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)-$$
175 declare ha_stop_file=$ha_tmp_dir/stop
176 declare ha_fail_file=$ha_tmp_dir/fail
177 declare ha_status_file_prefix=$ha_tmp_dir/status
178 declare -a ha_status_files
179 declare ha_machine_file=$ha_tmp_dir/machine_file
180 declare ha_lfsck_log=$ha_tmp_dir/lfsck.log
181 declare ha_lfsck_lock=$ha_tmp_dir/lfsck.lock
182 declare ha_lfsck_stop=$ha_tmp_dir/lfsck.stop
183 declare ha_lfsck_bg=${LFSCK_BG:-false}
184 declare ha_lfsck_after=${LFSCK_AFTER:-false}
185 declare ha_lfsck_node=${LFSCK_NODE:-""}
186 declare ha_lfsck_device=${LFSCK_DEV:-""}
187 declare ha_lfsck_types=${LFSCK_TYPES:-"namespace layout"}
188 declare ha_lfsck_custom_params=${LFSCK_CUSTOM_PARAMS:-""}
189 declare ha_lfsck_wait=${LFSCK_WAIT:-1200}
190 declare ha_lfsck_fail_on_repaired=${LFSCK_FAIL_ON_REPAIRED:-false}
191 declare ha_power_down_cmd=${POWER_DOWN:-"pm -0"}
192 declare ha_power_up_cmd=${POWER_UP:-"pm -1"}
193 declare ha_power_delay=${POWER_DELAY:-60}
194 declare ha_failback_delay=${DELAY:-5}
195 declare ha_failback_cmd=${FAILBACK:-""}
196 declare ha_stripe_params=${STRIPEPARAMS:-"-c 0"}
197 declare ha_dir_stripe_count=${DSTRIPECOUNT:-"1"}
198 declare ha_mdt_index=${MDTINDEX:-"0"}
199 declare ha_mdt_index_random=${MDTINDEXRAND:-false}
200 declare -a ha_clients
201 declare -a ha_servers
202 declare -a ha_victims
203 declare ha_test_dir=/mnt/lustre/$(basename $0)-$$
204 declare ha_start_time=$(date +%s)
205 declare ha_expected_duration=$((60 * 60 * 24))
206 declare ha_max_failover_period=10
207 declare ha_nr_loops=0
208 declare ha_stop_signals="SIGINT SIGTERM SIGHUP"
209 declare ha_load_timeout=$((60 * 10))
210 declare ha_workloads_only=false
211 declare ha_workloads_dry_run=false
212 declare ha_simultaneous=false
214 declare ha_mpi_instances=${ha_mpi_instances:-1}
216 declare ha_mpi_loads=${ha_mpi_loads="ior simul mdtest"}
217 declare -a ha_mpi_load_tags=($ha_mpi_loads)
218 declare -a ha_mpiusers=(${ha_mpi_users="mpiuser"})
220 declare ha_ior_params=${IORP:-'" -b $ior_blockSize -t 2m -w -W -T 1"'}
221 declare ha_simul_params=${SIMULP:-'" -n 10"'}
222 declare ha_mdtest_params=${MDTESTP:-'" -i 1 -n 1000"'}
223 declare ha_mpirun_options=${MPIRUN_OPTIONS:-""}
224 declare ha_clients_stripe=${CLIENTSSTRIPE:-'"$STRIPEPARAMS"'}
225 declare ha_nclientsset=${NCLIENTSSET:-1}
227 declare ha_racer_params=${RACERP:-"MDSCOUNT=1"}
229 eval ha_params_ior=($ha_ior_params)
230 eval ha_params_simul=($ha_simul_params)
231 eval ha_params_mdtest=($ha_mdtest_params)
232 eval ha_stripe_clients=($ha_clients_stripe)
234 declare ha_nparams_ior=${#ha_params_ior[@]}
235 declare ha_nparams_simul=${#ha_params_simul[@]}
236 declare ha_nparams_mdtest=${#ha_params_mdtest[@]}
237 declare ha_nstripe_clients=${#ha_stripe_clients[@]}
239 declare -A ha_mpi_load_cmds=(
240 [ior]="$IOR -o {}/f.ior {params}"
241 [simul]="$SIMUL {params} -d {}"
242 [mdtest]="$MDTEST {params} -d {}"
245 declare racer=${RACER:-"$(dirname $0)/racer/racer.sh"}
247 declare ha_nonmpi_loads=${ha_nonmpi_loads="dd tar iozone"}
