2 # -*- mode: Bash; tab-width: 4; indent-tabs-mode: t; -*-
3 # vim:shiftwidth=4:softtabstop=4:tabstop=4:
7 # ha.sh - test Lustre HA (aka failover) configurations
15 # ha.sh tests Lustre HA (aka failover) configurations with a CRM.
23 # Specify client nodes.
26 # Specify server nodes.
29 # Specify victim nodes to be rebooted.
32 # Choose a parent of the test directory. "/mnt/lustre" if not specified.
35 # Define a duration for the test. 86400 seconds if not specified.
38 # Define a max failover period. 10 minutes if not set.
41 # Only run the workloads; no failure will be introduced.
42 # -v, -s are ignored in this case.
44 # Workloads dry run for several seconds; no failures will be introduced.
45 # This option is useful to verify the loads.
46 # -u is ignored in this case
48 # Reboot victim nodes simultaneously.
53 # A Lustre file system is up and mounted on all client nodes. This script
54 # does not mount or unmount any Lustre targets or clients, let alone format
57 # Each target has a failnode, so that workloads can continue after a power
60 # CRM could be configured by 2 ways:
62 # Targets are automatically failed back when their primary node is back. This
63 # assumption avoids calling CRM-specific commands to trigger failbacks, making
64 # this script more CRM-neural.
66 # Targets are not automatically failed back when their primary node is back.
67 # CRM-specific command is executed to trigger failbacks.
69 # A crash dump mechanism is configured to catch LBUGs, panics, etc.
73 # Each client runs set of MPI and non-MPI workloads. These
74 # applications are run in short loops so that their exit status can be waited
75 # for and checked within reasonable time by ha_wait_loads.
76 # The set of MPI and non-MPI workloads are configurable by parameters:
78 # default set: dd, tar, iozone
80 # default set: ior, simul, mdtest
82 # The number of clients run MPI loads is configured by parameter
83 # ha_mpi_instances. Only one client runs MPI workloads by default.
85 # MPI workloads can be run from several users. The list of users to use is
86 # configured by parameter ha_mpi_users, default is "mpiuser".
88 # PROCESS STRUCTURE AND IPC
90 # On the node where this script is run, the processes look like this:
94 # ~ ha.sh (ha_repeat_mpi_load ior)
96 # ~ ha.sh (ha_repeat_mpi_load simul)
98 # ~ ha.sh (ha_repeat_mpi_load mdtest)
100 # ~ ... (one for each MPI load)
102 # ~ ha.sh (ha_repeat_nonmpi_load client2 dbench)
103 # ~ pdsh client2 dbench
104 # ~ ha.sh (ha_repeat_nonmpi_load client2 iozone)
105 # ~ pdsh client2 iozone
106 # ~ ha.sh (ha_repeat_nonmpi_load client5 iozone)
107 # ~ pdsh client5 iozone
108 # ~ ... (one for each non-MPI load on each client)
110 # Each tilde represents a process. Indentations imply parent-children
113 # IPC is done by files in the temporary directory.
