2 # vim:shiftwidth=4:softtabstop=4:tabstop=4:
5 # For duration of 24 hours repeatedly failover a random MDS at
6 # 10 minute intervals and verify that no application errors occur.
8 # Test runs one of CLIENT_LOAD progs on remote clients.
13 # bug number for skipped test:
14 ALWAYS_EXCEPT="$RECOVERY_MDS_SCALE_EXCEPT"
15 # UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
17 LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
18 . $LUSTRE/tests/test-framework.sh
20 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
23 remote_mds_nodsh && skip_env "remote MDS with nodsh" && exit 0
24 remote_ost_nodsh && skip_env "remote OST with nodsh" && exit 0
26 [ -z "$CLIENTS" -o $CLIENTCOUNT -lt 3 ] &&
27 skip_env "need three or more clients" && exit 0
29 if [ -z "$SHARED_DIRECTORY" ] || ! check_shared_dir $SHARED_DIRECTORY; then
30 skip_env "SHARED_DIRECTORY should be specified with a shared directory \
31 which is accessable on all of the nodes"
35 ERRORS_OK="" # No application failures should occur during this test.
37 if [ "$SLOW" = "no" ]; then
38 DURATION=${DURATION:-$((60 * 30))}
40 DURATION=${DURATION:-$((60 * 60 * 24))}
42 SERVER_FAILOVER_PERIOD=${SERVER_FAILOVER_PERIOD:-$((60 * 10))} # 10 minutes
44 MINSLEEP=${MINSLEEP:-120}
45 REQFAIL_PERCENT=${REQFAIL_PERCENT:-3} # bug17839 comment 62
46 # round up the result of integer division: C=(A + (B - 1)) / B
47 REQFAIL=${REQFAIL:-$(((DURATION * REQFAIL_PERCENT + (SERVER_FAILOVER_PERIOD *
48 100 - 1 )) / SERVER_FAILOVER_PERIOD / 100))}
50 END_RUN_FILE=${END_RUN_FILE:-$SHARED_DIRECTORY/end_run_file}
51 LOAD_PID_FILE=${LOAD_PID_FILE:-$TMP/client-load.pid}
52 VMSTAT_PID_FILE=${VMSTAT_PID_FILE:-$TMP/vmstat.pid}
54 server_numfailovers () {
56 local var=${facet}_numfailovers
59 [[ ${!var} ]] && val=${!var}
63 servers_numfailovers () {
67 for facet in ${MDTS//,/ } ${OSTS//,/ }; do
68 echo "$facet: $(server_numfailovers $facet) times"
72 summary_and_cleanup () {
76 # Having not empty END_RUN_FILE means the failed loads only
77 if [ -s $END_RUN_FILE ]; then
78 print_end_run_file $END_RUN_FILE
82 echo $(date +'%F %H:%M:%S') Terminating clients loads ...
83 echo "$0" >> $END_RUN_FILE
85 [ $rc -eq 0 ] || result=FAIL
87 log "Duration: $DURATION
88 Server failover period: $SERVER_FAILOVER_PERIOD seconds
89 Exited after: $ELAPSED seconds
90 Number of failovers before exit:
91 $(servers_numfailovers)
92 Status: $result: rc=$rc"
94 # stop vmstat on OSS nodes
95 [ "$VMSTAT" ] && stop_process $(comma_list $(osts_nodes)) $VMSTAT_PID_FILE
97 # stop the client loads
98 stop_client_loads $NODES_TO_USE $LOAD_PID_FILE
100 if [ $rc -ne 0 ]; then
101 # we are interested in only on failed clients and servers
102 local failedclients=$(cat $END_RUN_FILE | grep -v $0)
103 gather_logs $(comma_list $(all_server_nodes) $failedclients)
110 local flavor=${1:-"MDS"}
115 [ "$flavor" = "MDS" ] && servers=$MDTS || servers=$OSTS
117 trap summary_and_cleanup EXIT INT
119 # start vmstat on OSS nodes
120 [ "$VMSTAT" ] && start_vmstat $(comma_list $(osts_nodes)) $VMSTAT_PID_FILE
124 start_client_loads $NODES_TO_USE
126 echo client loads pids:
127 do_nodesv $NODES_TO_USE "cat $LOAD_PID_FILE" || exit 3
133 local start_ts=$(date +%s)
134 local current_ts=$start_ts
136 while [ $ELAPSED -lt $DURATION -a ! -e $END_RUN_FILE ]; do
137 # In order to perform the
138 # expected number of failovers, we need to account the following:
139 # 1) the time that has elapsed during the client load checking
140 # 2) time takes for failover
141 it_time_start=$(date +%s)
143 serverfacet=$(get_random_entry $servers)
144 var=${serverfacet}_numfailovers
146 # Check that our client loads are still running. If any have died,
147 # that means they have died outside of recovery, which is unacceptable.