248 declare -a ha_nonmpi_load_tags=($ha_nonmpi_loads)
249 declare -A ha_nonmpi_load_cmds=(
250 [dd]="dd if=/dev/zero of={}/f.dd bs=1M count=256"
251 [tar]="tar cf - /etc | tar xf - -C {}"
252 [iozone]="iozone -a -e -+d -s $iozone_SIZE {}/f.iozone"
253 [racer]="$ha_racer_params $racer {}"
258 ha_info "Usage: $0 -c HOST[,...] -s HOST[,...]" \
259 "-v HOST[,...] [-d DIRECTORY] [-u SECONDS]"
262 ha_process_arguments()
266 while getopts hc:s:v:d:p:u:wrm opt; do
273 ha_clients=(${OPTARG//,/ })
276 ha_servers=(${OPTARG//,/ })
279 ha_victims=(${OPTARG//,/ })
282 ha_test_dir=$OPTARG/$(basename $0)-$$
285 ha_expected_duration=$OPTARG
288 ha_max_failover_period=$OPTARG
291 ha_workloads_only=true
294 ha_workloads_dry_run=true
306 if [ -z "${ha_clients[*]}" ]; then
307 ha_error "-c is mandatory"
311 if ! ($ha_workloads_dry_run ||
312 $ha_workloads_only) &&
313 ([ -z "${ha_servers[*]}" ] ||
314 [ -z "${ha_victims[*]}" ]); then
315 ha_error "-s, and -v are all mandatory"
329 # -S is to be used here to track the
330 # remote command return values
332 pdsh -S -w $nodes PATH=/usr/local/sbin:/usr/local/bin:/sbin:\
333 /bin:/usr/sbin:/usr/bin "$@" ||
340 touch "$ha_stop_file"
342 if [ -e "$ha_fail_file" ]; then
343 ha_info "Test directory $ha_test_dir not removed"
344 ha_info "Temporary directory $ha_tmp_dir not removed"
346 ha_on ${ha_clients[0]} rm -rf "$ha_test_dir"
347 ha_info "Please find the results in the directory $ha_tmp_dir"
351 ha_trap_stop_signals()
353 ha_info "${ha_stop_signals// /,} received"
354 touch "$ha_stop_file"
361 ha_info "Sleeping for ${n}s"
363 # sleep(1) could interrupted.
372 while [ -e $lock ]; do
381 until mkdir "$lock" >/dev/null 2>&1; do
382 ha_sleep 1 >/dev/null
395 local nodes=${1// /,}
396 local file=/tmp/$(basename $0)-$$-$(date +%s).dk
397 local lock=$ha_tmp_dir/lock-dump-logs
401 ha_info "Dumping lctl log to $file"
404 # some nodes could crash, so
405 # do not exit with error if not all logs are dumped
407 ha_on $nodes "lctl dk >>$file" || rc=$?
410 ha_error "not all logs are dumped! Some nodes are unreachable."
421 local stripeparams=$6
423 local tag=${ha_mpi_load_tags[$load]}
424 local cmd=${ha_mpi_load_cmds[$tag]}
425 local dir=$ha_test_dir/$client-$tag
426 local log=$ha_tmp_dir/$client-$tag
429 local avg_loop_time=0
430 local start_time=$(date +%s)
432 cmd=${cmd//"{}"/$dir}
433 cmd=${cmd//"{params}"/$parameter}
435 ha_info "Starting $tag"
437 machines="-machinefile $machines"
438 while [ ! -e "$ha_stop_file" ] && ((rc == 0)); do
441 if $ha_mdt_index_random && [ $ha_mdt_index -ne 0 ]; then
442 mdt_index=$(ha_rand $ha_mdt_index)
444 mdt_index=$ha_mdt_index
446 ha_on $client $LFS mkdir -i$mdt_index -c$ha_dir_stripe_count "$dir" &&
447 ha_on $client $LFS getdirstripe "$dir" &&
448 ha_on $client $LFS setstripe $stripeparams $dir &&
449 ha_on $client $LFS getstripe $dir &&
450 ha_on $client chmod a+xwr $dir &&
451 ha_on $client "su $mpiuser sh -c \" $mpirun $ha_mpirun_options \
452 -np $((${#ha_clients[@]} * mpi_threads_per_client )) \
453 $machines $cmd \" " &&
454 ha_on $client rm -rf "$dir";
455 } >>"$log" 2>&1 || rc=$?
