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/$NAME.sh}
19 TESTSUITELOG=${TESTSUITELOG:-$TMP/$(basename $0 .sh)}
20 DEBUGLOG=$TESTSUITELOG.debug
25 echo "--- env ---" >&2
27 echo "--- env ---" >&2
30 [ "$SHARED_DIRECTORY" ] || \
31 { skip "$0: Empty SHARED_DIRECTORY" && exit 0; }
33 [ -n "$CLIENTS" ] || { skip "$0 Need two or more remote clients" && exit 0; }
34 [ $CLIENTCOUNT -ge 3 ] || \
35 { skip "$0 Need two or more remote clients, have $CLIENTCOUNT" && exit 0; }
37 END_RUN_FILE=${END_RUN_FILE:-$SHARED_DIRECTORY/end_run_file}
38 LOAD_PID_FILE=${LOAD_PID_FILE:-$TMP/client-load.pid}
40 remote_mds_nodsh && skip "remote MDS with nodsh" && exit 0
41 remote_ost_nodsh && skip "remote OST with nodsh" && exit 0
45 check_and_setup_lustre
46 rm -rf $DIR/[df][0-9]*
48 # the test node needs to be insulated from a lustre failure as much as possible,
49 # so not even loading the lustre modules is ideal.
51 # -- remove hostname from clients list
52 zconf_umount $(hostname) $MOUNT
53 NODES_TO_USE=${NODES_TO_USE:-$CLIENTS}
54 NODES_TO_USE=$(exclude_items_from_list $NODES_TO_USE $(hostname))
56 check_progs_installed $NODES_TO_USE ${CLIENT_LOADS[@]}
58 MDTS=$(get_facets MDS)
59 OSTS=$(get_facets OST)
61 ERRORS_OK="" # No application failures should occur during this test.
62 FLAVOR=${FLAVOR:-"MDS"}
64 if [ "$FLAVOR" == "MDS" ]; then
70 if [ "$SLOW" = "no" ]; then
71 DURATION=${DURATION:-$((60 * 30))}
72 SERVER_FAILOVER_PERIOD=${SERVER_FAILOVER_PERIOD:-$((60 * 5))}
74 DURATION=${DURATION:-$((60 * 60 * 24))}
75 SERVER_FAILOVER_PERIOD=${SERVER_FAILOVER_PERIOD:-$((60 * 10))} # 10 minutes
80 vmstatLOG=${TESTSUITELOG}_$(basename $0 .sh).vmstat
82 server_numfailovers () {
86 for facet in $MDTS ${OSTS//,/ }; do
90 echo "$facet failed over $val times"
95 summary_and_cleanup () {
101 # Having not empty END_RUN_FILE means the failed loads only
102 if [ -s $END_RUN_FILE ]; then
103 echo "Found the END_RUN_FILE file: $END_RUN_FILE"
106 read END_RUN_NODE < $END_RUN_FILE
108 # a client load will end (i.e. fail) if it finds
109 # the end run file. that does not mean that that client load
110 # actually failed though. the first node in the END_RUN_NODE is
111 # the one we are really interested in.
112 if [ -n "$END_RUN_NODE" ]; then
113 var=$(client_var_name $END_RUN_NODE)_load
114 echo "Client load failed on node $END_RUN_NODE"
116 echo "client $END_RUN_NODE load stdout and debug files :
117 ${TESTSUITELOG}_run_${!var}.sh-${END_RUN_NODE}
118 ${TESTSUITELOG}_run_${!var}.sh-${END_RUN_NODE}.debug"
123 echo $(date +'%F %H:%M:%S') Terminating clients loads ...
