X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Ftests%2Flarge-scale.sh;h=9fd35a76434f51828fa85f5c13bf878d4d04b5e5;hb=1155c2fee4c197bfc295f761d52f9b748c37ad79;hp=8da978380460637f109d7d268e639818d2877fb0;hpb=dee5f24114531ec34fc56ce2826ada9e5690aabc;p=fs%2Flustre-release.git diff --git a/lustre/tests/large-scale.sh b/lustre/tests/large-scale.sh index 8da9783..9fd35a7 100644 --- a/lustre/tests/large-scale.sh +++ b/lustre/tests/large-scale.sh @@ -19,13 +19,12 @@ init_test_env $@ . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh} init_logging -remote_mds_nodsh && log "SKIP: remote MDS with nodsh" && exit 0 +remote_mds_nodsh && skip "remote MDS with nodsh" -[ -z "$CLIENTS" ] && skip_env "$TESTSUITE: Need two or more clients" && exit 0 +[ -z "$CLIENTS" ] && skip_env "$TESTSUITE: Need two or more clients" [ $CLIENTCOUNT -lt 2 ] && - skip_env "$TESTSUITE: Need 2+ clients, have only $CLIENTCOUNT" && exit 0 + skip_env "$TESTSUITE: Need 2+ clients, have only $CLIENTCOUNT" -# [ "$SLOW" = "no" ] && EXCEPT_SLOW="" MOUNT_2="" @@ -71,42 +70,46 @@ test_3a() { local num=$increment - while [ $num -le $CLIENTCOUNT ]; do - list=$(comma_list ${nodes[@]:0:$num}) - - generate_machine_file $list $machinefile || - { error "can not generate machinefile"; exit 1; } - - for i in $(seq $iters); do - mdsrate_cleanup $num $machinefile $nfiles $dir 'f%%d' --ignore - - COMMAND="${MDSRATE} --create --nfiles $nfiles --dir $dir --filefmt 'f%%d'" - mpi_run -np $((num * nthreads)) -machinefile $machinefile ${COMMAND} | tee ${LOG} & - - pid=$! - echo "pid=$pid" - - # 2 threads 100000 creates 117 secs - sleep 20 - - log "$i : Starting failover on $SINGLEMDS" - facet_failover $SINGLEMDS - if ! wait_recovery_complete $SINGLEMDS $((TIMEOUT * 10)); then - echo "$SINGLEMDS recovery is not completed!" - kill -9 $pid - exit 7 - fi - - duration=$(do_facet $SINGLEMDS lctl get_param -n $procfile | grep recovery_duration) - - res=( "${res[@]}" "$num" ) - res=( "${res[@]}" "$duration" ) - echo "RECOVERY TIME: NFILES=$nfiles number of clients: $num $duration" - wait $pid - - done - num=$((num + increment)) - done + while [ $num -le $CLIENTCOUNT ]; do + list=$(comma_list ${nodes[@]:0:$num}) + + generate_machine_file $list $machinefile || + { error "can not generate machinefile"; exit 1; } + + for i in $(seq $iters); do + mdsrate_cleanup $num $machinefile $nfiles $dir 'f%%d' \ + --ignore + + COMMAND="${MDSRATE} --create --nfiles $nfiles --dir + $dir --filefmt 'f%%d'" + mpi_run ${MACHINEFILE_OPTION} $machinefile \ + -np $((num * nthreads)) ${COMMAND} | tee ${LOG}& + + pid=$! + echo "pid=$pid" + + # 2 threads 100000 creates 117 secs + sleep 20 + + log "$i : Starting failover on $SINGLEMDS" + facet_failover $SINGLEMDS + if ! wait_recovery_complete $SINGLEMDS \ + $((TIMEOUT * 10)); then + echo "$SINGLEMDS recovery is not completed!" + kill -9 $pid + exit 7 + fi + + duration=$(do_facet $SINGLEMDS lctl get_param -n \ + $procfile | grep recovery_duration) + + res=( "${res[@]}" "$num" ) + res=( "${res[@]}" "$duration" ) + echo "RECOVERY TIME: NFILES=$nfiles number of clients: $num $duration" + wait $pid + done + num=$((num + increment)) + done mdsrate_cleanup $num $machinefile $nfiles $dir 'f%%d' --ignore