460 touch "$ha_fail_file"
461 touch "$ha_stop_file"
462 ha_dump_logs "${ha_clients[*]} ${ha_servers[*]}"
466 nr_loops=$((nr_loops + 1))
469 [ $nr_loops -ne 0 ] &&
470 avg_loop_time=$((($(date +%s) - start_time) / nr_loops))
472 ha_info "$tag stopped: rc $rc avg loop time $avg_loop_time"
488 # ha_mpi_instances defines the number of
489 # clients start mpi loads; should be <= ${#ha_clients[@]}
491 # ha_mpi_instances = 0
493 # ${#ha_mpi_load_tags[@]} =0
494 local inst=$ha_mpi_instances
495 (( inst == 0 )) || (( ${#ha_mpi_load_tags[@]} == 0 )) &&
496 ha_info "no mpi load to start" &&
499 (( inst <= ${#ha_clients[@]} )) || inst=${#ha_clients[@]}
501 # Define names for machinefiles for each client set
502 for (( n=0; n < $ha_nclientsset; n++ )); do
503 mach[$n]=$ha_machine_file$n
506 for ((n = 0; n < ${#ha_clients[@]}; n++)); do
507 m=$(( n % ha_nclientsset))
509 ha_info machine_file=$machines
510 echo ${ha_clients[n]} >> $machines
512 local dirname=$(dirname $ha_machine_file)
513 for client in ${ha_clients[@]}; do
514 ha_on $client mkdir -p $dirname
515 scp $ha_machine_file* $client:$dirname
518 for ((n = 0; n < $inst; n++)); do
519 client=${ha_clients[n]}
520 mpiuser=${ha_mpiusers[$((n % ${#ha_mpiusers[@]}))]}
521 for ((load = 0; load < ${#ha_mpi_load_tags[@]}; load++)); do
522 tag=${ha_mpi_load_tags[$load]}
523 status=$ha_status_file_prefix-$tag-$client
526 local num=ha_nparams_$tag
528 local aref=ha_params_$tag[nparam]
529 local parameter=${!aref}
530 local nstripe=$((n % ha_nstripe_clients))
531 aref=ha_stripe_clients[nstripe]
532 local stripe=${!aref}
533 local m=$(( n % ha_nclientsset))
535 ha_repeat_mpi_load $client $load $status "$parameter" $machines "$stripe" "$mpiuser" &
536 ha_status_files+=("$status")
541 ha_repeat_nonmpi_load()
546 local tag=${ha_nonmpi_load_tags[$load]}
547 local cmd=${ha_nonmpi_load_cmds[$tag]}
548 local dir=$ha_test_dir/$client-$tag
549 local log=$ha_tmp_dir/$client-$tag
552 local avg_loop_time=0
553 local start_time=$(date +%s)
555 cmd=${cmd//"{}"/$dir}
557 ha_info "Starting $tag on $client"
559 while [ ! -e "$ha_stop_file" ] && ((rc == 0)); do
560 ha_on $client "mkdir -p $dir && \
562 rm -rf $dir" >>"$log" 2>&1 || rc=$?
