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/$NAME.sh}
22 TESTSUITELOG=${TESTSUITELOG:-$TMP/$(basename $0 .sh)}
23 DEBUGLOG=$TESTSUITELOG.debug
28 echo "--- env ---" >&2
30 echo "--- env ---" >&2
33 [ "$SHARED_DIRECTORY" ] || \
34 { FAIL_ON_ERROR=true skip_env "$0 Empty SHARED_DIRECTORY" && exit 0; }
36 [ -n "$CLIENTS" ] || \
37 { FAIL_ON_ERROR=true skip_env "$0 Need two or more remote clients" && exit 0; }
39 [ $CLIENTCOUNT -ge 3 ] || \
40 { FAIL_ON_ERROR=true skip_env "$0 Need two or more remote clients, have $((CLIENTCOUNT - 1))" && exit 0; }
42 END_RUN_FILE=${END_RUN_FILE:-$SHARED_DIRECTORY/end_run_file}
43 LOAD_PID_FILE=${LOAD_PID_FILE:-$TMP/client-load.pid}
45 remote_mds_nodsh && skip "remote MDS with nodsh" && exit 0
47 [[ $FAILURE_MODE = SOFT ]] && \
48 log "WARNING: $0 is not functional with FAILURE_MODE = SOFT, bz22797"
52 check_and_setup_lustre
53 rm -rf $DIR/[df][0-9]*
55 max_recov_time=$(max_recovery_time)
57 # the test node needs to be insulated from a lustre failure as much as possible,
58 # so not even loading the lustre modules is ideal.
60 # -- remove hostname from clients list
61 zconf_umount $(hostname) $MOUNT
62 NODES_TO_USE=${NODES_TO_USE:-$CLIENTS}
63 NODES_TO_USE=$(exclude_items_from_list $NODES_TO_USE $(hostname))
65 check_progs_installed $NODES_TO_USE ${CLIENT_LOADS[@]}
67 MDTS=$(get_facets MDS)
69 if [ "$SLOW" = "no" ]; then
70 DURATION=${DURATION:-$((60 * 30))}
71 SERVER_FAILOVER_PERIOD=${SERVER_FAILOVER_PERIOD:-$((60 * 5))}
73 DURATION=${DURATION:-$((60 * 60 * 24))}
74 SERVER_FAILOVER_PERIOD=${SERVER_FAILOVER_PERIOD:-$((60 * 10))} # 10 minutes
79 vmstatLOG=${TESTSUITELOG}_$(basename $0 .sh).vmstat
85 for facet in $MDTS ${failed_clients//,/ }; do
89 echo "$facet failed over $val times"
94 # list is comma separated
98 do_nodes $list "node=\\\$(hostname)
100 log=${TESTSUITELOG}_run_${!var}.sh-\\\$node.debug
101 if [ -e \\\$log ] ; then
102 echo Node \\\$node debug log:
107 summary_and_cleanup () {
113 # Having not empty END_RUN_FILE means the failed loads only
114 if [ -s $END_RUN_FILE ]; then
115 echo "Found the END_RUN_FILE file: $END_RUN_FILE"
118 read END_RUN_NODE < $END_RUN_FILE
120 # a client load will end (i.e. fail) if it finds
121 # the end run file. that does not mean that that client load
122 # actually failed though. the first node in the END_RUN_NODE is
123 # the one we are really interested in.
124 if [ -n "$END_RUN_NODE" ]; then
125 var=$(node_var_name $END_RUN_NODE)_load
126 echo "Client load failed on node $END_RUN_NODE"
128 echo "client $END_RUN_NODE load stdout and debug files :
129 ${TESTSUITELOG}_run_${!var}.sh-${END_RUN_NODE}
130 ${TESTSUITELOG}_run_${!var}.sh-${END_RUN_NODE}.debug"
136 echo $(date +'%F %H:%M:%S') Terminating clients loads ...
137 echo "$0" >> $END_RUN_FILE
139 [ $rc -eq 0 ] || result=FAIL
141 log "Duration: $DURATION
142 Server failover period: $SERVER_FAILOVER_PERIOD seconds
143 Exited after: $ELAPSED seconds
144 Number of failovers before exit:
146 Status: $result: rc=$rc"
148 # stop the vmstats on the OSTs
149 if [ "$VMSTAT" ]; then
150 do_nodes $(comma_list $(osts_nodes)) "test -f /tmp/vmstat.pid && \
151 { kill -s TERM \$(cat /tmp/vmstat.pid); rm -f /tmp/vmstat.pid; \
152 gzip -f9 $vmstatLOG-\$(hostname); }"
155 # make sure the client loads die
156 do_nodes $NODES_TO_USE "set -x; test -f $LOAD_PID_FILE && \
157 { kill -s TERM \$(cat $LOAD_PID_FILE) || true; }"
159 # and free up the pdshes that started them, if any are still around
160 if [ -n "$CLIENT_LOAD_PIDS" ]; then
161 kill $CLIENT_LOAD_PIDS || true
163 kill -9 $CLIENT_LOAD_PIDS || true
166 if [ $rc -ne 0 ]; then
167 print_logs $NODES_TO_USE
168 # we are interested in only on failed clients and servers
169 local failedclients=$(cat $END_RUN_FILE | grep -v $0)
170 # FIXME: need ostfailover-s nodes also for FLAVOR=OST
171 local product=$(gather_logs $(comma_list $(osts_nodes) \
172 $mds_HOST $mdsfailover_HOST $failedclients))
173 echo logs files $product
176 [ $rc -eq 0 ] && zconf_mount $(hostname) $MOUNT
184 log "-----============= $0 starting =============-----"
186 trap summary_and_cleanup EXIT # INT
191 if [ "$VMSTAT" ]; then
192 do_nodes $(comma_list $(osts_nodes)) "vmstat 1 > $vmstatLOG-\$(hostname) 2>/dev/null </dev/null & echo \$! > /tmp/vmstat.pid"
195 # Start client loads.