124 echo "$0" >> $END_RUN_FILE
126 [ $rc -eq 0 ] || result=FAIL
128 log "Duraion: $DURATION
129 Server failover period: $SERVER_FAILOVER_PERIOD seconds
130 Exited after: $ELAPSED seconds
131 Number of failovers before exit:
132 $(server_numfailovers)
133 Status: $result: rc=$rc"
135 # stop the vmstats on the OSTs
136 if [ "$VMSTAT" ]; then
137 do_nodes $(comma_list $(osts_nodes)) "test -f /tmp/vmstat.pid && \
138 { kill -s TERM \$(cat /tmp/vmstat.pid); rm -f /tmp/vmstat.pid; \
139 gzip -f9 $vmstatLOG-\$(hostname); }"
142 # make sure the client loads die
143 do_nodes $NODES_TO_USE "set -x; test -f $LOAD_PID_FILE && \
144 { kill -s TERM \$(cat $LOAD_PID_FILE) || true; }"
146 # and free up the pdshes that started them, if any are still around
147 if [ -n "$CLIENT_LOAD_PIDS" ]; then
148 kill $CLIENT_LOAD_PIDS || true
150 kill -9 $CLIENT_LOAD_PIDS || true
152 if [ $rc -ne 0 ]; then
153 # we are interested in only on failed clients and servers
154 local failedclients=$(cat $END_RUN_FILE | grep -v $0)
155 # FIXME: need ostfailover-s nodes also for FLAVOR=OST
156 local product=$(gather_logs $(comma_list $(osts_nodes) \
157 $(mdts_nodes) $mdsfailover_HOST $failedclients))
158 echo logs files $product
161 [ $rc -eq 0 ] && zconf_mount $(hostname) $MOUNT
169 log "-----============= $0 starting =============-----"
171 trap summary_and_cleanup EXIT INT
177 if [ "$VMSTAT" ]; then
178 do_nodes $(comma_list $(osts_nodes)) "vmstat 1 > $vmstatLOG-\$(hostname) 2>/dev/null </dev/null & echo \$! > /tmp/vmstat.pid"
181 # Start client loads.
182 start_client_loads $NODES_TO_USE
184 echo clients load pids:
185 if ! do_nodesv $NODES_TO_USE "cat $LOAD_PID_FILE"; then
192 MINSLEEP=${MINSLEEP:-120}
193 REQFAIL_PERCENT=${REQFAIL_PERCENT:-3} # bug17839 comment 62
194 REQFAIL=${REQFAIL:-$(( DURATION / SERVER_FAILOVER_PERIOD * REQFAIL_PERCENT / 100))}
197 while [ $ELAPSED -lt $DURATION -a ! -e $END_RUN_FILE ]; do
199 # In order to perform the
200 # expected number of failovers, we need to account the following :
201 # 1) the time that has elapsed during the client load checking
202 # 2) time takes for failover
204 it_time_start=$(date +%s)
206 SERVERFACET=$(get_random_entry $SERVERS)
207 var=${SERVERFACET}_nums
209 # Check that our client loads are still running. If any have died,
210 # that means they have died outside of recovery, which is unacceptable.
212 log "==== Checking the clients loads BEFORE failover -- failure NOT OK \
213 ELAPSED=$ELAPSED DURATION=$DURATION PERIOD=$SERVER_FAILOVER_PERIOD"
215 if ! check_client_loads $NODES_TO_USE; then
219 log "Wait $SERVERFACET recovery complete before doing next failover ...."
220 if [[ $NUM_FAILOVERS != 0 ]]; then
221 if ! wait_recovery_complete $SERVERFACET ; then
222 echo "$SERVERFACET recovery is not completed!"
227 log "Checking clients are in FULL state before doing next failover"
228 if ! wait_clients_import_state $NODES_TO_USE $SERVERFACET FULL; then
229 echo "Clients import not FULL, please consider to increase SERVER_FAILOVER_PERIOD=$SERVER_FAILOVER_PERIOD !"
232 log "Starting failover on $SERVERFACET"
234 facet_failover "$SERVERFACET" || exit 1
236 # Check that our client loads are still running during failover.
237 # No application failures should occur.
239 log "==== Checking the clients loads AFTER failover -- failure NOT OK"
240 if ! check_client_loads $NODES_TO_USE; then
241 log "Client load failed during failover. Exiting"
245 # Increment the number of failovers
246 NUM_FAILOVERS=$((NUM_FAILOVERS+1))
250 CURRENT_TS=$(date +%s)
251 ELAPSED=$((CURRENT_TS - START_TS))
253 sleep=$((SERVER_FAILOVER_PERIOD-(CURRENT_TS - it_time_start)))
255 # keep count the number of itterations when
256 # time spend to failover and two client loads check exceeded
257 # the value ( SERVER_FAILOVER_PERIOD - MINSLEEP )
258 if [ $sleep -lt $MINSLEEP ]; then
259 reqfail=$((reqfail +1))
260 log "WARNING: failover and two check_client_loads time exceeded SERVER_FAILOVER_PERIOD - MINSLEEP !
261 Failed to load the filesystem with I/O for a minimum period of $MINSLEEP $reqfail times ( REQFAIL=$REQFAIL ).
262 This iteration, the load was only applied for sleep=$sleep seconds.
263 Probably the hardware is taking excessively long to boot.
264 Try to increase SERVER_FAILOVER_PERIOD (current is $SERVER_FAILOVER_PERIOD), bug 20918"
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 ... "