565 ha_dump_logs "${ha_clients[*]} ${ha_servers[*]}"
566 touch "$ha_fail_file"
567 touch "$ha_stop_file"
571 nr_loops=$((nr_loops + 1))
574 [ $nr_loops -ne 0 ] &&
575 avg_loop_time=$((($(date +%s) - start_time) / nr_loops))
577 ha_info "$tag on $client stopped: rc $rc avg loop time ${avg_loop_time}s"
580 ha_start_nonmpi_loads()
587 for client in ${ha_clients[@]}; do
588 for ((load = 0; load < ${#ha_nonmpi_load_tags[@]}; load++)); do
589 tag=${ha_nonmpi_load_tags[$load]}
590 status=$ha_status_file_prefix-$tag-$client
591 ha_repeat_nonmpi_load $client $load $status &
592 ha_status_files+=("$status")
603 [ -f $ha_lfsck_stop ] && ha_info "LFSCK stopped" && break
604 [ -f $ha_stop_file ] &&
605 ha_info "$ha_stop_file found! LFSCK not started" &&
607 ha_start_lfsck 2>&1 | tee -a $ha_lfsck_log
611 ha_info LFSCK BG PID: $LFSCK_BG_PID
614 ha_wait_lfsck_completed () {
616 local -a types=($ha_lfsck_types)
620 local nodes="${ha_servers[@]}"
623 # -A start LFSCK on all nodes
625 [ ${#types[@]} -eq 0 ] && types=(namespace layout)
626 ha_info "Waiting LFSCK completed in $ha_lfsck_wait sec: types ${types[@]}"
627 for type in ${types[@]}; do
629 for (( i=0; i<=ha_lfsck_wait; i++)); do
630 status=($(ha_on $nodes lctl get_param -n *.*.lfsck_$type 2>/dev/null | \
631 awk '/status/ { print $3 }'))
632 for (( s=0; s<${#status[@]}; s++ )); do
633 # "partial" is expected after HARD failover
634 [[ "${status[s]}" = "completed" ]] ||
635 [[ "${status[s]}" = "partial" ]] || break
637 [[ $s -eq ${#status[@]} ]] && eval var_$type=1 && break
640 ha_info "LFSCK $type status in $i sec:"
641 ha_on $nodes lctl get_param -n *.*.lfsck_$type 2>/dev/null | grep status
645 for type in ${types[@]}; do
647 ha_on $nodes lctl get_param -n *.*.lfsck_$type 2>/dev/null
648 [[ ${!var} -eq 1 ]] ||
649 { ha_info "lfsck not completed in $ha_lfsck_wait sec";
657 local -a types=($ha_lfsck_types)
660 # -A: start LFSCK on all nodes via the specified MDT device
661 # (see "-M" option) by single LFSCK command
662 local params=" -A -r $ha_lfsck_custom_params"
664 # use specified device if set
665 [ -n "$ha_lfsck_device" ] && params="-M $ha_lfsck_device $params"
667 # -t: check type(s) to be performed (default all)
668 # check only specified types if set
669 if [ ${#types[@]} -ne 0 ]; then
670 local type="${types[@]}"
671 params="$params -t ${type// /,}"
674 ha_info "LFSCK start $params"
675 ha_on $ha_lfsck_node "lctl lfsck_start $params" || rc=1
676 if [ $rc -ne 0 ]; then
677 if [ -e $ha_lfsck_lock ]; then
679 ha_wait_unlock $ha_lfsck_lock
681 ha_on $ha_lfsck_node "lctl lfsck_start $params" || rc=1
686 { touch "$ha_fail_file"; touch "$ha_stop_file";
687 touch $ha_lfsck_stop; return 1; }
689 ha_wait_lfsck_completed ||
690 { touch "$ha_fail_file"; touch "$ha_stop_file";
691 touch $ha_lfsck_stop; return 1; }
700 n=$(cat $ha_lfsck_log | awk '/repaired/ {print $3}' |\
701 awk '{sum += $1} END { print sum }')
703 { ha_info "Total repaired: $n";
704 touch "$ha_fail_file"; return 1; }
710 $ha_lfsck_bg && ha_lfsck_bg
711 trap ha_trap_stop_signals $ha_stop_signals
712 ha_start_nonmpi_loads
719 # true because of lfsck_bg could be stopped already
720 $ha_lfsck_bg && wait $LFSCK_BG_PID || true
721 trap - $ha_stop_signals
722 ha_info "Waiting for workloads to stop"
729 local end=$(($(date +%s) + ha_load_timeout))
731 ha_info "Waiting for workload status"
732 rm -f "${ha_status_files[@]}"
735 # return immediately if ha_stop_file exists,
736 # all status_files not needed to be checked
738 for file in "${ha_status_files[@]}"; do
739 if [ -e "$ha_stop_file" ]; then
740 ha_info "$ha_stop_file found! Stop."
744 # Wait status file created during ha_load_timeout.
745 # Existing file guarantees that some application
746 # is completed. If no status file was created
747 # this function guarantees that we allow
748 # applications to continue after/before
749 # failover/failback during ha_load_timeout time.
