4 # For duration of 24 hours repeatedly failover a random MDS at
5 # 10 minute intervals and verify that no application errors occur.
7 # Test runs one of CLIENT_LOAD progs on remote clients.
9 LUSTRE=${LUSTRE:-`dirname $0`/..}
11 CLEANUP=${CLEANUP:-""}
12 . $LUSTRE/tests/test-framework.sh
16 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
19 DEBUGLOG=$TESTLOG_PREFIX.suite_debug_log.$(hostname -s).log
22 echo "--- env ---" >&2
24 echo "--- env ---" >&2
27 [ "$SHARED_DIRECTORY" ] || \
28 { FAIL_ON_ERROR=true skip_env "$0 Empty SHARED_DIRECTORY" && exit 0; }
30 check_shared_dir $SHARED_DIRECTORY ||
31 error "$SHARED_DIRECTORY isn't a shared directory"
33 [ -n "$CLIENTS" ] || \
34 { FAIL_ON_ERROR=true skip_env "$0 Need two or more remote clients" && exit 0; }
36 [ $CLIENTCOUNT -ge 3 ] || \
37 { FAIL_ON_ERROR=true skip_env "$0 Need two or more remote clients, have $((CLIENTCOUNT - 1))" && exit 0; }
39 END_RUN_FILE=${END_RUN_FILE:-$SHARED_DIRECTORY/end_run_file}
40 LOAD_PID_FILE=${LOAD_PID_FILE:-$TMP/client-load.pid}
41 VMSTAT_PID_FILE=${VMSTAT_PID_FILE:-$TMP/vmstat.pid}
43 remote_mds_nodsh && skip "remote MDS with nodsh" && exit 0
44 remote_ost_nodsh && skip "remote OST with nodsh" && exit 0
48 check_and_setup_lustre
49 rm -rf $DIR/[df][0-9]*
51 max_recov_time=$(max_recovery_time)
53 # the test node needs to be insulated from a lustre failure as much as possible,
54 # so not even loading the lustre modules is ideal.
56 # -- remove hostname from clients list
57 zconf_umount $(hostname) $MOUNT
58 NODES_TO_USE=${NODES_TO_USE:-$CLIENTS}
59 NODES_TO_USE=$(exclude_items_from_list $NODES_TO_USE $(hostname))
61 check_progs_installed $NODES_TO_USE ${CLIENT_LOADS[@]}
63 MDTS=$(get_facets MDS)
64 OSTS=$(get_facets OST)
66 ERRORS_OK="" # No application failures should occur during this test.
67 FLAVOR=${FLAVOR:-"MDS"}
69 if [ "$FLAVOR" == "MDS" ]; then
75 if [ "$SLOW" = "no" ]; then
76 DURATION=${DURATION:-$((60 * 30))}
77 SERVER_FAILOVER_PERIOD=${SERVER_FAILOVER_PERIOD:-$((60 * 5))}
79 DURATION=${DURATION:-$((60 * 60 * 24))}
80 SERVER_FAILOVER_PERIOD=${SERVER_FAILOVER_PERIOD:-$((60 * 10))} # 10 minutes
85 server_numfailovers () {
87 local var=${facet}_numfailovers
90 [[ ${!var} ]] && val=${!var}
94 servers_numfailovers () {
98 for facet in ${MDTS//,/ } ${OSTS//,/ }; do
99 echo "$facet: $(server_numfailovers $facet) times"
103 summary_and_cleanup () {
108 # Having not empty END_RUN_FILE means the failed loads only
109 if [ -s $END_RUN_FILE ]; then
110 echo "Found the END_RUN_FILE file: $END_RUN_FILE"
113 read END_RUN_NODE < $END_RUN_FILE
115 # A client load will stop if it found the END_RUN_FILE file.
116 # That does not mean the client load actually failed though.
117 # The first node in END_RUN_FILE is the one we are interested in.
118 if [ -n "$END_RUN_NODE" ]; then
119 var=$(node_var_name $END_RUN_NODE)_load
120 echo "Client load failed on node $END_RUN_NODE"
122 echo "Client $END_RUN_NODE load stdout and debug files:
123 $TESTLOG_PREFIX.run_${!var}_stdout.$END_RUN_NODE.log
124 $TESTLOG_PREFIX.run_${!var}_debug.$END_RUN_NODE.log"
129 echo $(date +'%F %H:%M:%S') Terminating clients loads ...
