3 # client failure does not affect other clients
5 # Start load on clients (each client works on it's own directory).
6 # At defined (5-10 minutes) interval fail one random client and then fail mds.
7 # Reintegrate failed client after recovery completed,
8 # application errors are allowed for that client but not on other clients.
9 # 10 minute intervals and verify that no application errors occur.
11 # Test runs one of CLIENT_LOAD progs on remote clients.
13 LUSTRE=${LUSTRE:-`dirname $0`/..}
15 CLEANUP=${CLEANUP:-""}
16 . $LUSTRE/tests/test-framework.sh
20 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
23 DEBUGLOG=$TESTLOG_PREFIX.suite_debug_log.$(hostname -s).log
26 echo "--- env ---" >&2
28 echo "--- env ---" >&2
31 [ "$SHARED_DIRECTORY" ] || \
32 { FAIL_ON_ERROR=true skip_env "$0 Empty SHARED_DIRECTORY" && exit 0; }
34 check_shared_dir $SHARED_DIRECTORY ||
35 error "$SHARED_DIRECTORY isn't a shared directory"
37 [ -n "$CLIENTS" ] || \
38 { FAIL_ON_ERROR=true skip_env "$0 Need two or more remote clients" && exit 0; }
40 [ $CLIENTCOUNT -ge 3 ] || \
41 { FAIL_ON_ERROR=true skip_env "$0 Need two or more remote clients, have $((CLIENTCOUNT - 1))" && exit 0; }
43 END_RUN_FILE=${END_RUN_FILE:-$SHARED_DIRECTORY/end_run_file}
44 LOAD_PID_FILE=${LOAD_PID_FILE:-$TMP/client-load.pid}
45 VMSTAT_PID_FILE=${VMSTAT_PID_FILE:-$TMP/vmstat.pid}
47 remote_mds_nodsh && skip "remote MDS with nodsh" && exit 0
49 [[ $FAILURE_MODE = SOFT ]] && \
50 log "WARNING: $0 is not functional with FAILURE_MODE = SOFT, bz22797"
54 check_and_setup_lustre
55 rm -rf $DIR/[df][0-9]*
57 max_recov_time=$(max_recovery_time)
59 # the test node needs to be insulated from a lustre failure as much as possible,
60 # so not even loading the lustre modules is ideal.
62 # -- remove hostname from clients list
63 zconf_umount $(hostname) $MOUNT
64 NODES_TO_USE=${NODES_TO_USE:-$CLIENTS}
65 NODES_TO_USE=$(exclude_items_from_list $NODES_TO_USE $(hostname))
67 check_progs_installed $NODES_TO_USE ${CLIENT_LOADS[@]}
69 MDTS=$(get_facets MDS)
71 if [ "$SLOW" = "no" ]; then
72 DURATION=${DURATION:-$((60 * 30))}
73 SERVER_FAILOVER_PERIOD=${SERVER_FAILOVER_PERIOD:-$((60 * 5))}
75 DURATION=${DURATION:-$((60 * 60 * 24))}
76 SERVER_FAILOVER_PERIOD=${SERVER_FAILOVER_PERIOD:-$((60 * 10))} # 10 minutes
85 for facet in $MDTS ${failed_clients//,/ }; do
89 echo "$facet failed over $val times"
94 summary_and_cleanup () {
99 # Having not empty END_RUN_FILE means the failed loads only
100 if [ -s $END_RUN_FILE ]; then
101 echo "Found the END_RUN_FILE file: $END_RUN_FILE"
104 read END_RUN_NODE < $END_RUN_FILE
106 # A client load will stop if it found the END_RUN_FILE file.
107 # That does not mean the client load actually failed though.
108 # The first node in END_RUN_FILE is the one we are interested in.
109 if [ -n "$END_RUN_NODE" ]; then
110 var=$(node_var_name $END_RUN_NODE)_load
111 echo "Client load failed on node $END_RUN_NODE"
113 echo "Client $END_RUN_NODE load stdout and debug files:
114 $TESTLOG_PREFIX.run_${!var}_stdout.$END_RUN_NODE.log
115 $TESTLOG_PREFIX.run_${!var}_debug.$END_RUN_NODE.log"
120 echo $(date +'%F %H:%M:%S') Terminating clients loads ...
121 echo "$0" >> $END_RUN_FILE
123 [ $rc -eq 0 ] || result=FAIL
125 log "Duration: $DURATION
126 Server failover period: $SERVER_FAILOVER_PERIOD seconds
127 Exited after: $ELAPSED seconds
128 Number of failovers before exit:
130 Status: $result: rc=$rc"
132 # stop the vmstats on the OSTs
133 if [ "$VMSTAT" ]; then
134 do_nodes $(comma_list $(osts_nodes)) "test -f $VMSTAT_PID_FILE &&
135 { kill -s TERM \\\$(cat $VMSTAT_PID_FILE);
136 rm -f $VMSTAT_PID_FILE || true; }"
139 # make sure the client loads die
140 do_nodes $NODES_TO_USE "set -x; test -f $LOAD_PID_FILE &&
141 { kill -s TERM \\\$(cat $LOAD_PID_FILE);
142 rm -f $LOAD_PID_FILE || true; }"
144 # and free up the pdshes that started them, if any are still around
145 if [ -n "$CLIENT_LOAD_PIDS" ]; then
146 kill $CLIENT_LOAD_PIDS || true
148 kill -9 $CLIENT_LOAD_PIDS || true
151 if [ $rc -ne 0 ]; then
152 # we are interested in only on failed clients and servers
153 local failedclients=$(cat $END_RUN_FILE | grep -v $0)
154 # FIXME: need ostfailover-s nodes also for FLAVOR=OST
155 local product=$(gather_logs $(comma_list $(osts_nodes) \
156 $(mdts_nodes) $mdsfailover_HOST $failedclients) 1)
160 [ $rc -eq 0 ] && zconf_mount $(hostname) $MOUNT
168 log "-----============= $0 starting =============-----"
170 trap summary_and_cleanup EXIT # INT
175 if [ "$VMSTAT" ]; then
176 do_nodes $(comma_list $(osts_nodes)) \
177 "vmstat 1 > $TESTLOG_PREFIX.vmstat.\\\$(hostname -s).log \
178 2>/dev/null </dev/null & echo \\\$! > $VMSTAT_PID_FILE"
181 # Start client loads.
