set -e
-# bug 5494 7288 5493
-ALWAYS_EXCEPT="24 27 52 $RECOVERY_SMALL_EXCEPT"
+# bug 5494 5493
+ALWAYS_EXCEPT="24 52 $RECOVERY_SMALL_EXCEPT"
PTLDEBUG=${PTLDEBUG:--1}
LUSTRE=${LUSTRE:-`dirname $0`/..}
. $LUSTRE/tests/test-framework.sh
init_test_env $@
. ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
+init_logging
-if [ "$FAILURE_MODE" = "HARD" ] && mixed_ost_devs; then
- CONFIG_EXCEPTIONS="52"
- echo -n "Several ost services on one ost node are used with FAILURE_MODE=$FAILURE_MODE. "
- echo "Except the tests: $CONFIG_EXCEPTIONS"
- ALWAYS_EXCEPT="$ALWAYS_EXCEPT $CONFIG_EXCEPTIONS"
-fi
-
-remote_mds_nodsh && skip "remote MDS with nodsh" && exit 0
+require_dsh_mds || exit 0
# also long tests: 19, 21a, 21e, 21f, 23, 27
# 1 2.5 2.5 4 4 (min)"
remote_ost_nodsh && skip "remote OST with nodsh" && return 0
# With adaptive timeouts, bulk_get won't expire until adaptive_timeout_max
- if at_is_valid && at_is_enabled; then
+ if at_is_enabled; then
at_max_saved=$(at_max_get ost1)
at_max_set $TIMEOUT ost1
fi
# OST bulk will time out here, client retries
do_facet ost1 lctl set_param fail_loc=0x80000503
# need to ensure we send an RPC
- do_facet client cp /etc/termcap $DIR/$tfile
+ do_facet client cp $SAMPLE_FILE $DIR/$tfile
sync
# with AT, client will wait adaptive_max*factor+net_latency before
do_facet ost1 lctl set_param fail_loc=0
do_facet client "df $DIR"
# expect cmp to succeed, client resent bulk
- do_facet client "cmp /etc/termcap $DIR/$tfile" || return 3
+ do_facet client "cmp $SAMPLE_FILE $DIR/$tfile" || return 3
do_facet client "rm $DIR/$tfile" || return 4
- [ $at_max_saved -ne 0 ] && $(at_max_set $at_max_saved ost1)
+ [ $at_max_saved -ne 0 ] && at_max_set $at_max_saved ost1
return 0
}
run_test 17 "timeout bulk get, don't evict client (2732)"
test_18a() {
- [ -z ${ost2_svc} ] && skip "needs 2 osts" && return 0
+ [ -z ${ost2_svc} ] && skip_env "needs 2 osts" && return 0
do_facet client mkdir -p $DIR/$tdir
f=$DIR/$tdir/$tfile
pgcache_empty || return 1
# 1 stripe on ost2
- lfs setstripe $f -s $((128 * 1024)) -i 1 -c 1
- get_stripe_info client $f
+ $LFS setstripe -i 1 -c 1 $f
+ stripe_index=$($LFS getstripe -i $f)
if [ $stripe_index -ne 1 ]; then
- lfs getstripe $f
- error "$f: different stripe offset ($stripe_index)" && return
+ $LFS getstripe $f
+ error "$f: stripe_index $stripe_index != 1" && return
fi
do_facet client cp $SAMPLE_FILE $f
cancel_lru_locks osc
pgcache_empty || return 1
- # shouldn't have to set stripe size of count==1
- lfs setstripe $f -s $((128 * 1024)) -i 0 -c 1
- get_stripe_info client $f
+ $LFS setstripe -i 0 -c 1 $f
+ stripe_index=$($LFS getstripe -i $f)
if [ $stripe_index -ne 0 ]; then
- lfs getstripe $f
- error "$f: different stripe offset ($stripe_index)" && return
+ $LFS getstripe $f
+ error "$f: stripe_index $stripe_index != 0" && return
fi
do_facet client cp $SAMPLE_FILE $f
cancel_lru_locks osc
pgcache_empty || return 1
- # shouldn't have to set stripe size of count==1
- lfs setstripe $f -s $((128 * 1024)) -i 0 -c 1
- get_stripe_info client $f
+ $LFS setstripe -i 0 -c 1 $f
+ stripe_index=$($LFS getstripe -i $f)
if [ $stripe_index -ne 0 ]; then
- lfs getstripe $f
- error "$f: different stripe offset ($stripe_index)" && return
+ $LFS getstripe $f
+ error "$f: stripe_index $stripe_index != 0" && return
fi
do_facet client cp $SAMPLE_FILE $f
# lost reply to connect request
do_facet ost1 lctl set_param fail_loc=0x80000225
# force reconnect
+ sleep 1
df $MOUNT > /dev/null 2>&1
sleep 2
# my understanding is that there should be nothing in the page
remote_ost_nodsh && skip "remote OST with nodsh" && return 0
mkdir -p $DIR/$tdir
- lfs setstripe $DIR/$tdir/${tfile} -i 0 -c 1
+ $LFS setstripe -i 0 -c 1 $DIR/$tdir/${tfile}
multiop_bg_pause $DIR/$tdir/${tfile} O_wc || return 1
MULTI_PID=$!