130 echo "$0" >> $END_RUN_FILE
132 [ $rc -eq 0 ] || result=FAIL
134 log "Duration: $DURATION
135 Server failover period: $SERVER_FAILOVER_PERIOD seconds
136 Exited after: $ELAPSED seconds
137 Number of failovers before exit:
138 $(servers_numfailovers)
139 Status: $result: rc=$rc"
141 # stop the vmstats on the OSTs
142 if [ "$VMSTAT" ]; then
143 do_nodes $(comma_list $(osts_nodes)) "test -f $VMSTAT_PID_FILE &&
144 { kill -s TERM \\\$(cat $VMSTAT_PID_FILE);
145 rm -f $VMSTAT_PID_FILE || true; }"
148 # make sure the client loads die
149 do_nodes $NODES_TO_USE "set -x; test -f $LOAD_PID_FILE &&
150 { kill -s TERM \\\$(cat $LOAD_PID_FILE);
151 rm -f $LOAD_PID_FILE || true; }"
153 # and free up the pdshes that started them, if any are still around
154 if [ -n "$CLIENT_LOAD_PIDS" ]; then
155 kill $CLIENT_LOAD_PIDS || true
157 kill -9 $CLIENT_LOAD_PIDS || true
160 if [ $rc -ne 0 ]; then
161 # we are interested in only on failed clients and servers
162 local failedclients=$(cat $END_RUN_FILE | grep -v $0)
163 # FIXME: need ostfailover-s nodes also for FLAVOR=OST
164 local product=$(gather_logs $(comma_list $(osts_nodes) \
165 $(mdts_nodes) $mdsfailover_HOST $failedclients) 1)
169 [ $rc -eq 0 ] && zconf_mount $(hostname) $MOUNT
177 log "-----============= $0 starting =============-----"
179 trap summary_and_cleanup EXIT INT
184 if [ "$VMSTAT" ]; then
185 do_nodes $(comma_list $(osts_nodes)) \
186 "vmstat 1 > $TESTLOG_PREFIX.vmstat.\\\$(hostname -s).log \
187 2>/dev/null </dev/null & echo \\\$! > $VMSTAT_PID_FILE"
190 # Start client loads.
191 start_client_loads $NODES_TO_USE
193 echo clients load pids:
194 if ! do_nodesv $NODES_TO_USE "cat $LOAD_PID_FILE"; then
198 MINSLEEP=${MINSLEEP:-120}
199 REQFAIL_PERCENT=${REQFAIL_PERCENT:-3} # bug17839 comment 62
200 REQFAIL=${REQFAIL:-$(( DURATION / SERVER_FAILOVER_PERIOD * REQFAIL_PERCENT / 100))}
207 while [ $ELAPSED -lt $DURATION -a ! -e $END_RUN_FILE ]; do
209 # In order to perform the
210 # expected number of failovers, we need to account the following :
211 # 1) the time that has elapsed during the client load checking
212 # 2) time takes for failover
214 it_time_start=$(date +%s)
216 SERVERFACET=$(get_random_entry $SERVERS)
217 var=${SERVERFACET}_numfailovers
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 "Wait $SERVERFACET recovery complete before doing next failover ...."
231 if ! wait_recovery_complete $SERVERFACET ; then
232 echo "$SERVERFACET recovery is not completed!"
236 log "Checking clients are in FULL state before doing next failover"
237 if ! wait_clients_import_state $NODES_TO_USE $SERVERFACET FULL; then
238 echo "Clients import not FULL, please consider to increase SERVER_FAILOVER_PERIOD=$SERVER_FAILOVER_PERIOD !"
241 log "Starting failover on $SERVERFACET"
243 facet_failover "$SERVERFACET" || exit 1
245 # Check that our client loads are still running during failover.
246 # No application failures should occur.
248 log "==== Checking the clients loads AFTER failover -- failure NOT OK"
249 if ! check_client_loads $NODES_TO_USE; then
250 log "Client load failed during failover. Exiting"
254 # Increment the number of failovers
258 CURRENT_TS=$(date +%s)
259 ELAPSED=$((CURRENT_TS - START_TS))
261 sleep=$((SERVER_FAILOVER_PERIOD-(CURRENT_TS - it_time_start)))
263 # keep count the number of itterations when
264 # time spend to failover and two client loads check exceeded
265 # the value ( SERVER_FAILOVER_PERIOD - MINSLEEP )
266 if [ $sleep -lt $MINSLEEP ]; then
267 reqfail=$((reqfail +1))
268 log "WARNING: failover and two check_client_loads time exceeded SERVER_FAILOVER_PERIOD - MINSLEEP !
269 Failed to load the filesystem with I/O for a minimum period of $MINSLEEP $reqfail times ( REQFAIL=$REQFAIL ).
270 This iteration, the load was only applied for sleep=$sleep seconds.
271 Estimated max recovery time : $max_recov_time
272 Probably the hardware is taking excessively long to boot.
273 Try to increase SERVER_FAILOVER_PERIOD (current is $SERVER_FAILOVER_PERIOD), bug 20918"
274 [ $reqfail -gt $REQFAIL ] && exit 6
277 log "$SERVERFACET has failed over ${!var} times, and counting..."
279 if [ $((ELAPSED + sleep)) -ge $DURATION ]; then
283 if [ $sleep -gt 0 ]; then
284 echo "sleeping $sleep seconds ... "