#!/bin/bash
-
+#
# Was Test 11 in cmd3.
# For duration of 24 hours repeatedly failover a random MDS at
# 10 minute intervals and verify that no application errors occur.
# Test runs one of CLIENT_LOAD progs on remote clients.
+set -e
-LUSTRE=${LUSTRE:-`dirname $0`/..}
-SETUP=${SETUP:-""}
-CLEANUP=${CLEANUP:-""}
-. $LUSTRE/tests/test-framework.sh
-
-init_test_env $@
-
-. ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
-
-TESTSUITELOG=${TESTSUITELOG:-$TMP/recovery-mds-scale}
-DEBUGLOG=$TESTSUITELOG.debug
-exec 2>$DEBUGLOG
-echo "--- env ---" >&2
-env >&2
-echo "--- env ---" >&2
-set -x
-
-[ "$SHARED_DIRECTORY" ] || \
- { skip "$0: Empty SHARED_DIRECTORY" && exit 0; }
+ONLY=${ONLY:-"$*"}
-[ -n "$CLIENTS" ] || { skip "$0 Need two or more remote clients" && exit 0; }
-[ $CLIENTCOUNT -ge 3 ] || \
- { skip "$0 Need two or more clients, have $CLIENTCOUNT" && exit 0; }
-
-END_RUN_FILE=${END_RUN_FILE:-$SHARED_DIRECTORY}/end_run_file}
-LOAD_PID_FILE=${LOAD_PID_FILE:-$TMP/client-load.pid}
+LUSTRE=${LUSTRE:-$(dirname $0)/..}
+. $LUSTRE/tests/test-framework.sh
+. $LUSTRE/tests/recovery-scale-lib.sh
+init_test_env "$@"
+init_logging
-remote_mds_nodsh && skip "remote MDS with nodsh" && exit 0
-remote_ost_nodsh && skip "remote OST with nodsh" && exit 0
+# bug number for skipped test:
+ALWAYS_EXCEPT="$RECOVERY_MDS_SCALE_EXCEPT "
+# UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
build_test_filter
-check_and_setup_lustre
-rm -rf $DIR/[df][0-9]*
-
-# 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
-# -- remove hostname from clients list
-zconf_umount $(hostname) $MOUNT
-NODES_TO_USE=${NODES_TO_USE:-$CLIENTS}
-NODES_TO_USE=$(exclude_item_from_list $NODES_TO_USE $(hostname))
-
-check_progs_installed $NODES_TO_USE ${CLIENT_LOADS[@]}
-
-MDTS=""
-for ((i=1; i<=$MDSCOUNT; i++)) do
- MDTS="$MDTS mds$i"
-done
-MDTS=$(comma_list $MDTS)
-
-OSTS=""
-for ((i=1; i<=$OSTCOUNT; i++)) do
- OSTS="$OSTS ost$i"
-done
-OSTS=$(comma_list $OSTS)
-
-ERRORS_OK="" # No application failures should occur during this test.
-FLAVOR=${FLAVOR:-"MDS"}
-
-rm -f $END_RUN_FILE
-
-vmstatLOG=${TESTSUITELOG}_$(basename $0 .sh).vmstat
-
-server_numfailovers () {
- local facet
- local var
-
- for facet in $MDTS ${OSTS//,/ }; do
- var=${facet}_nums
- val=${!var}
- if [ "$val" ] ; then
- echo "$facet failed over $val times"
- fi
- done
-}
-
-summary_and_cleanup () {
-
- local rc=$?
