return returns 1 if at least one process exits in error unless returns 0
2030 function wait_background
() {
2031 # We extract the PIDS from the variable name
2035 for pid
in $pids; do
2036 if ! wait $pid; then
2037 # If one process failed then return 1
2042 # We empty the variable reporting that all process ended
2049 function test_wait_background
() {
2051 run_in_background pids bash
-c "sleep 1; exit 1"
2052 run_in_background pids bash
-c "sleep 2; exit 0"
2053 wait_background pids
2054 if [ $?
-ne 1 ]; then return 1; fi
2056 run_in_background pids bash
-c "sleep 1; exit 0"
2057 run_in_background pids bash
-c "sleep 2; exit 0"
2058 wait_background pids
2059 if [ $?
-ne 0 ]; then return 1; fi
2061 if [ ! -z "$pids" ]; then return 1; fi
2064 function flush_pg_stats
()
2066 local timeout
=${1:-$TIMEOUT}
2071 seq=`ceph tell osd.$osd flush_pg_stats`
2072 seqs
="$seqs $osd-$seq"
2076 osd
=`echo $s | cut -d - -f 1`
2077 seq=`echo $s | cut -d - -f 2`
2078 echo "waiting osd.$osd seq $seq"
2079 while test $
(ceph osd last-stat-seq
$osd) -lt $seq; do
2081 if [ $
((timeout--
)) -eq 0 ]; then
2088 function test_flush_pg_stats
()
2092 setup
$dir ||
return 1
2093 run_mon
$dir a
--osd_pool_default_size=1 ||
return 1
2094 run_mgr
$dir x ||
return 1
2095 run_osd
$dir 0 ||
return 1
2096 create_rbd_pool ||
return 1
2097 rados
-p rbd put obj
/etc
/group
2098 flush_pg_stats ||
return 1
2099 local jq_filter
='.pools | .[] | select(.name == "rbd") | .stats'
2100 stored
=`ceph df detail --format=json | jq "$jq_filter.stored"`
2101 stored_raw
=`ceph df detail --format=json | jq "$jq_filter.stored_raw"`
2102 test $stored -gt 0 ||
return 1
2103 test $stored == $stored_raw ||
return 1
2107 #######################################################################
2110 # Call the **run** function (which must be defined by the caller) with
2111 # the **dir** argument followed by the caller argument list.
2113 # If the **run** function returns on error, all logs found in **dir**
2114 # are displayed for diagnostic purposes.
2116 # **teardown** function is called when the **run** function returns
2117 # (on success or on error), to cleanup leftovers. The CEPH_CONF is set
2118 # to /dev/null and CEPH_ARGS is unset so that the tests are protected from
2119 # external interferences.
2121 # It is the responsibility of the **run** function to call the
2122 # **setup** function to prepare the test environment (create a temporary
2125 # The shell is required (via PS4) to display the function and line
2126 # number whenever a statement is executed to help debugging.
2128 # @param dir directory in which all data is stored
2129 # @param ... arguments passed transparently to **run**
2130 # @return 0 on success, 1 on error
2137 PS4
='${BASH_SOURCE[0]}:$LINENO: ${FUNCNAME[0]}: '
2139 export PATH
=.
:$PATH # make sure program from sources are preferred
2140 export PYTHONWARNINGS
=ignore
2141 export CEPH_CONF
=/dev
/null
2145 if run
$dir "$@" ; then
2150 teardown
$dir $code ||
return 1
2154 #######################################################################
2156 function run_tests
() {
2158 PS4
='${BASH_SOURCE[0]}:$LINENO: ${FUNCNAME[0]}: '
2160 export .
:$PATH # make sure program from sources are preferred
2162 export CEPH_MON
="127.0.0.1:7109" # git grep '\<7109\>' : there must be only one
2164 CEPH_ARGS
+=" --fsid=$(uuidgen) --auth-supported=none "
2165 CEPH_ARGS
+="--mon-host=$CEPH_MON "
2166 export CEPH_CONF
=/dev
/null
2168 local funcs
=${@:-$(set | sed -n -e 's/^\(test_[0-9a-z_]*\) .*/\1/p')}
2169 local dir
=td
/ceph-helpers
2171 for func
in $funcs ; do
2172 if ! $func $dir; then
2179 if test "$1" = TESTS
; then
2186 # jq only support --exit-status|-e from version 1.4 forwards, which makes
2187 # returning on error waaaay prettier and straightforward.
2188 # However, the current automated upstream build is running with v1.3,
2189 # which has no idea what -e is. Hence the convoluted error checking we
2191 # The next time someone changes this code, please check if v1.4 is now
2192 # a thing, and, if so, please change these to use -e. Thanks.
2194 # jq '.all.supported | select([.[] == "foo"] | any)'
2195 function jq_success
() {
2200 in_escaped
=$
(printf %s
"$input" |
sed "s/'/'\\\\''/g")
2201 filter_escaped
=$
(printf %s
"$filter" |
sed "s/'/'\\\\''/g")
2203 ret
=$
(echo "$in_escaped" | jq
"$filter_escaped")
2204 if [[ "$ret" == "true" ]]; then
2206 elif [[ -n "$expects" ]]; then
2207 if [[ "$ret" == "$expects" ]]; then
2216 ret
="$(echo $input | jq \"$filter\")"
2217 if [[ "$ret" == "true" ]]; then
2219 elif [[ -n "$expects" && "$ret" == "$expects" ]]; then
2225 function inject_eio
() {
2239 local -a initial_osds
=($
(get_osds
$poolname $objname))
2240 local osd_id
=${initial_osds[$shard_id]}
2241 if [ "$pooltype" != "ec" ]; then
2244 type=$
(cat $dir/$osd_id/type)
2245 set_config osd
$osd_id ${type}_debug_inject_read_err true ||
return 1
2247 while ( CEPH_ARGS
='' ceph
--admin-daemon $
(get_asok_path osd.
$osd_id) \
2248 inject
${which}err
$poolname $objname $shard_id |
grep -q Invalid
); do
2249 loop
=$
(expr $loop + 1)
2250 if [ $loop = "10" ]; then
2257 function multidiff
() {
2259 if [ "$DIFFCOLOPTS" = "" ]; then
2262 diff $DIFFCOLOPTS $@
2266 function create_ec_pool
() {
2269 local allow_overwrites
=$1
2272 ceph osd erasure-code-profile
set myprofile crush-failure-domain
=osd
"$@" ||
return 1
2274 create_pool
"$poolname" 1 1 erasure myprofile ||
return 1
2276 if [ "$allow_overwrites" = "true" ]; then
2277 ceph osd pool
set "$poolname" allow_ec_overwrites true ||
return 1
2280 wait_for_clean ||
return 1
2285 # compile-command: "cd ../../src ; make -j4 && ../qa/standalone/ceph-helpers.sh TESTS # test_get_config"