118 SIMUL=${SIMUL:-$(which simul 2> /dev/null || true)}
119 IOR=${IOR:-$(which IOR 2> /dev/null || true)}
120 MDTEST=${MDTEST:-$(which mdtest 2> /dev/null || true)}
122 ior_blockSize=${ior_blockSize:-6g}
123 mpi_threads_per_client=${mpi_threads_per_client:-2}
125 iozone_SIZE=${iozone_SIZE:-262144} # 256m
127 mpirun=${MPIRUN:-$(which mpirun)}
128 LFS=${LFS:-$(which lfs)}
132 for ((load = 0; load < ${#ha_mpi_load_tags[@]}; load++)); do
133 local tag=${ha_mpi_load_tags[$load]}
134 local bin=$(echo $tag | tr '[:lower:]' '[:upper:]')
135 if [ x${!bin} = x ]; then
136 ha_error ha_mpi_loads: ${ha_mpi_loads}, $bin is not set
144 echo "$0: $(date +%H:%M:%S' '%s):" "$@"
149 local nodes=${1// /,}
151 ha_on $nodes "lctl mark $*"
163 ha_error "Trap ERR triggered by:"
164 ha_error " $BASH_COMMAND"
165 ha_error "Call trace:"
166 for ((i = 0; i < ${#FUNCNAME[@]}; i++)); do
167 ha_error " ${FUNCNAME[$i]} [${BASH_SOURCE[$i]}:${BASH_LINENO[$i]}]"
174 declare ha_tmp_dir=/tmp/$(basename $0)-$$
175 declare ha_stop_file=$ha_tmp_dir/stop
176 declare ha_fail_file=$ha_tmp_dir/fail
177 declare ha_pm_states=$ha_tmp_dir/ha_pm_states
178 declare ha_status_file_prefix=$ha_tmp_dir/status
179 declare -a ha_status_files
180 declare ha_machine_file=$ha_tmp_dir/machine_file
181 declare ha_lfsck_log=$ha_tmp_dir/lfsck.log
182 declare ha_lfsck_lock=$ha_tmp_dir/lfsck.lock
183 declare ha_lfsck_stop=$ha_tmp_dir/lfsck.stop
184 declare ha_lfsck_bg=${LFSCK_BG:-false}
185 declare ha_lfsck_after=${LFSCK_AFTER:-false}
186 declare ha_lfsck_node=${LFSCK_NODE:-""}
187 declare ha_lfsck_device=${LFSCK_DEV:-""}
188 declare ha_lfsck_types=${LFSCK_TYPES:-"namespace layout"}
189 declare ha_lfsck_custom_params=${LFSCK_CUSTOM_PARAMS:-""}
190 declare ha_lfsck_wait=${LFSCK_WAIT:-1200}
191 declare ha_lfsck_fail_on_repaired=${LFSCK_FAIL_ON_REPAIRED:-false}
192 declare ha_power_down_cmd=${POWER_DOWN:-"pm -0"}
193 declare ha_power_up_cmd=${POWER_UP:-"pm -1"}
194 declare ha_power_delay=${POWER_DELAY:-60}
195 declare ha_node_up_delay=${NODE_UP_DELAY:-10}
196 declare ha_pm_host=${PM_HOST:-$(hostname)}
197 declare ha_failback_delay=${DELAY:-5}
198 declare ha_failback_cmd=${FAILBACK:-""}
199 declare ha_stripe_params=${STRIPEPARAMS:-"-c 0"}
200 declare ha_test_dir_stripe_count=${TDSTRIPECOUNT:-"1"}
201 declare ha_test_dir_mdt_index=${TDMDTINDEX:-"0"}
202 declare ha_test_dir_mdt_index_random=${TDMDTINDEXRAND:-false}
203 declare ha_dir_stripe_count=${DSTRIPECOUNT:-"1"}
204 declare ha_mdt_index=${MDTINDEX:-"0"}
205 declare ha_mdt_index_random=${MDTINDEXRAND:-false}
206 declare -a ha_clients
207 declare -a ha_servers
208 declare -a ha_victims
209 declare -a ha_victims_pair
210 declare ha_test_dir=/mnt/lustre/$(basename $0)-$$
211 declare -a ha_testdirs=(${ha_test_dirs="$ha_test_dir"})
213 for ((i=0; i<${#ha_testdirs[@]}; i++)); do
214 echo I=$i ${ha_testdirs[i]}
215 ha_testdirs[i]="${ha_testdirs[i]}/$(basename $0)-$$"
216 echo i=$i ${ha_testdirs[i]}
219 declare ha_cleanup=${CLEANUP:-true}
220 declare ha_start_time=$(date +%s)
221 declare ha_expected_duration=$((60 * 60 * 24))
222 declare ha_max_failover_period=10
223 declare ha_nr_loops=0
224 declare ha_stop_signals="SIGINT SIGTERM SIGHUP"
225 declare ha_load_timeout=${LOAD_TIMEOUT:-$((60 * 10))}
226 declare ha_workloads_only=false
227 declare ha_workloads_dry_run=false
228 declare ha_simultaneous=false
