set -x
[ "$SHARED_DIRECTORY" ] || \
- { skip "$0: Empty SHARED_DIRECTORY" && exit 0; }
+ { FAIL_ON_ERROR=true skip_env "$0 Empty SHARED_DIRECTORY" && exit 0; }
+
+[ -n "$CLIENTS" ] || \
+ { FAIL_ON_ERROR=true skip_env "$0 Need two or more remote clients" && exit 0; }
-[ -n "$CLIENTS" ] || { skip "$0 Need two or more remote clients" && exit 0; }
[ $CLIENTCOUNT -ge 3 ] || \
- { skip "$0 Need two or more clients, have $CLIENTCOUNT" && exit 0; }
+ { FAIL_ON_ERROR=true skip_env "$0 Need two or more remote clients, have $((CLIENTCOUNT - 1))" && exit 0; }
END_RUN_FILE=${END_RUN_FILE:-$SHARED_DIRECTORY/end_run_file}
LOAD_PID_FILE=${LOAD_PID_FILE:-$TMP/client-load.pid}
remote_mds_nodsh && skip "remote MDS with nodsh" && exit 0
+[[ $FAILURE_MODE = SOFT ]] && \
+ log "WARNING: $0 is not functional with FAILURE_MODE = SOFT, bz22797"
+
build_test_filter
check_and_setup_lustre
rm -rf $DIR/[df][0-9]*
+max_recov_time=$(max_recovery_time)
+
# the test node needs to be insulated from a lustre failure as much as possible,
# so not even loading the lustre modules is ideal.
# -- umount lustre
# 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 $END_RUN_NODE)_load
+ var=$(node_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 :
local result=PASS
[ $rc -eq 0 ] || result=FAIL
- log "Duraion: $DURATION
+ log "Duration: $DURATION
Server failover period: $SERVER_FAILOVER_PERIOD seconds
Exited after: $ELAPSED seconds
Number of failovers before exit:
start_client_loads $NODES_TO_USE
echo clients load pids:
-if ! do_nodes $NODES_TO_USE "set -x; echo \$(hostname): && cat $LOAD_PID_FILE"; then
- if [ -e $DEBUGLOG ]; then
- exec 2<&-
- cat $DEBUGLOG
+if ! do_nodesv $NODES_TO_USE "cat $LOAD_PID_FILE"; then
exit 3
- fi
fi
START_TS=$(date +%s)
it_time_start=$(date +%s)
FAIL_CLIENT=$(get_random_entry $NODES_TO_USE)
- client_var=$(client_var_name $FAIL_CLIENT)_nums
+ client_var=$(node_var_name $FAIL_CLIENT)_nums
# store the list of failed clients
# lists are comma separated
log "Starting failover on $SERVERFACET"
facet_failover "$SERVERFACET" || exit 1
- if ! wait_recovery_complete $SERVERFACET $((TIMEOUT * 10)); then
+ if ! wait_recovery_complete $SERVERFACET ; then
echo "$SERVERFACET recovery is not completed!"
exit 7
fi
log "WARNING: failover, client reintegration and check_client_loads time exceeded SERVER_FAILOVER_PERIOD - MINSLEEP !
Failed to load the filesystem with I/O for a minimum period of $MINSLEEP $reqfail times ( REQFAIL=$REQFAIL ).
This iteration, the load was only applied for sleep=$sleep seconds.
+Estimated max recovery time : $max_recov_time
Probably the hardware is taking excessively long to boot.
Try to increase SERVER_FAILOVER_PERIOD (current is $SERVER_FAILOVER_PERIOD), bug 20918"
[ $reqfail -gt $REQFAIL ] && exit 6
log " Number of failovers:
$(numfailovers) and counting..."
- if [ $((ELAPSED + sleep)) -gt $DURATION ]; then
+ if [ $((ELAPSED + sleep)) -ge $DURATION ]; then
break
fi