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_items_from_list $NODES_TO_USE $(hostname))
52 check_progs_installed $NODES_TO_USE ${CLIENT_LOADS[@]}
54 MDTS=$(get_facets MDS)
55 OSTS=$(get_facets OST)
57 ERRORS_OK="" # No application failures should occur during this test.
58 FLAVOR=${FLAVOR:-"MDS"}
60 if [ "$FLAVOR" == "MDS" ]; then
66 if [ "$SLOW" = "no" ]; then
67 DURATION=${DURATION:-$((60 * 30))}
68 SERVER_FAILOVER_PERIOD=${SERVER_FAILOVER_PERIOD:-$((60 * 5))}
70 DURATION=${DURATION:-$((60 * 60 * 24))}
71 SERVER_FAILOVER_PERIOD=${SERVER_FAILOVER_PERIOD:-$((60 * 10))} # 10 minutes
76 vmstatLOG=${TESTSUITELOG}_$(basename $0 .sh).vmstat
78 server_numfailovers () {
82 for facet in $MDTS ${OSTS//,/ }; do
86 echo "$facet failed over $val times"
91 summary_and_cleanup () {
97 # Having not empty END_RUN_FILE means the failed loads only
98 if [ -s $END_RUN_FILE ]; then
99 echo "Found the END_RUN_FILE file: $END_RUN_FILE"
102 read END_RUN_NODE < $END_RUN_FILE
104 # a client load will end (i.e. fail) if it finds
105 # the end run file. that does not mean that that client load
106 # actually failed though. the first node in the END_RUN_NODE is
107 # the one we are really interested in.
108 if [ -n "$END_RUN_NODE" ]; then
109 var=$(client_var_name $END_RUN_NODE)_load
110 echo "Client load failed on node $END_RUN_NODE"
112 echo "client $END_RUN_NODE load stdout and debug files :
113 ${TESTSUITELOG}_run_${!var}.sh-${END_RUN_NODE}
114 ${TESTSUITELOG}_run_${!var}.sh-${END_RUN_NODE}.debug"
119 echo $(date +'%F %H:%M:%S') Terminating clients loads ...
120 echo "$0" >> $END_RUN_FILE
122 [ $rc -eq 0 ] || result=FAIL
124 log "Duraion: $DURATION
125 Server failover period: $SERVER_FAILOVER_PERIOD seconds
126 Exited after: $ELAPSED seconds
127 Number of failovers before exit:
128 $(server_numfailovers)
129 Status: $result: rc=$rc"
131 # stop the vmstats on the OSTs
132 if [ "$VMSTAT" ]; then
133 do_nodes $(comma_list $(osts_nodes)) "test -f /tmp/vmstat.pid && \
134 { kill -s TERM \$(cat /tmp/vmstat.pid); rm -f /tmp/vmstat.pid; \
135 gzip -f9 $vmstatLOG-\$(hostname); }"
138 # make sure the client loads die
139 do_nodes $NODES_TO_USE "set -x; test -f $LOAD_PID_FILE && \
140 { kill -s TERM \$(cat $LOAD_PID_FILE) || true; }"
142 # and free up the pdshes that started them, if any are still around
143 if [ -n "$CLIENT_LOAD_PIDS" ]; then
144 kill $CLIENT_LOAD_PIDS || true
146 kill -9 $CLIENT_LOAD_PIDS || true
148 [ $rc -eq 0 ] && zconf_mount $(hostname) $MOUNT
156 log "-----============= $0 starting =============-----"
158 trap summary_and_cleanup EXIT INT
164 if [ "$VMSTAT" ]; then
165 do_nodes $(comma_list $(osts_nodes)) "vmstat 1 > $vmstatLOG-\$(hostname) 2>/dev/null </dev/null & echo \$! > /tmp/vmstat.pid"
168 # Start client loads.
169 start_client_loads $NODES_TO_USE
171 echo clients load pids:
172 if ! do_nodes $NODES_TO_USE "set -x; echo \$(hostname): && cat $LOAD_PID_FILE"; then
173 if [ -e $DEBUGLOG ]; then
183 MINSLEEP=${MINSLEEP:-120}
184 REQFAIL_PERCENT=${REQFAIL_PERCENT:-3} # bug17839 comment 62
185 REQFAIL=${REQFAIL:-$(( DURATION / SERVER_FAILOVER_PERIOD * REQFAIL_PERCENT / 100))}
188 while [ $ELAPSED -lt $DURATION -a ! -e $END_RUN_FILE ]; do
190 # In order to perform the
191 # expected number of failovers, we need to account the following :
192 # 1) the time that has elapsed during the client load checking
193 # 2) time takes for failover
195 it_time_start=$(date +%s)
197 SERVERFACET=$(get_random_entry $SERVERS)
198 var=${SERVERFACET}_nums
200 # Check that our client loads are still running. If any have died,
201 # that means they have died outside of recovery, which is unacceptable.
203 log "==== Checking the clients loads BEFORE failover -- failure NOT OK \
204 ELAPSED=$ELAPSED DURATION=$DURATION PERIOD=$SERVER_FAILOVER_PERIOD"
206 if ! check_client_loads $NODES_TO_USE; then
210 log "Starting failover on $SERVERFACET"
212 facet_failover "$SERVERFACET" || exit 1
214 # Check that our client loads are still running during failover.
215 # No application failures should occur.
217 log "==== Checking the clients loads AFTER failover -- failure NOT OK"
218 if ! check_client_loads $NODES_TO_USE; then
219 log "Client load failed during failover. Exiting"
223 # Increment the number of failovers
224 NUM_FAILOVERS=$((NUM_FAILOVERS+1))
228 CURRENT_TS=$(date +%s)
229 ELAPSED=$((CURRENT_TS - START_TS))
231 sleep=$((SERVER_FAILOVER_PERIOD-(CURRENT_TS - it_time_start)))
233 # keep count the number of itterations when
234 # time spend to failover and two client loads check exceeded
235 # the value ( SERVER_FAILOVER_PERIOD - MINSLEEP )
236 if [ $sleep -lt $MINSLEEP ]; then
237 reqfail=$((reqfail +1))
238 log "WARNING: failover and two check_client_loads time exceeded SERVER_FAILOVER_PERIOD - MINSLEEP !
239 Failed to meet interval $reqfail times ( REQFAIL=$REQFAIL ); have sleep=$sleep"
240 [ $reqfail -gt $REQFAIL ] && exit 6
243 log "$SERVERFACET has failed over ${!var} times, and counting..."
244 if [ $sleep -gt 0 ]; then
245 echo "sleeping $sleep seconds ... "