230 declare ha_mpi_instances=${ha_mpi_instances:-1}
232 declare ha_mpi_loads=${ha_mpi_loads="ior simul mdtest"}
233 declare -a ha_mpi_load_tags=($ha_mpi_loads)
234 declare -a ha_mpiusers=(${ha_mpi_users="mpiuser"})
236 declare -A ha_mpiopts
238 for ((i=0; i<${#ha_mpiusers[@]}; i++)); do
239 u=${ha_mpiusers[i]%%:*}
241 # user gets empty option if ha_mpi_users does not specify it explicitly
242 [[ ${ha_mpiusers[i]} =~ : ]] && o=${ha_mpiusers[i]##*:}
244 ha_mpiopts[$u]+=" $o"
246 ha_users=(${!ha_mpiopts[@]})
248 declare ha_ior_params=${IORP:-'" -b $ior_blockSize -t 2m -w -W -T 1"'}
249 declare ha_simul_params=${SIMULP:-'" -n 10"'}
250 declare ha_mdtest_params=${MDTESTP:-'" -i 1 -n 1000"'}
251 declare ha_mpirun_options=${MPIRUN_OPTIONS:-""}
252 declare ha_clients_stripe=${CLIENTSSTRIPE:-'"$STRIPEPARAMS"'}
253 declare ha_nclientsset=${NCLIENTSSET:-1}
254 declare ha_ninstmustfail=${NINSTMUSTFAIL:-0}
256 declare ha_racer_params=${RACERP:-"MDSCOUNT=1"}
258 eval ha_params_ior=($ha_ior_params)
259 eval ha_params_simul=($ha_simul_params)
260 eval ha_params_mdtest=($ha_mdtest_params)
261 eval ha_stripe_clients=($ha_clients_stripe)
263 declare ha_nparams_ior=${#ha_params_ior[@]}
264 declare ha_nparams_simul=${#ha_params_simul[@]}
265 declare ha_nparams_mdtest=${#ha_params_mdtest[@]}
266 declare ha_nstripe_clients=${#ha_stripe_clients[@]}
268 declare -A ha_mpi_load_cmds=(
269 [ior]="$IOR -o {}/f.ior {params}"
270 [simul]="$SIMUL {params} -d {}"
271 [mdtest]="$MDTEST {params} -d {}"
274 declare racer=${RACER:-"$(dirname $0)/racer/racer.sh"}
276 declare ha_nonmpi_loads=${ha_nonmpi_loads="dd tar iozone"}
277 declare -a ha_nonmpi_load_tags=($ha_nonmpi_loads)
278 declare -A ha_nonmpi_load_cmds=(
279 [dd]="dd if=/dev/zero of={}/f.dd bs=1M count=256"
280 [tar]="tar cf - /etc | tar xf - -C {}"
281 [iozone]="iozone -a -e -+d -s $iozone_SIZE {}/f.iozone"
282 [racer]="$ha_racer_params $racer {}"
287 ha_info "Usage: $0 -c HOST[,...] -s HOST[,...]" \
288 "-v HOST[,...] -f HOST[,...] [-d DIRECTORY] [-u SECONDS]"
291 ha_process_arguments()
295 while getopts hc:s:v:d:p:u:wrmf: opt; do
302 ha_clients=(${OPTARG//,/ })
305 ha_servers=(${OPTARG//,/ })
308 ha_victims=(${OPTARG//,/ })
311 ha_test_dir=$OPTARG/$(basename $0)-$$
314 ha_expected_duration=$OPTARG
317 ha_max_failover_period=$OPTARG
320 ha_workloads_only=true
323 ha_workloads_dry_run=true
329 ha_victims_pair=(${OPTARG//,/ })
338 if [ -z "${ha_clients[*]}" ]; then
339 ha_error "-c is mandatory"
343 if ! ($ha_workloads_dry_run ||
344 $ha_workloads_only) &&
345 ([ -z "${ha_servers[*]}" ] ||
346 [ -z "${ha_victims[*]}" ]); then
347 ha_error "-s, and -v are all mandatory"
361 # -S is to be used here to track the
362 # remote command return values
364 pdsh -S -w $nodes "PATH=/usr/local/sbin:/usr/local/bin:/sbin:\
365 /bin:/usr/sbin:/usr/bin; $@" ||
372 touch "$ha_stop_file"
374 if [ -e "$ha_fail_file" ]; then
375 ha_info "Test directories ${ha_testdirs[@]} not removed"
376 ha_info "Temporary directory $ha_tmp_dir not removed"
379 ha_on ${ha_clients[0]} rm -rf ${ha_testdirs[@]} ||
380 ha_info "Test directories ${ha_testdirs[@]} not removed"
381 ha_info "Please find the results in the directory $ha_tmp_dir"
385 ha_trap_stop_signals()
387 ha_info "${ha_stop_signals// /,} received"
388 touch "$ha_stop_file"
395 ha_info "Sleeping for ${n}s"
397 # sleep(1) could interrupted.