cancel_lru_locks osc
remote_ost_nodsh && skip "remote OST with nodsh" && return 0
mkdir -p $DIR/$tdir
- lfs setstripe $DIR/$tdir/${tfile} -i 0 -c 1
+ $LFS setstripe -i 0 -c 1 $DIR/$tdir/${tfile}
cancel_lru_locks osc
#define OBD_FAIL_LDLM_ENQUEUE_EXTENT_ERR 0x308
do_facet ost1 lctl set_param fail_loc=0x80000308
remote_ost_nodsh && skip "remote OST with nodsh" && return 0
mkdir -p $DIR/$tdir
- lfs setstripe $DIR/$tdir -s 0 -i 0 -c 1
+ $LFS setstripe -i 0 -c 1 $DIR/$tdir
cancel_lru_locks osc
multiop_bg_pause $DIR/$tdir/$tfile Owy_wyc || return 1
MULTI_PID=$!
}
run_test 24 "fsync error (should return error)"
+wait_client_evicted () {
+ local facet=$1
+ local exports=$2
+ local varsvc=${facet}_svc
+
+ wait_update $(facet_active_host $facet) \
+ "lctl get_param -n *.${!varsvc}.num_exports | cut -d' ' -f2" \
+ $((exports - 1)) $3
+}
+
test_26a() { # was test_26 bug 5921 - evict dead exports by pinger
# this test can only run from a client on a separate node.
remote_ost || { skip "local OST" && return 0; }
remote_ost_nodsh && skip "remote OST with nodsh" && return 0
remote_mds || { skip "local MDS" && return 0; }
- OST_FILE=obdfilter.${ost1_svc}.num_exports
- OST_EXP="`do_facet ost1 lctl get_param -n $OST_FILE`"
- OST_NEXP1=`echo $OST_EXP | cut -d' ' -f2`
- echo starting with $OST_NEXP1 OST exports
+
+ if [ $(facet_host mgs) = $(facet_host ost1) ]; then
+ skip "msg and ost1 are at the same node"
+ return 0
+ fi
+
+ check_timeout || return 1
+
+ local OST_NEXP=$(do_facet ost1 lctl get_param -n obdfilter.${ost1_svc}.num_exports | cut -d' ' -f2)
+
+ echo starting with $OST_NEXP OST exports
# OBD_FAIL_PTLRPC_DROP_RPC 0x505
do_facet client lctl set_param fail_loc=0x505
# evictor takes PING_EVICT_TIMEOUT + 3 * PING_INTERVAL to evict.
# But if there's a race to start the evictor from various obds,
# the loser might have to wait for the next ping.
- echo Waiting for $(($TIMEOUT * 8)) secs
- sleep $(($TIMEOUT * 8))
- OST_EXP="`do_facet ost1 lctl get_param -n $OST_FILE`"
- OST_NEXP2=`echo $OST_EXP | cut -d' ' -f2`
- echo ending with $OST_NEXP2 OST exports
+
+ local rc=0
+ wait_client_evicted ost1 $OST_NEXP $((TIMEOUT * 2 + TIMEOUT * 3 / 4))
+ rc=$?
