X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Ftests%2Frecovery-small.sh;h=365be602462486be1460758dded6c43b1d8302fe;hb=b511c65f468698aeb6e68f091ee3f30ec27830d0;hp=e8cf1282d4cf2c62946bdda3c1c99be943ed9e5a;hpb=9a6b2622c932686b28d2d4a041c769c3f3f3c3f3;p=fs%2Flustre-release.git diff --git a/lustre/tests/recovery-small.sh b/lustre/tests/recovery-small.sh index e8cf128..365be60 100755 --- a/lustre/tests/recovery-small.sh +++ b/lustre/tests/recovery-small.sh @@ -2,9 +2,6 @@ set -e -# bug 5493 LU2034 -ALWAYS_EXCEPT="52 60 $RECOVERY_SMALL_EXCEPT" - export MULTIOP=${MULTIOP:-multiop} PTLDEBUG=${PTLDEBUG:--1} LUSTRE=${LUSTRE:-`dirname $0`/..} @@ -13,6 +10,13 @@ init_test_env $@ . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh} init_logging +ALWAYS_EXCEPT="$RECOVERY_SMALL_EXCEPT" +if [ "$MDSCOUNT" -gt "1" ]; then + # bug number for skipped test: LU-10931 + ALWAYS_EXCEPT+=" 136" + # UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT! +fi + require_dsh_mds || exit 0 # also long tests: 19, 21a, 21e, 21f, 23, 27 @@ -211,7 +215,7 @@ test_10c() { awk '{sub("_UUID", "", $2); print $2;}') #assume one client mdccli=$($LCTL dl | grep "${mdtname}-mdc" | awk '{print $4;}') - conn_uuid=$($LCTL get_param -n mdc.${mdccli}.mds_conn_uuid) + conn_uuid=$($LCTL get_param -n mdc.${mdccli}.conn_uuid) mdcpath="mdc.${mdccli}.import=connection=${conn_uuid}" drop_bl_callback_once "chmod 0777 ${workdir}" & @@ -1076,20 +1080,21 @@ test_26a() { # was test_26 bug 5921 - evict dead exports by pinger check_timeout || return 1 - local OST_NEXP=$(do_facet ost1 lctl get_param -n obdfilter.${ost1_svc}.num_exports | cut -d' ' -f2) - - echo starting with $OST_NEXP OST exports # OBD_FAIL_PTLRPC_DROP_RPC 0x505 do_facet client lctl set_param fail_loc=0x505 - # evictor takes PING_EVICT_TIMEOUT + 3 * PING_INTERVAL to evict. - # But if there's a race to start the evictor from various obds, - # the loser might have to wait for the next ping. - + local before=$(date +%s) local rc=0 - wait_client_evicted ost1 $OST_NEXP $((TIMEOUT * 2 + TIMEOUT * 3 / 4)) - rc=$? + + # 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. + sleep $((TIMEOUT * 2 + TIMEOUT * 3 / 4)) do_facet client lctl set_param fail_loc=0x0 - [ $rc -eq 0 ] || error "client not evicted from OST" + do_facet client df > /dev/null + + local oscs=$(lctl dl | awk '/-osc-/ {print $4}') + check_clients_evicted $before ${oscs[@]} + check_clients_full 10 ${oscs[@]} } run_test 26a "evict dead exports" @@ -1287,7 +1292,7 @@ run_test 52 "failover OST under load" # test of open reconstruct test_53() { touch $DIR/$tfile - drop_ldlm_reply "openfile -f O_RDWR:O_CREAT -m 0755 $DIR/$tfile" ||\ + drop_mdt_ldlm_reply "openfile -f O_RDWR:O_CREAT -m 0755 $DIR/$tfile" ||\ return 2 } run_test 53 "touch: drop rep" @@ -1311,6 +1316,13 @@ test_55() { mkdir -p $DIR/$tdir + # This test assumes relatively small max_dirty_mb setting + # which we want to walk away from, so just for it we will + # temporarily lower the value + local max_dirty_mb=$(lctl get_param -n osc.*.max_dirty_mb | head -n 1) + lctl set_param -n osc.*.max_dirty_mb=32 + stack_trap "lctl set_param osc.