406 while [ -e $lock ]; do
415 until mkdir "$lock" >/dev/null 2>&1; do
416 ha_sleep 1 >/dev/null
429 local nodes=${1// /,}
430 local file=/tmp/$(basename $0)-$$-$(date +%s).dk
431 local lock=$ha_tmp_dir/lock-dump-logs
435 ha_info "Dumping lctl log to $file"
438 # some nodes could crash, so
439 # do not exit with error if not all logs are dumped
441 ha_on $nodes "lctl dk >>$file" || rc=$?
444 ha_error "not all logs are dumped! Some nodes are unreachable."
455 local stripeparams=$6
458 local mpirunoptions=$9
460 local tag=${ha_mpi_load_tags[$load]}
461 local cmd=${ha_mpi_load_cmds[$tag]}
462 local dir=$test_dir/$client-$tag
463 local log=$ha_tmp_dir/$client-$tag
466 local avg_loop_time=0
467 local start_time=$(date +%s)
469 cmd=${cmd//"{}"/$dir}
470 cmd=${cmd//"{params}"/$parameter}
472 [[ -n "$ha_postcmd" ]] && ha_postcmd=${ha_postcmd//"{}"/$dir}
473 [[ -n "$ha_precmd" ]] && ha_precmd=${ha_precmd//"{}"/$dir}
474 ha_info "Starting $tag"
476 machines="-machinefile $machines"
477 while [ ! -e "$ha_stop_file" ] && ((rc == 0)); do
478 ha_info "$client Starts: $mpiuser: $cmd" 2>&1 | tee -a $log
481 if $ha_mdt_index_random && [ $ha_mdt_index -ne 0 ]; then
482 mdt_index=$(ha_rand $ha_mdt_index)
484 mdt_index=$ha_mdt_index
486 [[ -n "$ha_precmd" ]] && ha_info "$ha_precmd" &&
487 ha_on $client "$ha_precmd" >>"$log" 2>&1
488 ha_on $client $LFS mkdir -i$mdt_index -c$ha_dir_stripe_count "$dir" &&
489 ha_on $client $LFS getdirstripe "$dir" &&
490 ha_on $client $LFS setstripe $stripeparams $dir &&
491 ha_on $client $LFS getstripe $dir &&
492 ha_on $client chmod a+xwr $dir &&
493 ha_on $client "su $mpiuser sh -c \" $mpirun $mpirunoptions \
494 -np $((${#ha_clients[@]} * mpi_threads_per_client / ha_nclientsset)) \
495 $machines $cmd \" " || rc=$?
496 [[ -n "$ha_postcmd" ]] && ha_info "$ha_postcmd" &&
497 ha_on $client "$ha_postcmd" >>"$log" 2>&1
498 (( ((rc == 0)) && (( mustpass != 0 )) )) ||
499 (( ((rc != 0)) && (( mustpass == 0 )) )) &&
500 ha_on $client rm -rf "$dir";
501 } >>"$log" 2>&1 || rc=$?