do_facet client lctl set_param fail_loc=0x0
- [ $OST_NEXP1 -le $OST_NEXP2 ] && error "client not evicted"
- return 0
+ [ $rc -eq 0 ] || error "client not evicted from OST"
}
run_test 26a "evict dead exports"
test_26b() { # bug 10140 - evict dead exports by pinger
remote_ost_nodsh && skip "remote OST with nodsh" && return 0
- client_df
- zconf_mount `hostname` $MOUNT2 || error "Failed to mount $MOUNT2"
- sleep 1 # wait connections being established
- MDS_FILE=mdt.${mds1_svc}.num_exports
- MDS_NEXP1="`do_facet $SINGLEMDS lctl get_param -n $MDS_FILE | cut -d' ' -f2`"
- OST_FILE=obdfilter.${ost1_svc}.num_exports
- OST_NEXP1="`do_facet ost1 lctl get_param -n $OST_FILE | cut -d' ' -f2`"
- echo starting with $OST_NEXP1 OST and $MDS_NEXP1 MDS exports
- zconf_umount `hostname` $MOUNT2 -f
- # evictor takes PING_EVICT_TIMEOUT + 3 * PING_INTERVAL to evict.
- # But if there's a race to start the evictor from various obds,
- # the loser might have to wait for the next ping.
- echo Waiting for $(($TIMEOUT * 3)) secs
- sleep $(($TIMEOUT * 3))
- OST_NEXP2="`do_facet ost1 lctl get_param -n $OST_FILE | cut -d' ' -f2`"
- MDS_NEXP2="`do_facet $SINGLEMDS lctl get_param -n $MDS_FILE | cut -d' ' -f2`"
- echo ending with $OST_NEXP2 OST and $MDS_NEXP2 MDS exports
- [ $OST_NEXP1 -le $OST_NEXP2 ] && error "client not evicted from OST"
- [ $MDS_NEXP1 -le $MDS_NEXP2 ] && error "client not evicted from MDS"
- return 0
+ if [ $(facet_host mgs) = $(facet_host ost1) ]; then
+ skip "msg and ost1 are at the same node"
+ return 0
+ fi
+
+ check_timeout || return 1
+ clients_up
+ zconf_mount `hostname` $MOUNT2 ||
+ { error "Failed to mount $MOUNT2"; return 2; }
+ sleep 1 # wait connections being established
+
+ local MDS_NEXP=$(do_facet $SINGLEMDS lctl get_param -n mdt.${mds1_svc}.num_exports | cut -d' ' -f2)
+ local OST_NEXP=$(do_facet ost1 lctl get_param -n obdfilter.${ost1_svc}.num_exports | cut -d' ' -f2)
+
+ echo starting with $OST_NEXP OST and $MDS_NEXP MDS exports
+
+ zconf_umount `hostname` $MOUNT2 -f
+
+ # PING_INTERVAL max(obd_timeout / 4, 1U)
+ # PING_EVICT_TIMEOUT (PING_INTERVAL * 6)
+
+ # evictor takes PING_EVICT_TIMEOUT + 3 * PING_INTERVAL to evict.
+ # But if there's a race to start the evictor from various obds,
+ # the loser might have to wait for the next ping.
+ # = 9 * PING_INTERVAL + PING_INTERVAL
+ # = 10 PING_INTERVAL = 10 obd_timeout / 4 = 2.5 obd_timeout
+ # let's wait $((TIMEOUT * 3)) # bug 19887
+ local rc=0
+ wait_client_evicted ost1 $OST_NEXP $((TIMEOUT * 3)) || \
+ error "Client was not evicted by ost" rc=1
+ wait_client_evicted $SINGLEMDS $MDS_NEXP $((TIMEOUT * 3)) || \
+ error "Client was not evicted by mds"
}
run_test 26b "evict dead exports"
#define OBD_FAIL_OSC_SHUTDOWN 0x407
do_facet $SINGLEMDS lctl set_param fail_loc=0x80000407
# need to wait for reconnect
- echo -n waiting for fail_loc
- while [ $(do_facet $SINGLEMDS lctl get_param -n fail_loc) -eq -2147482617 ]; do
- sleep 1
- echo -n .
