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/$NAME.sh}
18 TESTSUITELOG=${TESTSUITELOG:-$TMP/recovery-mds-scale}
19 DEBUGLOG=$TESTSUITELOG.debug
21 echo "--- env ---" >&2
23 echo "--- env ---" >&2
26 [ "$SHARED_DIRECTORY" ] || \
27 { skip "$0: Empty SHARED_DIRECTORY" && exit 0; }
29 [ -n "$CLIENTS" ] || { skip "$0 Need two or more remote clients" && exit 0; }
30 [ $CLIENTCOUNT -ge 3 ] || \
31 { skip "$0 Need two or more clients, have $CLIENTCOUNT" && exit 0; }
33 END_RUN_FILE=${END_RUN_FILE:-$SHARED_DIRECTORY}/end_run_file}
34 LOAD_PID_FILE=${LOAD_PID_FILE:-$TMP/client-load.pid}
36 remote_mds_nodsh && skip "remote MDS with nodsh" && exit 0
37 remote_ost_nodsh && skip "remote OST with nodsh" && exit 0
41 check_and_setup_lustre
42 rm -rf $DIR/[df][0-9]*
44 # the test node needs to be insulated from a lustre failure as much as possible,
45 # so not even loading the lustre modules is ideal.
47 # -- remove hostname from clients list
48 zconf_umount $(hostname) $MOUNT
49 NODES_TO_USE=${NODES_TO_USE:-$CLIENTS}
50 NODES_TO_USE=$(exclude_item_from_list $NODES_TO_USE $(hostname))
52 check_progs_installed $NODES_TO_USE ${CLIENT_LOADS[@]}
55 for ((i=1; i<=$MDSCOUNT; i++)) do
58 MDTS=$(comma_list $MDTS)
61 for ((i=1; i<=$OSTCOUNT; i++)) do
64 OSTS=$(comma_list $OSTS)
66 ERRORS_OK="" # No application failures should occur during this test.
67 FLAVOR=${FLAVOR:-"MDS"}
71 vmstatLOG=${TESTSUITELOG}_$(basename $0 .sh).vmstat
73 server_numfailovers () {
77 for facet in $MDTS ${OSTS//,/ }; do
81 echo "$facet failed over $val times"
86 summary_and_cleanup () {
92 # Having not empty END_RUN_FILE means the failed loads only
93 if [ -s $END_RUN_FILE ]; then
94 echo "Found the END_RUN_FILE file: $END_RUN_FILE"
97 read END_RUN_NODE < $END_RUN_FILE
99 # a client load will end (i.e. fail) if it finds
100 # the end run file. that does not mean that that client load
101 # actually failed though. the first node in the END_RUN_NODE is
102 # the one we are really interested in.
103 if [ -n "$END_RUN_NODE" ]; then
104 var=${END_RUN_NODE}_load
105 echo "Client load failed on node $END_RUN_NODE"
107 echo "client $END_RUN_NODE load stdout and debug files :
108 ${TESTSUITELOG}_run_${!var}.sh-${END_RUN_NODE}
109 ${TESTSUITELOG}_run_${!var}.sh-${END_RUN_NODE}.debug"
114 echo $(date +'%F %H:%M:%S') Terminating clients loads ...