503 ha_info $client: rc=$rc mustpass=$mustpass
505 # mustpass=0 means that failure is expected
506 if (( rc !=0 )); then
507 if (( mustpass != 0 )); then
508 touch "$ha_fail_file"
509 touch "$ha_stop_file"
510 ha_dump_logs "${ha_clients[*]} ${ha_servers[*]}"
515 elif (( mustpass == 0 )); then
516 touch "$ha_fail_file"
517 touch "$ha_stop_file"
518 ha_dump_logs "${ha_clients[*]} ${ha_servers[*]}"
520 echo rc=$rc mustpass=$mustpass >"$status"
522 nr_loops=$((nr_loops + 1))
525 [ $nr_loops -ne 0 ] &&
526 avg_loop_time=$((($(date +%s) - start_time) / nr_loops))
528 ha_info "$tag stopped: rc=$rc mustpass=$mustpass \
529 avg loop time $avg_loop_time"
546 # ha_mpi_instances defines the number of
547 # clients start mpi loads; should be <= ${#ha_clients[@]}
549 # ha_mpi_instances = 0
551 # ${#ha_mpi_load_tags[@]} =0
552 local inst=$ha_mpi_instances
553 (( inst == 0 )) || (( ${#ha_mpi_load_tags[@]} == 0 )) &&
554 ha_info "no mpi load to start" &&
557 (( inst <= ${#ha_clients[@]} )) || inst=${#ha_clients[@]}
559 # Define names for machinefiles for each client set
560 for (( n=0; n < $ha_nclientsset; n++ )); do
561 mach[$n]=$ha_machine_file$n
564 for ((n = 0; n < ${#ha_clients[@]}; n++)); do
565 m=$(( n % ha_nclientsset))
567 ha_info machine_file=$machines
568 echo ${ha_clients[n]} >> $machines
570 local dirname=$(dirname $ha_machine_file)
571 for client in ${ha_clients[@]}; do
572 ha_on $client mkdir -p $dirname
573 scp $ha_machine_file* $client:$dirname
577 for ((n = 0; n < $inst; n++)); do
578 client=${ha_clients[n]}
579 nmpi=$((n % ${#ha_users[@]}))
580 mpiuser=${ha_users[nmpi]}
581 ndir=$((n % ${#ha_testdirs[@]}))
582 test_dir=${ha_testdirs[ndir]}
583 for ((load = 0; load < ${#ha_mpi_load_tags[@]}; load++)); do
584 tag=${ha_mpi_load_tags[$load]}
585 status=$ha_status_file_prefix-$tag-$client
588 local num=ha_nparams_$tag
590 local aref=ha_params_$tag[nparam]
591 local parameter=${!aref}
592 local nstripe=$((n % ha_nstripe_clients))
593 aref=ha_stripe_clients[nstripe]
594 local stripe=${!aref}
595 local m=$(( n % ha_nclientsset))
598 [[ $ha_ninstmustfail == 0 ]] ||
599 mustpass=$(( n % ha_ninstmustfail ))
600 ha_repeat_mpi_load $client $load $status "$parameter" \
601 $machines "$stripe" "$mpiuser" "$mustpass" \
602 "${ha_mpiopts[$mpiuser]} $ha_mpirun_options" "$test_dir" &
603 ha_status_files+=("$status")
608 ha_repeat_nonmpi_load()
613 local tag=${ha_nonmpi_load_tags[$load]}
614 local cmd=${ha_nonmpi_load_cmds[$tag]}
616 local dir=$test_dir/$client-$tag
617 local log=$ha_tmp_dir/$client-$tag
620 local avg_loop_time=0
621 local start_time=$(date +%s)
623 cmd=${cmd//"{}"/$dir}
625 ha_info "Starting $tag on $client on $dir"
627 while [ ! -e "$ha_stop_file" ] && ((rc == 0)); do
628 ha_info "$client Starts: $cmd" 2>&1 | tee -a $log
629 ha_on $client "mkdir -p $dir && \
631 rm -rf $dir" >>"$log" 2>&1 || rc=$?