- done
- do_facet $SINGLEMDS lctl get_param -n fail_loc
+ echo waiting for fail_loc
+ wait_update_facet $SINGLEMDS "lctl get_param -n fail_loc" "-2147482617"
facet_failover $SINGLEMDS
#no crashes allowed!
kill -USR1 $CLIENT_PID
#define OBD_FAIL_MDS_CLIENT_ADD 0x12f
do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012f"
# fail once (evicted), reconnect fail (fail_loc), ok
- df || (sleep 10; df) || (sleep 10; df) || error "reconnect failed"
+ client_up || (sleep 10; client_up) || (sleep 10; client_up) || error "reconnect failed"
rm -f $DIR/$tfile
fail $SINGLEMDS # verify MDS last_rcvd can be loaded
}
run_test 28 "handle error adding new clients (bug 6086)"
+test_29a() { # bug 22273 - error adding new clients
+ #define OBD_FAIL_TGT_CLIENT_ADD 0x711
+ do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000711"
+ # fail abort so client will be new again
+ fail_abort $SINGLEMDS
+ client_up || error "reconnect failed"
+ return 0
+}
+run_test 29a "error adding new clients doesn't cause LBUG (bug 22273)"
+
+test_29b() { # bug 22273 - error adding new clients
+ #define OBD_FAIL_TGT_CLIENT_ADD 0x711
+ do_facet ost1 "lctl set_param fail_loc=0x80000711"
+ # fail abort so client will be new again
+ fail_abort ost1
+ client_up || error "reconnect failed"
+ return 0
+}
+run_test 29b "error adding new clients doesn't cause LBUG (bug 22273)"
+
test_50() {
mkdir -p $DIR/$tdir
# put a load of file creates/writes/deletes
run_test 50 "failover MDS under load"
test_51() {
+ #define OBD_FAIL_MDS_SYNC_CAPA_SL 0x1310
+ do_facet ost1 lctl set_param fail_loc=0x00001310
+
mkdir -p $DIR/$tdir
# put a load of file creates/writes/deletes
writemany -q $DIR/$tdir/$tfile 0 5 &
mkdir -p $DIR/$tdir
# first dd should be finished quickly
- lfs setstripe $DIR/$tdir/$tfile-1 -c 1 -i 0
+ $LFS setstripe -c 1 -i 0 $DIR/$tdir/$tfile-1
dd if=/dev/zero of=$DIR/$tdir/$tfile-1 bs=32M count=4 &
DDPID=$!
count=0
done
echo "(dd_pid=$DDPID, time=$count)successful"
- lfs setstripe $DIR/$tdir/$tfile-2 -c 1 -i 0
+ $LFS setstripe -c 1 -i 0 $DIR/$tdir/$tfile-2
#define OBD_FAIL_OST_DROP_REQ 0x21d
do_facet ost1 lctl set_param fail_loc=0x0000021d
# second dd will be never finished
}
run_test 59 "Read cancel race on client eviction"
-equals_msg `basename $0`: test complete, cleaning up
+err17935 () {
+ # we assume that all md changes are in the MDT0 changelog
+ if [ $MDSCOUNT -gt 1 ]; then
+ error_ignore 17935 $*
+ else
+ error $*
+ fi
+}
+
+test_60() {
+ MDT0=$($LCTL get_param -n mdc.*.mds_server_uuid | \
+ awk '{gsub(/_UUID/,""); print $1}' | head -1)
+
+ NUM_FILES=15000
+ mkdir -p $DIR/$tdir
+
+ # Register (and start) changelog
+ USER=$(do_facet $SINGLEMDS lctl --device $MDT0 changelog_register -n)
+ echo "Registered as $MDT0 changelog user $USER"
+
+ # Generate a large number of changelog entries
+ createmany -o $DIR/$tdir/$tfile $NUM_FILES
+ sync
+ sleep 5
+
+ # Unlink files in the background
+ unlinkmany $DIR/$tdir/$tfile $NUM_FILES &
+ CLIENT_PID=$!