196 start_client_loads $NODES_TO_USE
198 echo clients load pids:
199 if ! do_nodesv $NODES_TO_USE "cat $LOAD_PID_FILE"; then
206 MINSLEEP=${MINSLEEP:-120}
207 REQFAIL_PERCENT=${REQFAIL_PERCENT:-3} # bug17839 comment 62
208 REQFAIL=${REQFAIL:-$(( DURATION / SERVER_FAILOVER_PERIOD * REQFAIL_PERCENT / 100))}
212 # This is used for FAIL_CLIENT only
214 while [ $ELAPSED -lt $DURATION -a ! -e $END_RUN_FILE ]; do
216 # In order to perform the
217 # expected number of failovers, we need to account the following :
218 # 1) the time that has elapsed during the client load checking
219 # 2) time takes for failover
221 it_time_start=$(date +%s)
223 FAIL_CLIENT=$(get_random_entry $NODES_TO_USE)
224 client_var=$(node_var_name $FAIL_CLIENT)_nums
226 # store the list of failed clients
227 # lists are comma separated
228 failed_clients=$(expand_list $failed_clients $FAIL_CLIENT)
230 SERVERFACET=$(get_random_entry $MDTS)
231 var=${SERVERFACET}_nums
233 # Check that our client loads are still running. If any have died,
234 # that means they have died outside of recovery, which is unacceptable.
236 log "==== Checking the clients loads BEFORE failover -- failure NOT OK \
237 ELAPSED=$ELAPSED DURATION=$DURATION PERIOD=$SERVER_FAILOVER_PERIOD"
239 if ! check_client_loads $NODES_TO_USE; then
243 log "FAIL CLIENT $FAIL_CLIENT ... "
244 shutdown_client $FAIL_CLIENT
246 log "Starting failover on $SERVERFACET"
248 facet_failover "$SERVERFACET" || exit 1
249 if ! wait_recovery_complete $SERVERFACET ; then
250 echo "$SERVERFACET recovery is not completed!"
254 boot_node $FAIL_CLIENT
255 echo "Reintegrating $FAIL_CLIENT"
256 zconf_mount $FAIL_CLIENT $MOUNT || exit $?
258 # Increment the number of failovers
261 val=$((${!client_var} + 1))
262 eval $client_var=$val
264 # load script on failed clients could create END_RUN_FILE
265 # We shuold remove it and ignore the failure if this
266 # file contains the failed client only.
267 # We can not use ERRORS_OK when start all loads at the start of this script
268 # because the application errors allowed for random failed client only, but
269 # not for all clients.
270 if [ -e $END_RUN_FILE ]; then
271 read END_RUN_NODE < $END_RUN_FILE
272 [[ $END_RUN_NODE = $FAIL_CLIENT ]] &&
273 rm -f $END_RUN_FILE || exit 13
276 restart_client_loads $FAIL_CLIENT $ERRORS_OK || exit $?
278 # Check that not failed clients loads are still running.
279 # No application failures should occur on clients that was not failed.
281 log "==== Checking the clients loads AFTER failed client reintegrated -- failure NOT OK"
282 if ! ERRORS_OK= check_client_loads $(exclude_items_from_list $NODES_TO_USE $FAIL_CLIENT); then
283 log "Client load failed. Exiting"
287 CURRENT_TS=$(date +%s)
288 ELAPSED=$((CURRENT_TS - START_TS))
290 sleep=$((SERVER_FAILOVER_PERIOD-(CURRENT_TS - it_time_start)))
292 # keep count the number of itterations when
293 # time spend to failover and two client loads check exceeded
294 # the value ( SERVER_FAILOVER_PERIOD - MINSLEEP )
295 if [ $sleep -lt $MINSLEEP ]; then
296 reqfail=$((reqfail +1))
297 log "WARNING: failover, client reintegration and check_client_loads time exceeded SERVER_FAILOVER_PERIOD - MINSLEEP !
298 Failed to load the filesystem with I/O for a minimum period of $MINSLEEP $reqfail times ( REQFAIL=$REQFAIL ).
299 This iteration, the load was only applied for sleep=$sleep seconds.
300 Estimated max recovery time : $max_recov_time
301 Probably the hardware is taking excessively long to boot.
302 Try to increase SERVER_FAILOVER_PERIOD (current is $SERVER_FAILOVER_PERIOD), bug 20918"
303 [ $reqfail -gt $REQFAIL ] && exit 6
306 log " Number of failovers:
307 $(numfailovers) and counting..."
309 if [ $((ELAPSED + sleep)) -ge $DURATION ]; then
313 if [ $sleep -gt 0 ]; then
314 echo "sleeping $sleep seconds ... "