634 ha_dump_logs "${ha_clients[*]} ${ha_servers[*]}"
635 touch "$ha_fail_file"
636 touch "$ha_stop_file"
640 nr_loops=$((nr_loops + 1))
643 [ $nr_loops -ne 0 ] &&
644 avg_loop_time=$((($(date +%s) - start_time) / nr_loops))
646 ha_info "$tag on $client stopped: rc $rc avg loop time ${avg_loop_time}s"
649 ha_start_nonmpi_loads()
659 for (( n = 0; n < ${#ha_clients[@]}; n++)); do
660 client=${ha_clients[n]}
661 ndir=$((n % ${#ha_testdirs[@]}))
662 test_dir=${ha_testdirs[ndir]}
663 for ((load = 0; load < ${#ha_nonmpi_load_tags[@]}; load++)); do
664 tag=${ha_nonmpi_load_tags[$load]}
665 status=$ha_status_file_prefix-$tag-$client
666 ha_repeat_nonmpi_load $client $load $status $test_dir &
667 ha_status_files+=("$status")
672 declare ha_bgcmd=${ha_bgcmd:-""}
673 declare ha_bgcmd_log=$ha_tmp_dir/bgcmdlog
676 [[ -z "$ha_bgcmd" ]] && return 0
677 for ((i=0; i<${#ha_testdirs[@]}; i++)); do
678 ha_bgcmd=${ha_bgcmd//"{}"/${ha_testdirs[i]}}
681 ha_info "BG cmd: $ha_bgcmd"
683 [ -f $ha_stop_file ] &&
684 ha_info "$ha_stop_file found! $ha_bgcmd no started" &&
686 eval $ha_bgcmd 2>&1 | tee -a $ha_bgcmd_log
690 ha_info CMD BG PID: $CMD_BG_PID
691 ps aux | grep $CMD_BG_PID
700 [ -f $ha_lfsck_stop ] && ha_info "LFSCK stopped" && break
701 [ -f $ha_stop_file ] &&
702 ha_info "$ha_stop_file found! LFSCK not started" &&
704 ha_start_lfsck 2>&1 | tee -a $ha_lfsck_log
708 ha_info LFSCK BG PID: $LFSCK_BG_PID
711 ha_wait_lfsck_completed () {
713 local -a types=($ha_lfsck_types)
717 local nodes="${ha_servers[@]}"
720 # -A start LFSCK on all nodes
722 [ ${#types[@]} -eq 0 ] && types=(namespace layout)
723 ha_info "Waiting LFSCK completed in $ha_lfsck_wait sec: types ${types[@]}"
724 for type in ${types[@]}; do
726 for (( i=0; i<=ha_lfsck_wait; i++)); do
727 status=($(ha_on $nodes lctl get_param -n *.*.lfsck_$type 2>/dev/null | \
728 awk '/status/ { print $3 }'))
729 for (( s=0; s<${#status[@]}; s++ )); do
730 # "partial" is expected after HARD failover
731 [[ "${status[s]}" = "completed" ]] ||
732 [[ "${status[s]}" = "partial" ]] || break
734 [[ $s -eq ${#status[@]} ]] && eval var_$type=1 && break
737 ha_info "LFSCK $type status in $i sec:"
738 ha_on $nodes lctl get_param -n *.*.lfsck_$type 2>/dev/null | grep status
742 for type in ${types[@]}; do
744 ha_on $nodes lctl get_param -n *.*.lfsck_$type 2>/dev/null
745 [[ ${!var} -eq 1 ]] ||
746 { ha_info "lfsck not completed in $ha_lfsck_wait sec";
754 local -a types=($ha_lfsck_types)
757 # -A: start LFSCK on all nodes via the specified MDT device
758 # (see "-M" option) by single LFSCK command
759 local params=" -A -r $ha_lfsck_custom_params"
761 # use specified device if set
762 [ -n "$ha_lfsck_device" ] && params="-M $ha_lfsck_device $params"
764 # -t: check type(s) to be performed (default all)
765 # check only specified types if set
766 if [ ${#types[@]} -ne 0 ]; then
767 local type="${types[@]}"
768 params="$params -t ${type// /,}"
771 ha_info "LFSCK start $params"
772 ha_on $ha_lfsck_node "lctl lfsck_start $params" || rc=1
773 if [ $rc -ne 0 ]; then
774 if [ -e $ha_lfsck_lock ]; then
776 ha_wait_unlock $ha_lfsck_lock
778 ha_on $ha_lfsck_node "lctl lfsck_start $params" || rc=1
783 { touch "$ha_fail_file"; touch "$ha_stop_file";
784 touch $ha_lfsck_stop; return 1; }
786 ha_wait_lfsck_completed ||
787 { touch "$ha_fail_file"; touch "$ha_stop_file";
788 touch $ha_lfsck_stop; return 1; }
797 n=$(cat $ha_lfsck_log | awk '/repaired/ {print $3}' |\
798 awk '{sum += $1} END { print sum }')
800 { ha_info "Total repaired: $n";
801 touch "$ha_fail_file"; return 1; }
808 $ha_lfsck_bg && ha_lfsck_bg
809 trap ha_trap_stop_signals $ha_stop_signals
810 ha_start_nonmpi_loads
817 [[ -n $CMD_BG_PID ]] && wait $CMD_BG_PID || true
818 # true because of lfsck_bg could be stopped already
819 $ha_lfsck_bg && wait $LFSCK_BG_PID || true
820 trap - $ha_stop_signals
821 ha_info "Waiting for workloads to stop"
828 local end=$(($(date +%s) + ha_load_timeout))
830 ha_info "Waiting $ha_load_timeout sec for workload status..."