+ sleep 1
+
+ # Failover the MDS while unlinks are happening
+ facet_failover $SINGLEMDS
+
+ # Wait for unlinkmany to finish
+ wait $CLIENT_PID
+
+ # Check if all the create/unlink events were recorded
+ # in the changelog
+ $LFS changelog $MDT0 >> $DIR/$tdir/changelog
+ local cl_count=$(grep UNLNK $DIR/$tdir/changelog | wc -l)
+ echo "$cl_count unlinks in $MDT0 changelog"
+
+ do_facet $SINGLEMDS lctl --device $MDT0 changelog_deregister $USER
+ USERS=$(( $(do_facet $SINGLEMDS lctl get_param -n \
+ mdd.$MDT0.changelog_users | wc -l) - 2 ))
+ if [ $USERS -eq 0 ]; then
+ [ $cl_count -eq $NUM_FILES ] || \
+ err17935 "Recorded ${cl_count} unlinks out of $NUM_FILES"
+ # Also make sure we can clear large changelogs
+ cl_count=$($LFS changelog $FSNAME | wc -l)
+ [ $cl_count -le 2 ] || \
+ error "Changelog not empty: $cl_count entries"
+ else
+ # If there are other users, there may be other unlinks in the log
+ [ $cl_count -ge $NUM_FILES ] || \
+ err17935 "Recorded ${cl_count} unlinks out of $NUM_FILES"
+ echo "$USERS other changelog users; can't verify clear"
+ fi
+}
+run_test 60 "Add Changelog entries during MDS failover"
+
+test_61()
+{
+ local mdtosc=$(get_mdtosc_proc_path $SINGLEMDS $FSNAME-OST0000)
+ mdtosc=${mdtosc/-MDT*/-MDT\*}
+ local cflags="osc.$mdtosc.connect_flags"
+ do_facet $SINGLEMDS "lctl get_param -n $cflags" |grep -q skip_orphan
+ [ $? -ne 0 ] && skip "don't have skip orphan feature" && return
+
+ mkdir -p $DIR/$tdir || error "mkdir dir $DIR/$tdir failed"
+ # Set the default stripe of $DIR/$tdir to put the files to ost1
+ $LFS setstripe -c 1 -i 0 $DIR/$tdir
+
+ replay_barrier $SINGLEMDS
+ createmany -o $DIR/$tdir/$tfile-%d 10
+ local oid=`do_facet ost1 "lctl get_param -n obdfilter.${ost1_svc}.last_id"`
+
+ fail_abort $SINGLEMDS
+
+ touch $DIR/$tdir/$tfile
+ local id=`$LFS getstripe $DIR/$tdir/$tfile | awk '$1 == 0 { print $2 }'`
+ [ $id -le $oid ] && error "the orphan objid was reused, failed"
+
+ # Cleanup
+ rm -rf $DIR/$tdir
+}
+run_test 61 "Verify to not reuse orphan objects - bug 17025"
+
+check_cli_ir_state()
+{
+ local NODE=${1:-$HOSTNAME}
+ local st
+ st=$(do_node $NODE "lctl get_param mgc.*.ir_state |
+ awk '/imperative_recovery:/ { print \\\$2}'")
+ [ $st != ON -o $st != OFF ] ||
+ error "Error state $st, must be ON or OFF"
+ echo -n $st
+}
+
+check_target_ir_state()
+{
+ local target=${1}
+ local name=${target}_svc
+ local recovery_proc=obdfilter.${!name}.recovery_status
+ local st
+
+ st=$(do_facet $target "lctl get_param -n $recovery_proc |
+ awk '/IR:/{ print \\\$2}'")
+ [ $st != ON -o $st != OFF ] ||
+ error "Error state $st, must be ON or OFF"
+ echo -n $st
+}
+
+set_ir_status()
+{
+ do_facet mgs lctl set_param -n mgs.MGS.live.$FSNAME="state=$1"
+}
+
+get_ir_status()
+{
+ local state=$(do_facet mgs "lctl get_param -n mgs.MGS.live.$FSNAME |
+ awk '/state:/{ print \\\$2 }'")
+ echo -n ${state/,/}
+}
+
+nidtbl_version_mgs()
+{