*.max_dirty_mb=$max_dirty_mb" EXIT + # Minimum pass speed is 2MBps local ddtimeout=64 # LU-2887/LU-3089 - set min pass speed to 500KBps @@ -1435,63 +1447,47 @@ test_59() { # bug 10589 } run_test 59 "Read cancel race on client eviction" -err17935 () { - # we assume that all md changes are in the MDT0 changelog - if [ $MDSCOUNT -gt 1 ]; then - error_ignore bz17935 $* - else - error $* - fi -} - test_60() { - MDT0=$($LCTL get_param -n mdc.*.mds_server_uuid | - awk '{ gsub(/_UUID/,""); print $1 }' | head -n1) - - NUM_FILES=15000 - mkdir -p $DIR/$tdir + local num_files=${COUNT:-5000} + test_mkdir $DIR/$tdir # Register (and start) changelog - USER=$(do_facet $SINGLEMDS lctl --device $MDT0 changelog_register -n) - echo "Registered as $MDT0 changelog user $USER" + changelog_register || error "changelog_register failed" # Generate a large number of changelog entries - createmany -o $DIR/$tdir/$tfile $NUM_FILES + createmany -o $DIR/$tdir/$tfile $num_files sync sleep 5 # Unlink files in the background - unlinkmany $DIR/$tdir/$tfile $NUM_FILES & - CLIENT_PID=$! + unlinkmany $DIR/$tdir/$tfile $num_files & + local client_pid=$! sleep 1 # Failover the MDS while unlinks are happening facet_failover $SINGLEMDS # Wait for unlinkmany to finish - wait $CLIENT_PID + 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" + local cl_count=$(changelog_dump | grep -c UNLNK) + echo "$cl_count unlinks in changelog" + + changelog_deregister || error "changelog_deregister failed" + if ! changelog_users $SINGLEMDS | grep -q "^cl"; then + [ $cl_count -eq $num_files ] || + error "Recorded $cl_count of $num_files unlinks" + # Also make sure we can clear large changelogs + cl_count=$(changelog_dump | wc -l) + [ $cl_count -le 2 ] || + error "Changelog not empty: $cl_count entries" + else # If other users, there may be other unlinks in the log + [ $cl_count -ge $num_files ] || + error "Recorded $cl_count of $num_files unlinks" + changelog_users $SINGLEMDS + echo "other changelog users; can't verify clear" fi } run_test 60 "Add Changelog entries during MDS failover" @@ -1535,6 +1531,39 @@ run_test 61 "Verify to not reuse orphan objects - bug 17025" #} #run_test 62 "Verify connection flags race - bug LU-1716" +test_65() { + mount_client $DIR2 + + #grant lock1, export2 + $SETSTRIPE -i -0 $DIR2/$tfile || return 1 + $MULTIOP $DIR2/$tfile Ow || return 2 + +#define OBD_FAIL_LDLM_BL_EVICT 0x31e + do_facet ost $LCTL set_param fail_loc=0x31e + #get waiting lock2, export1 + $MULTIOP $DIR/$tfile Ow & + PID1=$! + # let enqueue to get asleep + sleep 2 + + #get lock2 blocked + $MULTIOP $DIR2/$tfile Ow & + PID2=$! + sleep 2 + + #evict export1 + ost_evict_client + + sleep 2 + do_facet ost $LCTL set_param fail_loc=0 + + wait $PID1 + wait $PID2 + + umount_client $DIR2 +} +run_test 65 "lock enqueue for destroyed export" + test_66() { [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.7.51) ]] || @@ -1547,7 +1576,7 @@ test_66() # drop 1 reply with UPDATE lock mcreate $DIR/$tfile || error "mcreate failed: $?" - drop_ldlm_reply_once "stat $DIR/$tfile" & + drop_mdt_ldlm_reply_once "stat $DIR/$tfile" & sleep 2 # make the re-sent lock to sleep @@ -1555,8 +1584,8 @@ test_66() do_nodes $list $LCTL set_param fail_loc=0x80000136 #initiate the re-connect & re-send - local mdccli=$($LCTL dl | awk '/-mdc-/ {print $4;}') - local conn_uuid=$($LCTL get_param -n mdc.