182 start_client_loads $NODES_TO_USE
184 echo clients load pids:
185 if ! do_nodesv $NODES_TO_USE "cat $LOAD_PID_FILE"; then
192 MINSLEEP=${MINSLEEP:-120}
193 REQFAIL_PERCENT=${REQFAIL_PERCENT:-3} # bug17839 comment 62
194 REQFAIL=${REQFAIL:-$(( DURATION / SERVER_FAILOVER_PERIOD * REQFAIL_PERCENT / 100))}
198 # This is used for FAIL_CLIENT only
200 while [ $ELAPSED -lt $DURATION -a ! -e $END_RUN_FILE ]; do
202 # In order to perform the
203 # expected number of failovers, we need to account the following :
204 # 1) the time that has elapsed during the client load checking
205 # 2) time takes for failover
207 it_time_start=$(date +%s)
209 FAIL_CLIENT=$(get_random_entry $NODES_TO_USE)
210 client_var=$(node_var_name $FAIL_CLIENT)_nums
212 # store the list of failed clients
213 # lists are comma separated
214 failed_clients=$(expand_list $failed_clients $FAIL_CLIENT)
216 SERVERFACET=$(get_random_entry $MDTS)
217 var=${SERVERFACET}_nums
219 # Check that our client loads are still running. If any have died,
220 # that means they have died outside of recovery, which is unacceptable.
222 log "==== Checking the clients loads BEFORE failover -- failure NOT OK \
223 ELAPSED=$ELAPSED DURATION=$DURATION PERIOD=$SERVER_FAILOVER_PERIOD"
225 if ! check_client_loads $NODES_TO_USE; then
229 log "FAIL CLIENT $FAIL_CLIENT ... "
230 shutdown_client $FAIL_CLIENT
232 log "Starting failover on $SERVERFACET"
234 facet_failover "$SERVERFACET" || exit 1
235 if ! wait_recovery_complete $SERVERFACET ; then
236 echo "$SERVERFACET recovery is not completed!"
240 boot_node $FAIL_CLIENT
241 echo "Reintegrating $FAIL_CLIENT"
242 zconf_mount $FAIL_CLIENT $MOUNT || exit $?
244 # Increment the number of failovers
247 val=$((${!client_var} + 1))
248 eval $client_var=$val
250 # load script on failed clients could create END_RUN_FILE
251 # We shuold remove it and ignore the failure if this
252 # file contains the failed client only.
253 # We can not use ERRORS_OK when start all loads at the start of this script
254 # because the application errors allowed for random failed client only, but
255 # not for all clients.
256 if [ -e $END_RUN_FILE ]; then
257 read END_RUN_NODE < $END_RUN_FILE
258 [[ $END_RUN_NODE = $FAIL_CLIENT ]] &&
259 rm -f $END_RUN_FILE || exit 13
262 restart_client_loads $FAIL_CLIENT $ERRORS_OK || exit $?
264 # Check that not failed clients loads are still running.
265 # No application failures should occur on clients that was not failed.
267 log "==== Checking the clients loads AFTER failed client reintegrated -- failure NOT OK"
268 if ! ERRORS_OK= check_client_loads $(exclude_items_from_list $NODES_TO_USE $FAIL_CLIENT); then
269 log "Client load failed. Exiting"
273 CURRENT_TS=$(date +%s)
274 ELAPSED=$((CURRENT_TS - START_TS))
275 sleep=$((SERVER_FAILOVER_PERIOD-(CURRENT_TS - it_time_start)))
277 # keep count the number of itterations when
278 # time spend to failover and two client loads check exceeded
279 # the value ( SERVER_FAILOVER_PERIOD - MINSLEEP )
280 if [ $sleep -lt $MINSLEEP ]; then
281 reqfail=$((reqfail +1))
282 log "WARNING: failover, client reintegration and check_client_loads time exceeded SERVER_FAILOVER_PERIOD - MINSLEEP !
283 Failed to load the filesystem with I/O for a minimum period of $MINSLEEP $reqfail times ( REQFAIL=$REQFAIL ).
284 This iteration, the load was only applied for sleep=$sleep seconds.
285 Estimated max recovery time : $max_recov_time
286 Probably the hardware is taking excessively long to boot.
287 Try to increase SERVER_FAILOVER_PERIOD (current is $SERVER_FAILOVER_PERIOD), bug 20918"
288 [ $reqfail -gt $REQFAIL ] && exit 6
291 log " Number of failovers:
292 $(numfailovers) and counting..."
294 if [ $((ELAPSED + sleep)) -ge $DURATION ]; then
298 if [ $sleep -gt 0 ]; then
299 echo "sleeping $sleep seconds ... "