+ local ver=$(do_facet mgs "lctl get_param -n mgs.MGS.live.$FSNAME |
+ awk '/nidtbl_version:/{ print \\\$2 }'")
+ echo -n $ver
+}
+
+# nidtbl_version_client <mds1|client> [node]
+nidtbl_version_client()
+{
+ local cli=$1
+ local node=${2:-$HOSTNAME}
+
+ if [ X$cli = Xclient ]; then
+ cli=$FSNAME-client
+ else
+ local obdtype=${cli/%[0-9]*/}
+ [ $obdtype != mds ] && error "wrong parameters $cli"
+
+ node=$(facet_active_host $cli)
+ local t=${cli}_svc
+ cli=${!t}
+ fi
+
+ local vers=$(do_node $node "lctl get_param -n mgc.*.ir_state" |
+ awk "/$cli/{print \$6}" |sort -u)
+
+ # in case there are multiple mounts on the client node
+ local arr=($vers)
+ [ ${#arr[@]} -ne 1 ] && error "versions on client node mismatch"
+ echo -n $vers
+}
+
+nidtbl_versions_match()
+{
+ [ $(nidtbl_version_mgs) -eq $(nidtbl_version_client ${1:-client}) ]
+}
+
+target_instance_match()
+{
+ local srv=$1
+ local obdtype
+ local cliname
+
+ obdtype=${srv/%[0-9]*/}
+ case $obdtype in
+ mds)
+ obdname="mdt"
+ cliname="mdc"
+ ;;
+ ost)
+ obdname="obdfilter"
+ cliname="osc"
+ ;;
+ *)
+ error "invalid target type" $srv
+ return 1
+ ;;
+ esac
+
+ local target=${srv}_svc
+ local si=$(do_facet $srv lctl get_param -n $obdname.${!target}.instance)
+ local ci=$(lctl get_param -n $cliname.${!target}-${cliname}-*.import | \
+ awk '/instance/{ print $2 }' |head -1)
+
+ return $([ $si -eq $ci ])
+}
+
+test_100()
+{
+ do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
+ { skip "MGS without IR support"; return 0; }
+
+ # disable IR
+ set_ir_status disabled
+
+ local saved_FAILURE_MODE=$FAILURE_MODE
+ [ $(facet_host mgs) = $(facet_host ost1) ] && FAILURE_MODE="SOFT"
+ fail ost1
+
+ # valid check
+ nidtbl_versions_match &&
+ error "version must differ due to IR disabled"
+ target_instance_match ost1 || error "instance mismatch"
+
+ # restore env
+ set_ir_status full
+ FAILURE_MODE=$saved_FAILURE_MODE
+}
+run_test 100 "IR: Make sure normal recovery still works w/o IR"
+
+test_101()
+{
+ do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
+ { skip "MGS without IR support"; return 0; }
+
+ set_ir_status full
+
+ local OST1_IMP=$(get_osc_import_name client ost1)
+
+ # disable pinger recovery
+ lctl set_param -n osc.$OST1_IMP.pinger_recov=0
+
+ fail ost1
+
+ target_instance_match ost1 || error "instance mismatch"
+ nidtbl_versions_match || error "version must match"
+
+ lctl set_param -n osc.$OST1_IMP.pinger_recov=1
+}
+run_test 101 "IR: Make sure IR works w/o normal recovery"
+
+test_102()
+{
+ do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
+ { skip "MGS without IR support"; return 0; }
+
+ local clients=${CLIENTS:-$HOSTNAME}
+ local old_version
+ local new_version
+ local mgsdev=mgs
+
+ set_ir_status full
+
+ # let's have a new nidtbl version
+ fail ost1
+
+ # sleep for a while so that clients can see the failure of ost
+ # it must be MGC_TIMEOUT_MIN_SECONDS + MGC_TIMEOUT_RAND_CENTISEC.