${mdccli}.mds_conn_uuid) + local mdccli=$($LCTL dl | awk '/-MDT0000-mdc-/ {print $4;}') + local conn_uuid=$($LCTL get_param -n mdc.${mdccli}.conn_uuid) $LCTL set_param "mdc.${mdccli}.import=connection=${conn_uuid}" sleep 2 @@ -1568,38 +1597,24 @@ test_66() } run_test 66 "lock enqueue re-send vs client eviction" -test_65() { - mount_client $DIR2 - - #grant lock1, export2 - $SETSTRIPE -i -0 $DIR2/$tfile || return 1 - $MULTIOP $DIR2/$tfile Ow || return 2 - -#define OBD_FAIL_LDLM_BL_EVICT 0x31e - do_facet ost $LCTL set_param fail_loc=0x31e - #get waiting lock2, export1 - $MULTIOP $DIR/$tfile Ow & - PID1=$! - # let enqueue to get asleep - sleep 2 - - #get lock2 blocked - $MULTIOP $DIR2/$tfile Ow & - PID2=$! - sleep 2 - - #evict export1 - ost_evict_client +test_67() +{ +#define OBD_FAIL_PTLRPC_CONNECT_RACE 0x531 + $LCTL set_param fail_loc=0x80000531 + local mdtname="MDT0000" + local mdccli=$($LCTL dl | grep "${mdtname}-mdc" | awk '{print $4;}') + local conn_uuid=$($LCTL get_param -n mdc.${mdccli}.mds_conn_uuid) + $LCTL set_param "mdc.${mdccli}.import=connection=${conn_uuid}" & sleep 2 - do_facet ost $LCTL set_param fail_loc=0 - wait $PID1 - wait $PID2 + mds_evict_client + sleep 1 - umount_client $DIR2 + client_reconnect + wait } -run_test 65 "lock enqueue for destroyed export" +run_test 67 "connect vs import invalidate race" check_cli_ir_state() { @@ -1745,26 +1760,45 @@ test_100() } run_test 100 "IR: Make sure normal recovery still works w/o IR" -test_101() +test_101a() { - do_facet mgs $LCTL list_param mgs.*.ir_timeout || - { skip "MGS without IR support"; return 0; } + do_facet mgs $LCTL list_param mgs.*.ir_timeout || + skip "MGS without IR support" - set_ir_status full + set_ir_status full - local OST1_IMP=$(get_osc_import_name client ost1) + local ost1_imp=$(get_osc_import_name client ost1) - # disable pinger recovery - lctl set_param -n osc.$OST1_IMP.pinger_recov=0 + # disable pinger recovery + lctl set_param -n osc.$ost1_imp.pinger_recov=0 + stack_trap "$LCTL set_param -n osc.$ost1_imp.pinger_recov=1" EXIT - fail ost1 + fail ost1 - target_instance_match ost1 || error "instance mismatch" - nidtbl_versions_match || error "version must match" + target_instance_match ost1 || error "instance mismatch" + nidtbl_versions_match || error "version must match" +} +run_test 101a "IR: Make sure IR works w/o normal recovery" + +test_101b() +{ + do_facet mgs $LCTL list_param mgs.*.ir_timeout || + skip "MGS without IR support" + + set_ir_status full + + local ost1_imp=$(get_osc_import_name client ost1) - lctl set_param -n osc.$OST1_IMP.pinger_recov=1 +#define OBD_FAIL_OST_PREPARE_DELAY 0x247 + do_facet ost1 "$LCTL set_param fail_loc=0x247" + # disable pinger recovery + $LCTL set_param -n osc.$ost1_imp.pinger_recov=0 + stack_trap "$LCTL set_param -n osc.