148 log "==== Checking the clients loads BEFORE failover -- failure NOT OK \
149 ELAPSED=$ELAPSED DURATION=$DURATION PERIOD=$SERVER_FAILOVER_PERIOD"
150 check_client_loads $NODES_TO_USE || exit 4
152 log "Wait $serverfacet recovery complete before doing next failover..."
153 if ! wait_recovery_complete $serverfacet; then
154 echo "$serverfacet recovery is not completed!"
158 log "Checking clients are in FULL state before doing next failover..."
159 if ! wait_clients_import_state $NODES_TO_USE $serverfacet FULL; then
160 echo "Clients import not FULL, please consider to increase \
161 SERVER_FAILOVER_PERIOD=$SERVER_FAILOVER_PERIOD!"
164 log "Starting failover on $serverfacet"
165 facet_failover "$serverfacet" || exit 1
167 # Check that our client loads are still running during failover.
168 # No application failures should occur.
169 log "==== Checking the clients loads AFTER failover -- failure NOT OK"
170 if ! check_client_loads $NODES_TO_USE; then
171 log "Client load failed during failover. Exiting..."
175 # Increment the number of failovers.
179 current_ts=$(date +%s)
180 ELAPSED=$((current_ts - start_ts))
182 sleep=$((SERVER_FAILOVER_PERIOD - (current_ts - it_time_start)))
184 # Keep counting the number of iterations when
185 # time spent to failover and two client loads check exceeded
186 # the value ( SERVER_FAILOVER_PERIOD - MINSLEEP ).
187 if [ $sleep -lt $MINSLEEP ]; then
188 reqfail=$((reqfail + 1))
189 log "WARNING: failover and two check_client_loads time exceeded \
190 SERVER_FAILOVER_PERIOD - MINSLEEP!
191 Failed to load the filesystem with I/O for a minimum period of \
192 $MINSLEEP $reqfail times ( REQFAIL=$REQFAIL ).
193 This iteration, the load was only applied for sleep=$sleep seconds.
194 Estimated max recovery time: $MAX_RECOV_TIME
195 Probably the hardware is taking excessively long time to boot.
196 Try to increase SERVER_FAILOVER_PERIOD (current is $SERVER_FAILOVER_PERIOD), \
198 [ $reqfail -gt $REQFAIL ] && exit 6
201 log "$serverfacet has failed over ${!var} times, and counting..."
203 [ $((ELAPSED + sleep)) -ge $DURATION ] && break
205 if [ $sleep -gt 0 ]; then
206 echo "sleeping $sleep seconds... "
213 ################################## Main Flow ###################################
216 check_and_setup_lustre
217 rm -rf $DIR/[Rdfs][0-9]*
219 MAX_RECOV_TIME=$(max_recovery_time)
221 # The test node needs to be insulated from a lustre failure as much as possible,
222 # so not even loading the lustre modules is ideal.
224 # -- remove hostname from clients list
225 zconf_umount $HOSTNAME $MOUNT
226 NODES_TO_USE=${NODES_TO_USE:-$CLIENTS}
227 NODES_TO_USE=$(exclude_items_from_list $NODES_TO_USE $HOSTNAME)
229 check_progs_installed $NODES_TO_USE ${CLIENT_LOADS[@]}
231 MDTS=$(get_facets MDS)
232 OSTS=$(get_facets OST)
234 test_failover_mds() {
235 # failover a random MDS
238 run_test failover_mds "failover MDS"
240 test_failover_ost() {
241 # failover a random OST
244 run_test failover_ost "failover OST"
246 zconf_mount $HOSTNAME $MOUNT || error "mount $MOUNT on $HOSTNAME failed"
247 client_up || error "start client on $HOSTNAME failed"
250 check_and_cleanup_lustre