- 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=${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
+remote_mds_nodsh && skip_env "remote MDS with nodsh"
- log "Duraion: $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
- [ $rc -eq 0 ] && zconf_mount $(hostname) $MOUNT
-
- exit $rc
-}
-
-#
-# MAIN
-#
-log "-----============= $0 starting =============-----"
-
-trap summary_and_cleanup EXIT INT
-
-DURATION=${DURATION:-$((60*60*24))}
-ELAPSED=0
-NUM_FAILOVERS=0
-
-# vmstat the osts
-if [ "$VMSTAT" ]; then
- do_nodes $(comma_list $(osts_nodes)) "vmstat 1 > $vmstatLOG-\$(hostname) 2>/dev/null </dev/null & echo \$! > /tmp/vmstat.pid"
+if (( CLIENTCOUNT < 3 )); then
+ skip_env "need three or more clients"
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 $DEBUGLOG
- exit 3
- fi
+# SHARED_DIRECTORY should be specified with a shared directory which is
+# accessable on all of the nodes
+if [[ -z "$SHARED_DIRECTORY" ]] || ! check_shared_dir "$SHARED_DIRECTORY"; then
+ skip_env "SHARED_DIRECTORY not set"
fi
-START_TS=$(date +%s)
-CURRENT_TS=$START_TS
-
-if [ "$FLAVOR" == "MDS" ]; then
- SERVER_FAILOVER_PERIOD=$MDS_FAILOVER_PERIOD
- SERVERS=$MDTS
-else
- SERVER_FAILOVER_PERIOD=$OSS_FAILOVER_PERIOD
- SERVERS=$OSTS
-fi
-
-SERVER_FAILOVER_PERIOD=${SERVER_FAILOVER_PERIOD:-$((60 * 10))} # 10 minutes
-
-MINSLEEP=${MINSLEEP:-120}
-REQFAIL_PERCENT=${REQFAIL_PERCENT:-3} # bug17839 comment 62
-REQFAIL=${REQFAIL:-$(( DURATION / SERVER_FAILOVER_PERIOD * REQFAIL_PERCENT / 100))}
-reqfail=0
-sleep=0
-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}_nums
-
- # Check that our client loads are still running. If any have died,
- # that means they have died outside of recovery, which is unacceptable.
-
- log "==== Checking the clients loads BEFORE failover -- failure NOT OK \
- ELAPSED=$ELAPSED DURATION=$DURATION PERIOD=$SERVER_FAILOVER_PERIOD"
-
- if ! check_client_loads $NODES_TO_USE; then
- exit 4
- fi
-
- log "Starting failover on $SERVERNODE"
+ERRORS_OK="" # No application failures should occur during this test.
- facet_failover "$SERVERFACET" || exit 1
+check_and_setup_lustre
+rm -rf $DIR/[Rdfs][0-9]*
- # Check that our client loads are still running during failover.
- # No application failures should occur.
+insulate_clients
+check_progs_installed $NODES_TO_USE "${CLIENT_LOADS[@]}"
- log "==== Checking the clients loads AFTER failover -- failure NOT OK"
- if ! check_client_loads $NODES_TO_USE; then
- log "Client load failed during failover. Exiting"
- exit 5
- fi
+MAX_RECOV_TIME=$(max_recovery_time)
+MDTS=$(get_facets MDS)
+OSTS=$(get_facets OST)
- # Increment the number of failovers
- NUM_FAILOVERS=$((NUM_FAILOVERS+1))
- val=$((${!var} + 1))
- eval $var=$val
-
- CURRENT_TS=$(date +%s)
- ELAPSED=$((CURRENT_TS - START_TS))
-
- sleep=$((SERVER_FAILOVER_PERIOD-(CURRENT_TS - it_time_start)))
+# Print informaiton about settings
+run_info $SERVER_FAILOVER_PERIOD $DURATION $MINSLEEP $SLOW $REQFAIL \
+ $SHARED_DIRECTORY $END_RUN_FILE $LOAD_PID_FILE $VMSTAT_PID_FILE \
+ $CLIENTCOUNT $MDTS $OSTS
- # keep count the number of itterations when
- # time spend to failover and two client loads check exceeded
- # the value ( SERVER_FAILOVER_PERIOD - MINSLEEP )
- if [ $sleep -lt $MINSLEEP ]; then
- reqfail=$((reqfail +1))
- log "WARNING: failover and two check_client_loads time exceeded SERVER_FAILOVER_PERIOD - MINSLEEP !
-Failed to meet interval $reqfail times ( REQFAIL=$REQFAIL ); have sleep=$sleep"
- [ $reqfail -gt $REQFAIL ] && exit 6
- fi
+test_failover_mds() {
+ # failover a random MDS
+ failover_target MDS
+}
+run_test failover_mds "failover MDS"
- log "$SERVERFACET has failed over ${!var} times, and counting..."
- if [ $sleep -gt 0 ]; then
- echo "sleeping $sleep seconds ... "
- sleep $sleep
- fi
-done
+zconf_mount $HOSTNAME $MOUNT || error "mount $MOUNT on $HOSTNAME failed"
+client_up || error "start client on $HOSTNAME failed"
-exit 0
+complete_test $SECONDS
+check_and_cleanup_lustre
+exit_status