831 rm -f "${ha_status_files[@]}"
834 # return immediately if ha_stop_file exists,
835 # all status_files not needed to be checked
837 for file in "${ha_status_files[@]}"; do
838 if [ -e "$ha_stop_file" ]; then
839 ha_info "$ha_stop_file found! Stop."
843 # Wait status file created during ha_load_timeout.
844 # Existing file guarantees that some application
845 # is completed. If no status file was created
846 # this function guarantees that we allow
847 # applications to continue after/before
848 # failover/failback during ha_load_timeout time.
850 until [ -e "$file" ] || (($(date +%s) >= end)); do
852 # check ha_stop_file again, it could appear
853 # during ha_load_timeout
855 if [ -e "$ha_stop_file" ]; then
856 ha_info "$ha_stop_file found! Stop."
859 ha_sleep 1 >/dev/null
867 local expected_state=$2
873 # store pm -x -q $nodes results in a file to have
874 # more information about nodes statuses
875 ha_on $ha_pm_host pm -x -q $nodes | awk '{print $2 $3}' > $ha_pm_states
879 while IFS=": " read node state; do
880 [[ "$state" = "$expected_state" ]] && {
881 nodes=${nodes/$node/}
888 if [ -n "$nodes" ]; then
895 ha_power_down_cmd_fn()
900 case $ha_power_down_cmd in
901 # format is: POWER_DOWN=sysrqcrash
902 sysrqcrash) cmd="pdsh -S -w $nodes 'echo c > /proc/sysrq-trigger' &" ;;
903 *) cmd="$ha_power_down_cmd $nodes" ;;
916 case $ha_power_down_cmd in
918 sysrqcrash) state=off ;;
922 if $ha_lfsck_bg && [[ ${nodes//,/ /} =~ $ha_lfsck_node ]]; then
923 ha_info "$ha_lfsck_node down, delay start LFSCK"
924 ha_lock $ha_lfsck_lock
927 ha_info "Powering down $nodes : cmd: $ha_power_down_cmd"
928 for (( i=0; i<10; i++ )) {
929 ha_info "attempt: $i"
930 ha_power_down_cmd_fn $nodes &&
931 ha_powermanage $nodes $state && rc=0 && break
932 sleep $ha_power_delay
936 ha_info "Failed Powering down in $i attempts:" \
948 for ((i=0; i<${#ha_victims[@]}; i++)) {
949 [[ ${ha_victims[i]} == $node ]] && echo ${ha_victims_pair[i]} &&
952 [[ $i -ne ${#ha_victims[@]} ]] ||
953 ha_error "No pair found!"
959 local end=$(($(date +%s) + ha_node_up_delay))
962 if [[ ${#ha_victims_pair[@]} -eq 0 ]]; then
963 ha_sleep $ha_node_up_delay
967 # Check CRM status on failover pair
968 while (($(date +%s) <= end)); do
970 for n in ${nodes//,/ }; do
971 local pair=$(ha_get_pair $n)
972 local status=$(ha_on $pair crm_mon -1rQ | \
973 grep -w $n | head -1)
975 ha_info "$n pair: $pair status: $status"
976 [[ "$status" == *OFFLINE* ]] ||
981 if [[ $rc -eq 0 ]]; then
982 ha_info "CRM: Got all victims status OFFLINE"
988 ha_info "$nodes CRM status not OFFLINE"
989 for n in ${nodes//,/ }; do
990 local pair=$(ha_get_pair $n)
993 ha_on $pair crm_mon -1rQ
995 ha_error "CRM: some of $nodes are not OFFLINE in $ha_node_up_delay sec"
1005 ha_power_up_delay $nodes
1006 ha_info "Powering up $nodes : cmd: $ha_power_up_cmd"
1007 for (( i=0; i<10; i++ )) {
1008 ha_info "attempt: $i"
1009 $ha_power_up_cmd $nodes &&
1010 ha_powermanage $nodes on && rc=0 && break
1011 sleep $ha_power_delay
1015 ha_info "Failed Powering up in $i attempts: $ha_power_up_cmd"
1024 # Print a random integer within [0, MAX).