+ # int mgc_request.c:
+ # define MGC_TIMEOUT_MIN_SECONDS 5
+ # define MGC_TIMEOUT_RAND_CENTISEC 0x1ff /* ~500 *
+ local count=30 # 20 seconds at most
+ while [ $count -gt 0 ]; do
+ nidtbl_versions_match && break
+ sleep 1
+ count=$((count-1))
+ done
+
+ nidtbl_versions_match || error "nidtbl mismatch"
+
+ # get the version #
+ old_version=$(nidtbl_version_client client)
+
+ zconf_umount_clients $clients $MOUNT || error "Cannot umount client"
+
+ # restart mgs
+ combined_mgs_mds && mgsdev=mds1
+ remount_facet $mgsdev
+ fail ost1
+
+ zconf_mount_clients $clients $MOUNT || error "Cannot mount client"
+
+ # check new version
+ new_version=$(nidtbl_version_client client)
+ [ $new_version -lt $old_version ] &&
+ error "nidtbl version wrong after mgs restarts"
+ return 0
+}
+run_test 102 "IR: New client gets updated nidtbl after MGS restart"
+
+test_103()
+{
+ do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
+ { skip "MGS without IR support"; return 0; }
+
+ combined_mgs_mds && skip "mgs and mds on the same target" && return 0
+
+ # workaround solution to generate config log on the mds
+ remount_facet mds1
+
+ stop mgs
+ stop mds1
+
+ # We need this test because mds is like a client in IR context.
+ start mds1 $MDSDEV1 || error "MDS should start w/o mgs"
+
+ # start mgs and remount mds w/ ir
+ start mgs $MGSDEV
+ clients_up
+
+ # remount client so that fsdb will be created on the MGS
+ umount_client $MOUNT || error "umount failed"
+ mount_client $MOUNT || error "mount failed"
+
+ # sleep 30 seconds so the MDS has a chance to detect MGS restarting
+ local count=30
+ while [ $count -gt 0 ]; do
+ [ $(nidtbl_version_client mds1) -ne 0 ] && break
+ sleep 1
+ count=$((count-1))
+ done
+
+ # after a while, mds should be able to reconnect to mgs and fetch
+ # up-to-date nidtbl version
+ nidtbl_versions_match mds1 || error "mds nidtbl mismatch"
+
+ # reset everything
+ set_ir_status full
+}
+run_test 103 "IR: MDS can start w/o MGS and get updated nidtbl later"
+
+test_104()
+{
+ do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
+ { skip "MGS without IR support"; return 0; }
+
+ set_ir_status full
+
+ stop ost1
+ start ost1 $(ostdevname 1) "$OST_MOUNT_OPTS -onoir" ||
+ error "OST1 cannot start"
+ clients_up
+
+ local ir_state=$(check_target_ir_state ost1)
+ [ $ir_state = "OFF" ] || error "ir status on ost1 should be OFF"
+}
+run_test 104 "IR: ost can disable IR voluntarily"
+
+test_105()
+{
+ [ -z "$RCLIENTS" ] && skip "Needs multiple clients" && return 0
+ do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
+ { skip "MGS without IR support"; return 0; }
+
+ set_ir_status full
+
+ # get one of the clients from client list
+ local rcli=$(echo $RCLIENTS |cut -d' ' -f 1)
+
+ local old_MOUNTOPT=$MOUNTOPT
+ MOUNTOPT=${MOUNTOPT},noir
+ zconf_umount $rcli $MOUNT || error "umount failed"
+ zconf_mount $rcli $MOUNT || error "mount failed"
+
+ # make sure lustre mount at $rcli disabling IR
+ local ir_state=$(check_cli_ir_state $rcli)
+ [ $ir_state = OFF ] || error "IR state must be OFF at $rcli"
+
+ # make sure MGS's state is Partial
+ [ $(get_ir_status) = "partial" ] || error "MGS IR state must be partial"
+
+ fail ost1
+ # make sure IR on ost1 is OFF
+ local ir_state=$(check_target_ir_state ost1)
+ [ $ir_state = "OFF" ] || error "IR status on ost1 should be OFF"
+
+ # restore it
+ MOUNTOPT=$old_MOUNTOPT
+ zconf_umount $rcli $MOUNT || error "umount failed"
+ zconf_mount $rcli $MOUNT || error "mount failed"
+
+ # make sure MGS's state is full
+ [ $(get_ir_status) = "full" ] || error "MGS IR status must be full"
+
+ fail ost1
+ # make sure IR on ost1 is ON
+ local ir_state=$(check_target_ir_state ost1)
+ [ $ir_state = "ON" ] || error "IR status on ost1 should be OFF"
+
+ return 0
+}
+run_test 105 "IR: NON IR clients support"
+
+complete $(basename $0) $SECONDS
check_and_cleanup_lustre
-[ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG && grep -q FAIL $TESTSUITELOG && exit 1 || true
+exit_status