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:-$TMP/$(basename $0 .sh)}
19 DEBUGLOG=$TESTSUITELOG.debug
24 echo "--- env ---" >&2
26 echo "--- env ---" >&2
29 [ "$SHARED_DIRECTORY" ] || \
30 { skip "$0: Empty SHARED_DIRECTORY" && exit 0; }
32 [ -n "$CLIENTS" ] || { skip "$0 Need two or more remote clients" && exit 0; }
33 [ $CLIENTCOUNT -ge 3 ] || \
34 { skip "$0 Need two or more remote clients, have $CLIENTCOUNT" && exit 0; }
36 END_RUN_FILE=${END_RUN_FILE:-$SHARED_DIRECTORY/end_run_file}
37 LOAD_PID_FILE=${LOAD_PID_FILE:-$TMP/client-load.pid}
39 remote_mds_nodsh && skip "remote MDS with nodsh" && exit 0
40 remote_ost_nodsh && skip "remote OST with nodsh" && exit 0
44 check_and_setup_lustre
45 rm -rf $DIR/[df][0-9]*
47 # the test node needs to be insulated from a lustre failure as much as possible,
48 # so not even loading the lustre modules is ideal.
50 # -- remove hostname from clients list
51 zconf_umount $(hostname) $MOUNT
52 NODES_TO_USE=${NODES_TO_USE:-$CLIENTS}
53 NODES_TO_USE=$(exclude_items_from_list $NODES_TO_USE $(hostname))
55 check_progs_installed $NODES_TO_USE ${CLIENT_LOADS[@]}
57 MDTS=$(get_facets MDS)
58 OSTS=$(get_facets OST)
60 ERRORS_OK="" # No application failures should occur during this test.
61 FLAVOR=${FLAVOR:-"MDS"}
63 if [ "$FLAVOR" == "MDS" ]; then
69 if [ "$SLOW" = "no" ]; then
70 DURATION=${DURATION:-$((60 * 30))}
71 SERVER_FAILOVER_PERIOD=${SERVER_FAILOVER_PERIOD:-$((60 * 5))}
73 DURATION=${DURATION:-$((60 * 60 * 24))}
74 SERVER_FAILOVER_PERIOD=${SERVER_FAILOVER_PERIOD:-$((60 * 10))} # 10 minutes
79 vmstatLOG=${TESTSUITELOG}_$(basename $0 .sh).vmstat
81 server_numfailovers () {
85 for facet in $MDTS ${OSTS//,/ }; do
89 echo "$facet failed over $val times"
94 summary_and_cleanup () {
100 # Having not empty END_RUN_FILE means the failed loads only
101 if [ -s $END_RUN_FILE ]; then
102 echo "Found the END_RUN_FILE file: $END_RUN_FILE"
105 read END_RUN_NODE < $END_RUN_FILE
107 # a client load will end (i.e. fail) if it finds
108 # the end run file. that does not mean that that client load
109 # actually failed though. the first node in the END_RUN_NODE is
110 # the one we are really interested in.
111 if [ -n "$END_RUN_NODE" ]; then
112 var=$(client_var_name $END_RUN_NODE)_load
113 echo "Client load failed on node $END_RUN_NODE"
115 echo "client $END_RUN_NODE load stdout and debug files :
116 ${TESTSUITELOG}_run_${!var}.sh-${END_RUN_NODE}
117 ${TESTSUITELOG}_run_${!var}.sh-${END_RUN_NODE}.debug"
122 echo $(date +'%F %H:%M:%S') Terminating clients loads ...
