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 REQFAIL=${REQFAIL:-$((DURATION / SERVER_FAILOVER_PERIOD *
47 REQFAIL_PERCENT / 100))}
49 END_RUN_FILE=${END_RUN_FILE:-$SHARED_DIRECTORY/end_run_file}
50 LOAD_PID_FILE=${LOAD_PID_FILE:-$TMP/client-load.pid}
51 VMSTAT_PID_FILE=${VMSTAT_PID_FILE:-$TMP/vmstat.pid}
53 server_numfailovers () {
55 local var=${facet}_numfailovers
58 [[ ${!var} ]] && val=${!var}
62 servers_numfailovers () {
66 for facet in ${MDTS//,/ } ${OSTS//,/ }; do
67 echo "$facet: $(server_numfailovers $facet) times"
71 summary_and_cleanup () {
75 # Having not empty END_RUN_FILE means the failed loads only
76 if [ -s $END_RUN_FILE ]; then
77 print_end_run_file $END_RUN_FILE
81 echo $(date +'%F %H:%M:%S') Terminating clients loads ...
82 echo "$0" >> $END_RUN_FILE
84 [ $rc -eq 0 ] || result=FAIL
86 log "Duration: $DURATION
87 Server failover period: $SERVER_FAILOVER_PERIOD seconds
88 Exited after: $ELAPSED seconds
89 Number of failovers before exit:
90 $(servers_numfailovers)
91 Status: $result: rc=$rc"
93 # stop vmstat on OSS nodes
94 [ "$VMSTAT" ] && stop_process $(comma_list $(osts_nodes)) $VMSTAT_PID_FILE
96 # stop the client loads
97 stop_client_loads $NODES_TO_USE $LOAD_PID_FILE
99 if [ $rc -ne 0 ]; then
100 # we are interested in only on failed clients and servers
101 local failedclients=$(cat $END_RUN_FILE | grep -v $0)
102 gather_logs $(comma_list $(all_server_nodes) $failedclients)
109 local flavor=${1:-"MDS"}
114 [ "$flavor" = "MDS" ] && servers=$MDTS || servers=$OSTS
116 trap summary_and_cleanup EXIT INT
118 # start vmstat on OSS nodes
119 [ "$VMSTAT" ] && start_vmstat $(comma_list $(osts_nodes)) $VMSTAT_PID_FILE
123 start_client_loads $NODES_TO_USE
125 echo client loads pids:
126 do_nodesv $NODES_TO_USE "cat $LOAD_PID_FILE" || exit 3
132 local start_ts=$(date +%s)
133 local current_ts=$start_ts
135 while [ $ELAPSED -lt $DURATION -a ! -e $END_RUN_FILE ]; do
136 # In order to perform the
137 # expected number of failovers, we need to account the following:
138 # 1) the time that has elapsed during the client load checking
139 # 2) time takes for failover
140 it_time_start=$(date +%s)
142 serverfacet=$(get_random_entry $servers)
143 var=${serverfacet}_numfailovers
145 # Check that our client loads are still running. If any have died,
146 # that means they have died outside of recovery, which is unacceptable.
147 log "==== Checking the clients loads BEFORE failover -- failure NOT OK \
148 ELAPSED=$ELAPSED DURATION=$DURATION PERIOD=$SERVER_FAILOVER_PERIOD"
149 check_client_loads $NODES_TO_USE || exit 4
151 log "Wait $serverfacet recovery complete before doing next failover..."
152 if ! wait_recovery_complete $serverfacet; then
153 echo "$serverfacet recovery is not completed!"
157 log "Checking clients are in FULL state before doing next failover..."
158 if ! wait_clients_import_state $NODES_TO_USE $serverfacet FULL; then
159 echo "Clients import not FULL, please consider to increase \
160 SERVER_FAILOVER_PERIOD=$SERVER_FAILOVER_PERIOD!"
163 log "Starting failover on $serverfacet"
164 facet_failover "$serverfacet" || exit 1
166 # Check that our client loads are still running during failover.
167 # No application failures should occur.
168 log "==== Checking the clients loads AFTER failover -- failure NOT OK"
169 if ! check_client_loads $NODES_TO_USE; then
170 log "Client load failed during failover. Exiting..."
174 # Increment the number of failovers.
178 current_ts=$(date +%s)
179 ELAPSED=$((current_ts - start_ts))
181 sleep=$((SERVER_FAILOVER_PERIOD - (current_ts - it_time_start)))
183 # Keep counting the number of iterations when
184 # time spent to failover and two client loads check exceeded
185 # the value ( SERVER_FAILOVER_PERIOD - MINSLEEP ).
186 if [ $sleep -lt $MINSLEEP ]; then
187 reqfail=$((reqfail + 1))
188 log "WARNING: failover and two check_client_loads time exceeded \
189 SERVER_FAILOVER_PERIOD - MINSLEEP!
190 Failed to load the filesystem with I/O for a minimum period of \
191 $MINSLEEP $reqfail times ( REQFAIL=$REQFAIL ).
192 This iteration, the load was only applied for sleep=$sleep seconds.
193 Estimated max recovery time: $MAX_RECOV_TIME
194 Probably the hardware is taking excessively long time to boot.
195 Try to increase SERVER_FAILOVER_PERIOD (current is $SERVER_FAILOVER_PERIOD), \
197 [ $reqfail -gt $REQFAIL ] && exit 6
200 log "$serverfacet has failed over ${!var} times, and counting..."
202 [ $((ELAPSED + sleep)) -ge $DURATION ] && break
204 if [ $sleep -gt 0 ]; then
205 echo "sleeping $sleep seconds... "
212 ################################## Main Flow ###################################
215 check_and_setup_lustre
216 rm -rf $DIR/[Rdfs][0-9]*
218 MAX_RECOV_TIME=$(max_recovery_time)
220 # The test node needs to be insulated from a lustre failure as much as possible,
221 # so not even loading the lustre modules is ideal.
223 # -- remove hostname from clients list
224 zconf_umount $HOSTNAME $MOUNT
225 NODES_TO_USE=${NODES_TO_USE:-$CLIENTS}
226 NODES_TO_USE=$(exclude_items_from_list $NODES_TO_USE $HOSTNAME)
228 check_progs_installed $NODES_TO_USE ${CLIENT_LOADS[@]}
230 MDTS=$(get_facets MDS)
231 OSTS=$(get_facets OST)
233 test_failover_mds() {
234 # failover a random MDS
237 run_test failover_mds "failover MDS"
239 test_failover_ost() {
240 # failover a random OST
243 run_test failover_ost "failover OST"
245 zconf_mount $HOSTNAME $MOUNT || error "mount $MOUNT on $HOSTNAME failed"
246 client_up || error "start client on $HOSTNAME failed"
249 check_and_cleanup_lustre