1031 # See "5.2 Bash Variables" from "info bash".
1033 echo -n $((RANDOM * max / 32768))
1041 if $ha_simultaneous ; then
1042 nodes=$(echo ${ha_victims[@]})
1045 i=$(ha_rand ${#ha_victims[@]})
1046 nodes=${ha_victims[$i]}
1055 local end=$(($(date +%s) + 10 * 60))
1057 ha_info "Waiting for $nodes to boot up"
1058 until ha_on $nodes hostname >/dev/null 2>&1 ||
1059 [ -e "$ha_stop_file" ] ||
1060 (($(date +%s) >= end)); do
1061 ha_sleep 1 >/dev/null
1068 ha_info "Failback resources on $nodes in $ha_failback_delay sec"
1070 ha_sleep $ha_failback_delay
1071 [ "$ha_failback_cmd" ] ||
1073 ha_info "No failback command set, skiping"
1077 $ha_failback_cmd $nodes
1078 [ -e $ha_lfsck_lock ] && ha_unlock $ha_lfsck_lock || true
1083 ha_info "---------------8<---------------"
1085 ha_info " Duration: $(($(date +%s) - $ha_start_time))s"
1086 ha_info " Loops: $ha_nr_loops"
1093 while (($(date +%s) < ha_start_time + ha_expected_duration)) &&
1094 [ ! -e "$ha_stop_file" ]; do
1095 ha_info "---------------8<---------------"
1097 $ha_workloads_only || nodes=$(ha_aim)
1099 ha_info "Failing $nodes"
1100 $ha_workloads_only && ha_info " is skipped: workload only..."
1102 ha_sleep $(ha_rand $ha_max_failover_period)
1103 $ha_workloads_only || ha_power_down $nodes
1105 ha_wait_loads || return
1107 if [ -e $ha_stop_file ]; then
1108 $ha_workloads_only || ha_power_up $nodes
1112 ha_info "Bringing $nodes back"
1113 ha_sleep $(ha_rand 10)
1114 $ha_workloads_only ||
1117 ha_wait_nodes $nodes
1122 # Wait for the failback to start.
1125 ha_wait_loads || return
1127 ha_sleep $(ha_rand 20)
1129 ha_nr_loops=$((ha_nr_loops + 1))
1130 ha_info "Loop $ha_nr_loops done"
1137 ha_process_arguments "$@"
1140 ha_log "${ha_clients[*]} ${ha_servers[*]}" \
1141 "START: $0: $(date +%H:%M:%S' '%s)"
1142 trap ha_trap_exit EXIT
1146 if $ha_test_dir_mdt_index_random &&
1147 [ $ha_test_dir_mdt_index -ne 0 ]; then
1148 mdt_index=$(ha_rand $ha_test_dir_mdt_index)
1150 mdt_index=$ha_test_dir_mdt_index
1154 test_dir=${ha_testdirs[0]}
1155 ha_on ${ha_clients[0]} "$LFS mkdir -i$mdt_index \
1156 -c$ha_test_dir_stripe_count $test_dir"
1157 for ((i=0; i<${#ha_testdirs[@]}; i++)); do
1158 test_dir=${ha_testdirs[i]}
1159 ha_on ${ha_clients[0]} $LFS getdirstripe $test_dir
1160 ha_on ${ha_clients[0]} " \
1161 $LFS setstripe $ha_stripe_params $test_dir"
1167 if $ha_workloads_dry_run; then
1171 ha_dump_logs "${ha_clients[*]} ${ha_servers[*]}"
1176 $ha_lfsck_after && ha_start_lfsck | tee -a $ha_lfsck_log
1178 $ha_lfsck_fail_on_repaired && ha_lfsck_repaired
1180 if [ -e "$ha_fail_file" ]; then
1183 ha_log "${ha_clients[*]} ${ha_servers[*]}" \
1184 "END: $0: $(date +%H:%M:%S' '%s)"