751 until [ -e "$file" ] || (($(date +%s) >= end)); do
753 # check ha_stop_file again, it could appear
754 # during ha_load_timeout
756 if [ -e "$ha_stop_file" ]; then
757 ha_info "$ha_stop_file found! Stop."
760 ha_sleep 1 >/dev/null
771 if $ha_lfsck_bg && [[ ${nodes//,/ /} =~ $ha_lfsck_node ]]; then
772 ha_info "$ha_lfsck_node down, delay start LFSCK"
773 ha_lock $ha_lfsck_lock
776 ha_info "Powering down $nodes"
777 for i in $(seq 1 5); do
778 $ha_power_down_cmd $nodes && rc=0 && break
779 sleep $ha_power_delay
782 [ $rc -eq 0 ] || ha_info "Failed Powering down in $i attempts"
791 ha_info "Powering up $nodes"
792 for i in $(seq 1 5); do
793 $ha_power_up_cmd $nodes && rc=0 && break
794 sleep $ha_power_delay
797 [ $rc -eq 0 ] || ha_info "Failed Powering up in $i attempts"
803 # Print a random integer within [0, MAX).
810 # See "5.2 Bash Variables" from "info bash".
812 echo -n $((RANDOM * max / 32768))
820 if $ha_simultaneous ; then
821 nodes=$(echo ${ha_victims[@]})
824 i=$(ha_rand ${#ha_victims[@]})
825 nodes=${ha_victims[$i]}
834 local end=$(($(date +%s) + 10 * 60))
836 ha_info "Waiting for $nodes to boot up"
837 until ha_on $nodes hostname >/dev/null 2>&1 ||
838 [ -e "$ha_stop_file" ] ||
839 (($(date +%s) >= end)); do
840 ha_sleep 1 >/dev/null
847 ha_info "Failback resources on $nodes in $ha_failback_delay sec"
849 ha_sleep $ha_failback_delay
850 [ "$ha_failback_cmd" ] ||
852 ha_info "No failback command set, skiping"
856 $ha_failback_cmd $nodes
857 [ -e $ha_lfsck_lock ] && ha_unlock $ha_lfsck_lock || true
862 ha_info "---------------8<---------------"
864 ha_info " Duration: $(($(date +%s) - $ha_start_time))s"
865 ha_info " Loops: $ha_nr_loops"
872 while (($(date +%s) < ha_start_time + ha_expected_duration)) &&
873 [ ! -e "$ha_stop_file" ]; do
874 ha_info "---------------8<---------------"
876 $ha_workloads_only || nodes=$(ha_aim)
878 ha_info "Failing $nodes"
879 $ha_workloads_only && ha_info " is skipped: workload only..."
881 ha_sleep $(ha_rand $ha_max_failover_period)
882 $ha_workloads_only || ha_power_down $nodes
884 ha_wait_loads || return
886 if [ -e $ha_stop_file ]; then
887 $ha_workloads_only || ha_power_up $nodes
891 ha_info "Bringing $nodes back"
892 ha_sleep $(ha_rand 10)
893 $ha_workloads_only ||
901 # Wait for the failback to start.
904 ha_wait_loads || return
906 ha_sleep $(ha_rand 20)
908 ha_nr_loops=$((ha_nr_loops + 1))
909 ha_info "Loop $ha_nr_loops done"
916 ha_process_arguments "$@"
919 ha_log "${ha_clients[*]} ${ha_servers[*]}" \
920 "START: $0: $(date +%H:%M:%S' '%s)"
921 trap ha_trap_exit EXIT
923 ha_on ${ha_clients[0]} mkdir "$ha_test_dir"
924 ha_on ${ha_clients[0]} " \
925 $LFS setstripe $ha_stripe_params $ha_test_dir"
930 if $ha_workloads_dry_run; then
934 ha_dump_logs "${ha_clients[*]} ${ha_servers[*]}"
939 $ha_lfsck_after && ha_start_lfsck | tee -a $ha_lfsck_log
941 $ha_lfsck_fail_on_repaired && ha_lfsck_repaired
943 if [ -e "$ha_fail_file" ]; then
946 ha_log "${ha_clients[*]} ${ha_servers[*]}" \
947 "END: $0: $(date +%H:%M:%S' '%s)"