- local rc=$?
- local var
- trap 0
-
- # Having not empty END_RUN_FILE means the failed loads only
- if [ -s $END_RUN_FILE ]; then
- echo "Found the END_RUN_FILE file: $END_RUN_FILE"
- cat $END_RUN_FILE
- local END_RUN_NODE=
- read END_RUN_NODE < $END_RUN_FILE
-
- # a client load will end (i.e. fail) if it finds
- # the end run file. that does not mean that that client load
- # actually failed though. the first node in the END_RUN_NODE is
- # the one we are really interested in.
- if [ -n "$END_RUN_NODE" ]; then
- var=$(client_var_name $END_RUN_NODE)_load
- echo "Client load failed on node $END_RUN_NODE"
- echo
- echo "client $END_RUN_NODE load stdout and debug files :
- ${TESTSUITELOG}_run_${!var}.sh-${END_RUN_NODE}
- ${TESTSUITELOG}_run_${!var}.sh-${END_RUN_NODE}.debug"
- fi
- rc=1
- fi
-
- echo $(date +'%F %H:%M:%S') Terminating clients loads ...
- echo "$0" >> $END_RUN_FILE
- local result=PASS
- [ $rc -eq 0 ] || result=FAIL
-
- log "Duration: $DURATION
-Server failover period: $SERVER_FAILOVER_PERIOD seconds
-Exited after: $ELAPSED seconds
-Number of failovers before exit:
-$(server_numfailovers)
-Status: $result: rc=$rc"
-
- # stop the vmstats on the OSTs
- if [ "$VMSTAT" ]; then
- do_nodes $(comma_list $(osts_nodes)) "test -f /tmp/vmstat.pid && \
- { kill -s TERM \$(cat /tmp/vmstat.pid); rm -f /tmp/vmstat.pid; \
- gzip -f9 $vmstatLOG-\$(hostname); }"
- fi
-
- # make sure the client loads die
- do_nodes $NODES_TO_USE "set -x; test -f $LOAD_PID_FILE && \
- { kill -s TERM \$(cat $LOAD_PID_FILE) || true; }"
-
- # and free up the pdshes that started them, if any are still around
- if [ -n "$CLIENT_LOAD_PIDS" ]; then
- kill $CLIENT_LOAD_PIDS || true
- sleep 5
- kill -9 $CLIENT_LOAD_PIDS || true
- fi
- if [ $rc -ne 0 ]; then
- # we are interested in only on failed clients and servers
- local failedclients=$(cat $END_RUN_FILE | grep -v $0)
- # FIXME: need ostfailover-s nodes also for FLAVOR=OST
- local product=$(gather_logs $(comma_list $(osts_nodes) \
- $(mdts_nodes) $mdsfailover_HOST $failedclients))
- echo logs files $product
- fi
-
- [ $rc -eq 0 ] && zconf_mount $(hostname) $MOUNT
-
- exit $rc
+failover_target() {
+ local flavor=${1:-"MDS"}
+ local servers
+ local serverfacet
+ local var
+
+ [ "$flavor" = "MDS" ] && servers=$MDTS || servers=$OSTS
+
+ stack_trap summary_and_cleanup EXIT INT
+
+ # start vmstat on OSS nodes
+ [ "$VMSTAT" ] &&
+ start_vmstat $(comma_list $(osts_nodes)) $VMSTAT_PID_FILE
+
+ # start client loads
+ rm -f $END_RUN_FILE
+ start_client_loads $NODES_TO_USE
+
+ echo client loads pids:
+ do_nodesv $NODES_TO_USE "cat $LOAD_PID_FILE" || exit 3
+
+ ELAPSED=0
+ local sleep=0
+ local reqfail=0
+ local it_time_start
+ local start_ts=$(date +%s)
+ local current_ts=$start_ts
+
+ while [ $ELAPSED -lt $DURATION -a ! -e $END_RUN_FILE ]; do
+ # In order to perform the
+ # expected number of failovers, we need to account the
+ # following:
+ # 1) the time that has elapsed during the client load checking
+ # 2) time takes for failover
+ it_time_start=$(date +%s)
+
+ serverfacet=$(get_random_entry $servers)
+ var=${serverfacet}_numfailovers
+
+ # Check that our client loads are still running. If any have
+ # died, that means they have died outside of recovery, which
+ # is unacceptable.
+ log "==== Check clients loads BEFORE failover -- failure NOT OK \
+ ELAPSED=$ELAPSED DURATION=$DURATION \
+ PERIOD=$SERVER_FAILOVER_PERIOD"
+ check_client_loads $NODES_TO_USE || exit 4
+
+ log "Wait $serverfacet recovery complete before next failover"
+ if ! wait_recovery_complete $serverfacet; then
+ echo "$serverfacet recovery is not completed!"
+ exit 7
+ fi
+
+ local want="FULL\|IDLE"
+ log "Checking clients are in $want state before next failover"
+ wait_clients_import_state $NODES_TO_USE $serverfacet "$want" ||
+ echo "Client import not $want, please consider to" \
+ "increase SERVER_FAILOVER_PERIOD=" \
+ "$SERVER_FAILOVER_PERIOD!"
+
+ log "Starting failover on $serverfacet"
+ facet_failover "$serverfacet" || exit 1
+
+ # Check that our client loads are still running during failover.
+ # No application failures should occur.
+ log "==== Check clients loads AFTER failover -- failure NOT OK"
+ if ! check_client_loads $NODES_TO_USE; then
+ log "Client load failed during failover. Exiting..."
+ exit 5
+ fi
+
+ # Increment the number of failovers.
+ val=$((${!var} + 1))
+ eval $var=$val
+
+ current_ts=$(date +%s)
+ ELAPSED=$((current_ts - start_ts))
+
+ sleep=$((SERVER_FAILOVER_PERIOD - (current_ts - it_time_start)))
+
+ # Keep counting the number of iterations when
+ # time spent to failover and two client loads check exceeded
+ # the value ( SERVER_FAILOVER_PERIOD - MINSLEEP ).
+ if (( $sleep < $MINSLEEP )); then
+ reqfail=$((reqfail + 1))
+
+ cat <<- END_WARN
+ WARNING: failover and two check_client_loads time exceeded: SERVER_FAILOVER_PERIOD - MINSLEEP!
+ Failed loading I/O for a min period of $MINSLEEP $reqfail times (REQFAIL=$REQFAIL)
+ This iteration, load only applied for sleep=$sleep seconds.
+ Estimated max recovery time: $MAX_RECOV_TIME
+ Probably hardware is taking excessively long time to boot.
+ Try increasing SERVER_FAILOVER_PERIOD (current is $SERVER_FAILOVER_PERIOD), bug 20918
+ END_WARN
+ (( $reqfail > $REQFAIL )) && exit 6
+ fi
+
+ log "$serverfacet failed over ${!var} times, and counting..."
+
+ [ $((ELAPSED + sleep)) -ge $DURATION ] && break
+
+ if (( $sleep > 0 )); then
+ echo "sleeping $sleep seconds... "
+ sleep $sleep
+ fi
+ done
+ exit 0