123 echo "$0" >> $END_RUN_FILE
125 [ $rc -eq 0 ] || result=FAIL
127 log "Duraion: $DURATION
128 Server failover period: $SERVER_FAILOVER_PERIOD seconds
129 Exited after: $ELAPSED seconds
130 Number of failovers before exit:
131 $(server_numfailovers)
132 Status: $result: rc=$rc"
134 # stop the vmstats on the OSTs
135 if [ "$VMSTAT" ]; then
136 do_nodes $(comma_list $(osts_nodes)) "test -f /tmp/vmstat.pid && \
137 { kill -s TERM \$(cat /tmp/vmstat.pid); rm -f /tmp/vmstat.pid; \
138 gzip -f9 $vmstatLOG-\$(hostname); }"
141 # make sure the client loads die
142 do_nodes $NODES_TO_USE "set -x; test -f $LOAD_PID_FILE && \
143 { kill -s TERM \$(cat $LOAD_PID_FILE) || true; }"
145 # and free up the pdshes that started them, if any are still around
146 if [ -n "$CLIENT_LOAD_PIDS" ]; then
147 kill $CLIENT_LOAD_PIDS || true
149 kill -9 $CLIENT_LOAD_PIDS || true
151 if [ $rc -ne 0 ]; then
152 # we are interested in only on failed clients and servers
153 local failedclients=$(cat $END_RUN_FILE | grep -v $0)
154 # FIXME: need ostfailover-s nodes also for FLAVOR=OST
155 local product=$(gather_logs $(comma_list $(osts_nodes) \
156 $(mdts_nodes) $mdsfailover_HOST $failedclients))
157 echo logs files $product
160 [ $rc -eq 0 ] && zconf_mount $(hostname) $MOUNT
168 log "-----============= $0 starting =============-----"
170 trap summary_and_cleanup EXIT INT
176 if [ "$VMSTAT" ]; then
177 do_nodes $(comma_list $(osts_nodes)) "vmstat 1 > $vmstatLOG-\$(hostname) 2>/dev/null </dev/null & echo \$! > /tmp/vmstat.pid"
180 # Start client loads.
181 start_client_loads $NODES_TO_USE
183 echo clients load pids:
184 if ! do_nodes $NODES_TO_USE "set -x; echo \$(hostname): && cat $LOAD_PID_FILE"; then
185 if [ -e $DEBUGLOG ]; then
195 MINSLEEP=${MINSLEEP:-120}
196 REQFAIL_PERCENT=${REQFAIL_PERCENT:-3} # bug17839 comment 62
197 REQFAIL=${REQFAIL:-$(( DURATION / SERVER_FAILOVER_PERIOD * REQFAIL_PERCENT / 100))}
200 while [ $ELAPSED -lt $DURATION -a ! -e $END_RUN_FILE ]; do
202 # In order to perform the
203 # expected number of failovers, we need to account the following :
204 # 1) the time that has elapsed during the client load checking
205 # 2) time takes for failover
207 it_time_start=$(date +%s)
209 SERVERFACET=$(get_random_entry $SERVERS)
210 var=${SERVERFACET}_nums
212 # Check that our client loads are still running. If any have died,
213 # that means they have died outside of recovery, which is unacceptable.
215 log "==== Checking the clients loads BEFORE failover -- failure NOT OK \
216 ELAPSED=$ELAPSED DURATION=$DURATION PERIOD=$SERVER_FAILOVER_PERIOD"
218 if ! check_client_loads $NODES_TO_USE; then
222 log "Wait $SERVERFACET recovery complete before doing next failover ...."
223 if [[ $NUM_FAILOVERS != 0 ]]; then
224 if ! wait_recovery_complete $SERVERFACET ; then
225 echo "$SERVERFACET recovery is not completed!"
230 log "Checking clients are in FULL state before doing next failover"
231 if ! wait_clients_import_state $NODES_TO_USE $SERVERFACET FULL; then
232 echo "Clients import not FULL, please consider to increase SERVER_FAILOVER_PERIOD=$SERVER_FAILOVER_PERIOD !"
235 log "Starting failover on $SERVERFACET"
237 facet_failover "$SERVERFACET" || exit 1
239 # Check that our client loads are still running during failover.
240 # No application failures should occur.
242 log "==== Checking the clients loads AFTER failover -- failure NOT OK"
243 if ! check_client_loads $NODES_TO_USE; then
244 log "Client load failed during failover. Exiting"
248 # Increment the number of failovers
249 NUM_FAILOVERS=$((NUM_FAILOVERS+1))
253 CURRENT_TS=$(date +%s)
254 ELAPSED=$((CURRENT_TS - START_TS))
256 sleep=$((SERVER_FAILOVER_PERIOD-(CURRENT_TS - it_time_start)))
258 # keep count the number of itterations when
259 # time spend to failover and two client loads check exceeded
260 # the value ( SERVER_FAILOVER_PERIOD - MINSLEEP )
261 if [ $sleep -lt $MINSLEEP ]; then
262 reqfail=$((reqfail +1))
263 log "WARNING: failover and two check_client_loads time exceeded SERVER_FAILOVER_PERIOD - MINSLEEP !
264 Failed to meet interval $reqfail times ( REQFAIL=$REQFAIL ); have sleep=$sleep"
265 [ $reqfail -gt $REQFAIL ] && exit 6
268 log "$SERVERFACET has failed over ${!var} times, and counting..."
270 if [ $((ELAPSED + sleep)) -gt $DURATION ]; then
274 if [ $sleep -gt 0 ]; then
275 echo "sleeping $sleep seconds ... "