115 echo "$0" >> $END_RUN_FILE
117 [ $rc -eq 0 ] || result=FAIL
119 log "Duraion: $DURATION
120 Server failover period: $SERVER_FAILOVER_PERIOD seconds
121 Exited after: $ELAPSED seconds
122 Number of failovers before exit:
123 $(server_numfailovers)
124 Status: $result: rc=$rc"
126 # stop the vmstats on the OSTs
127 if [ "$VMSTAT" ]; then
128 do_nodes $(comma_list $(osts_nodes)) "test -f /tmp/vmstat.pid && \
129 { kill -s TERM \$(cat /tmp/vmstat.pid); rm -f /tmp/vmstat.pid; \
130 gzip -f9 $vmstatLOG-\$(hostname); }"
133 # make sure the client loads die
134 do_nodes $NODES_TO_USE "set -x; test -f $LOAD_PID_FILE && \
135 { kill -s TERM \$(cat $LOAD_PID_FILE) || true; }"
137 # and free up the pdshes that started them, if any are still around
138 if [ -n "$CLIENT_LOAD_PIDS" ]; then
139 kill $CLIENT_LOAD_PIDS || true
141 kill -9 $CLIENT_LOAD_PIDS || true
143 [ $rc -eq 0 ] && zconf_mount $(hostname) $MOUNT
151 log "-----============= $0 starting =============-----"
153 trap summary_and_cleanup EXIT INT
155 if [ "$SLOW" = "no" ]; then
156 DURATION=${DURATION:-$((60 * 30))}
158 DURATION=${DURATION:-$((60 * 60 * 24))}
165 if [ "$VMSTAT" ]; then
166 do_nodes $(comma_list $(osts_nodes)) "vmstat 1 > $vmstatLOG-\$(hostname) 2>/dev/null </dev/null & echo \$! > /tmp/vmstat.pid"
169 # Start client loads.
170 start_client_loads $NODES_TO_USE
172 echo clients load pids:
173 if ! do_nodes $NODES_TO_USE "set -x; echo \$(hostname): && cat $LOAD_PID_FILE"; then
174 if [ -e $DEBUGLOG ]; then
184 if [ "$SLOW" = "no" ]; then
185 SERVER_FAILOVER_PERIOD=${SERVER_FAILOVER_PERIOD:-$((60 * 5))}
187 SERVER_FAILOVER_PERIOD=${SERVER_FAILOVER_PERIOD:-$((60 * 10))} # 10 minutes
190 MINSLEEP=${MINSLEEP:-120}
191 REQFAIL_PERCENT=${REQFAIL_PERCENT:-3} # bug17839 comment 62
192 REQFAIL=${REQFAIL:-$(( DURATION / SERVER_FAILOVER_PERIOD * REQFAIL_PERCENT / 100))}
195 while [ $ELAPSED -lt $DURATION -a ! -e $END_RUN_FILE ]; do
197 # In order to perform the
198 # expected number of failovers, we need to account the following :
199 # 1) the time that has elapsed during the client load checking
200 # 2) time takes for failover
202 it_time_start=$(date +%s)
204 SERVERFACET=$(get_random_entry $SERVERS)
205 var=${SERVERFACET}_nums
207 # Check that our client loads are still running. If any have died,
208 # that means they have died outside of recovery, which is unacceptable.
210 log "==== Checking the clients loads BEFORE failover -- failure NOT OK \
211 ELAPSED=$ELAPSED DURATION=$DURATION PERIOD=$SERVER_FAILOVER_PERIOD"
213 if ! check_client_loads $NODES_TO_USE; then
217 log "Starting failover on $SERVERNODE"
219 facet_failover "$SERVERFACET" || exit 1
221 # Check that our client loads are still running during failover.
222 # No application failures should occur.
224 log "==== Checking the clients loads AFTER failover -- failure NOT OK"
225 if ! check_client_loads $NODES_TO_USE; then
226 log "Client load failed during failover. Exiting"
230 # Increment the number of failovers
231 NUM_FAILOVERS=$((NUM_FAILOVERS+1))
235 CURRENT_TS=$(date +%s)
236 ELAPSED=$((CURRENT_TS - START_TS))
238 sleep=$((SERVER_FAILOVER_PERIOD-(CURRENT_TS - it_time_start)))
240 # keep count the number of itterations when
241 # time spend to failover and two client loads check exceeded
242 # the value ( SERVER_FAILOVER_PERIOD - MINSLEEP )
243 if [ $sleep -lt $MINSLEEP ]; then
244 reqfail=$((reqfail +1))
245 log "WARNING: failover and two check_client_loads time exceeded SERVER_FAILOVER_PERIOD - MINSLEEP !
246 Failed to meet interval $reqfail times ( REQFAIL=$REQFAIL ); have sleep=$sleep"
247 [ $reqfail -gt $REQFAIL ] && exit 6
250 log "$SERVERFACET has failed over ${!var} times, and counting..."
251 if [ $sleep -gt 0 ]; then
252 echo "sleeping $sleep seconds ... "