init_test_env $@
. ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
+init_logging
+
+DEBUGLOG=$TESTLOG_PREFIX.suite_debug_log.$(hostname -s).log
-TESTSUITELOG=${TESTSUITELOG:-$TMP/recovery-random-scale}
-DEBUGLOG=$TESTSUITELOG.debug
exec 2>$DEBUGLOG
echo "--- env ---" >&2
env >&2
set -x
[ "$SHARED_DIRECTORY" ] || \
- { skip "$0: Empty SHARED_DIRECTORY" && exit 0; }
+ { FAIL_ON_ERROR=true skip_env "$0 Empty SHARED_DIRECTORY" && exit 0; }
+
+check_shared_dir $SHARED_DIRECTORY ||
+ error "$SHARED_DIRECTORY isn't a shared directory"
+
+[ -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:-$TMP/client-load.pid}
+VMSTAT_PID_FILE=${VMSTAT_PID_FILE:-$TMP/vmstat.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
rm -f $END_RUN_FILE
-vmstatLOG=${TESTSUITELOG}_$(basename $0 .sh).vmstat
-
numfailovers () {
local facet
local var
done
}
-# list is comma separated
-print_logs () {
- local list=$1
-
- do_nodes $list "node=\\\$(hostname)
-var=\\\${node}_load
-log=${TESTSUITELOG}_run_${!var}.sh-\\\$node.debug
-if [ -e \\\$log ] ; then
-echo Node \\\$node debug log:
-cat \\\$log
-fi"
-}
-
summary_and_cleanup () {
-
local rc=$?
local var
trap 0
local END_RUN_NODE=
read END_RUN_NODE < $END_RUN_FILE
- # a client load will end (i.e. fail) if it finds
- # the end run file. that does not mean that that client load
- # actually failed though. the first node in the END_RUN_NODE is
- # the one we are really interested in.
+ # A client load will stop if it found the END_RUN_FILE file.
+ # That does not mean the client load actually failed though.
+ # The first node in END_RUN_FILE is the one we are interested in.
if [ -n "$END_RUN_NODE" ]; then
- var=${END_RUN_NODE}_load
- echo "Client load failed on node $END_RUN_NODE"
+ var=$(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 :
- ${TESTSUITELOG}_run_${!var}.sh-${END_RUN_NODE}
- ${TESTSUITELOG}_run_${!var}.sh-${END_RUN_NODE}.debug"
+ echo "Client $END_RUN_NODE load stdout and debug files:
+ $TESTLOG_PREFIX.run_${!var}_stdout.$END_RUN_NODE.log
+ $TESTLOG_PREFIX.run_${!var}_debug.$END_RUN_NODE.log"
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 "Duraion: $DURATION
+ log "Duration: $DURATION
Server failover period: $SERVER_FAILOVER_PERIOD seconds
Exited after: $ELAPSED seconds
Number of failovers before exit:
# stop the vmstats on the OSTs
if [ "$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); }"
+ do_nodes $(comma_list $(osts_nodes)) "test -f $VMSTAT_PID_FILE &&
+ { kill -s TERM \\\$(cat $VMSTAT_PID_FILE);
+ rm -f $VMSTAT_PID_FILE || true; }"
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; }"
+ do_nodes $NODES_TO_USE "set -x; test -f $LOAD_PID_FILE &&
+ { kill -s TERM \\\$(cat $LOAD_PID_FILE);
+ rm -f $LOAD_PID_FILE || true; }"
# and free up the pdshes that started them, if any are still around
if [ -n "$CLIENT_LOAD_PIDS" ]; then
fi
if [ $rc -ne 0 ]; then
- print_logs $NODES_TO_USE
+ # we are interested in only on failed clients and servers
+ local failedclients=$(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) 1)
+ echo $product
fi
- if [ $rc -eq 0 ]; then
- zconf_mount $(hostname) $MOUNT
- else
- error "exited with rc=$rc"
- fi
+ [ $rc -eq 0 ] && zconf_mount $(hostname) $MOUNT
+
exit $rc
}
# vmstat the osts
if [ "$VMSTAT" ]; then
- do_nodes $(comma_list $(osts_nodes)) "vmstat 1 > $vmstatLOG-\$(hostname) 2>/dev/null </dev/null & echo \$! > /tmp/vmstat.pid"
+ do_nodes $(comma_list $(osts_nodes)) \
+ "vmstat 1 > $TESTLOG_PREFIX.vmstat.\\\$(hostname -s).log \
+ 2>/dev/null </dev/null & echo \\\$! > $VMSTAT_PID_FILE"
fi
# Start client loads.
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
- exit 3
- fi
+if ! do_nodesv $NODES_TO_USE "cat $LOAD_PID_FILE"; then
+ exit 3
fi
START_TS=$(date +%s)
it_time_start=$(date +%s)
FAIL_CLIENT=$(get_random_entry $NODES_TO_USE)
- client_var=${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
-
+
boot_node $FAIL_CLIENT
echo "Reintegrating $FAIL_CLIENT"
zconf_mount $FAIL_CLIENT $MOUNT || exit $?
# not for all clients.
if [ -e $END_RUN_FILE ]; then
read END_RUN_NODE < $END_RUN_FILE
- [[ $END_RUN_NODE = $FAIL_CLIENT ]] &&
+ [[ $END_RUN_NODE = $FAIL_CLIENT ]] &&
rm -f $END_RUN_FILE || exit 13
fi
-
+
restart_client_loads $FAIL_CLIENT $ERRORS_OK || exit $?
# Check that not failed clients loads are still running.
CURRENT_TS=$(date +%s)
ELAPSED=$((CURRENT_TS - START_TS))
-
sleep=$((SERVER_FAILOVER_PERIOD-(CURRENT_TS - it_time_start)))
# keep count the number of itterations when
# the value ( SERVER_FAILOVER_PERIOD - MINSLEEP )
if [ $sleep -lt $MINSLEEP ]; then
reqfail=$((reqfail +1))
- log "WARNING: failover, client reintegration and check_client_loads time
-exceeded SERVER_FAILOVER_PERIOD - MINSLEEP !
-Failed to meet interval $reqfail times ( REQFAIL=$REQFAIL ); have sleep=$sleep"
- [ $reqfail -gt $REQFAIL ] && exit 6
- 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
+ fi
log " Number of failovers:
$(numfailovers) and counting..."
- if [ $sleep -gt 0 ]; then
+ if [ $((ELAPSED + sleep)) -ge $DURATION ]; then
+ break
+ fi
+
+ if [ $sleep -gt 0 ]; then
echo "sleeping $sleep seconds ... "
sleep $sleep
fi
git://git.whamcloud.com / fs / lustre-release.git / blobdiff