$ost1_imp.pinger_recov=1" EXIT + +#OST may return EAGAIN if it is not configured yet + fail ost1 } -run_test 101 "IR: Make sure IR works w/o normal recovery" +run_test 101b "IR: Make sure IR works w/o normal recovery and proceed EAGAIN" test_102() { @@ -1820,7 +1854,7 @@ 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 + combined_mgs_mds && skip "needs separate mgs and mds" && return 0 # workaround solution to generate config log on the mds remount_facet mds1 @@ -1954,17 +1988,17 @@ test_106() { # LU-1789 touch $DIR2/$tfile || error "failed to create empty file" replay_barrier $SINGLEMDS - $LCTL set_param debug=console + $LCTL set_param debug=ha $LCTL clear facet_failover $SINGLEMDS - # lightweight connection must be evicted + # lightweight goes through LUSTRE_IMP_RECOVER during failover touch -c $DIR2/$tfile || true $LCTL dk $TMP/lustre-log-$TESTNAME.log - evicted=`awk '/This client was evicted by .*MDT0000/ { + recovered=`awk '/MDT0000-mdc-[0-9a-f]*: lwp recover/ { print; }' $TMP/lustre-log-$TESTNAME.log` - [ -z "$evicted" ] && error "lightweight client not evicted by mds" + [ -z "$recovered" ] && error "lightweight client was not recovered" # and all operations performed by lightweight client should be # synchronous, so the file created before mds restart should be there @@ -2019,21 +2053,21 @@ run_test 108 "client eviction don't crash" test_110a () { [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0 local remote_dir=$DIR/$tdir/remote_dir - local MDTIDX=1 + local mdtidx=1 local num #prepare for 110 test, which need set striped dir on remote MDT. for num in $(seq $MDSCOUNT); do do_facet mds$num \ - lctl set_param -n mdt.${FSNAME}*.enable_remote_dir=1 \ + lctl set_param -n mdt.$FSNAME*.enable_remote_dir=1 \ 2>/dev/null done mkdir -p $DIR/$tdir - drop_request "$LFS mkdir -i $MDTIDX -c2 $remote_dir" || - error "lfs mkdir failed" - local diridx=$($GETSTRIPE -M $remote_dir) - [ $diridx -eq $MDTIDX ] || error "$diridx != $MDTIDX" + drop_request "$LFS mkdir -i $mdtidx -c2 $remote_dir" || + error "lfs mkdir failed" + local diridx=$($LFS getstripe -m $remote_dir) + [ $diridx -eq $mdtidx ] || error "$diridx != $mdtidx" rm -rf $DIR/$tdir || error "rmdir failed" } @@ -2042,14 +2076,14 @@ run_test 110a "create remote directory: drop client req" test_110b () { [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0 local remote_dir=$DIR/$tdir/remote_dir - local MDTIDX=1 + local mdtidx=1 mkdir -p $DIR/$tdir - drop_reint_reply "$LFS mkdir -i $MDTIDX -c2 $remote_dir" || - error "lfs mkdir failed" + drop_reint_reply "$LFS mkdir -i $mdtidx -c2 $remote_dir" || + error "lfs mkdir failed" - diridx=$($GETSTRIPE -M $remote_dir) - [ $diridx -eq $MDTIDX ] || error "$diridx != $MDTIDX" + diridx=$($LFS getstripe -m $remote_dir) + [ $diridx -eq $mdtidx ] || error "$diridx != $mdtidx" rm -rf $DIR/$tdir || error "rmdir failed" } @@ -2058,14 +2092,14 @@ run_test 110b "create remote directory: drop Master rep" test_110c () { [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0 local remote_dir=$DIR/$tdir/remote_dir - local MDTIDX=1 + local mdtidx=1 mkdir -p $DIR/$tdir - drop_update_reply $MDTIDX "$LFS mkdir -i $MDTIDX -c2 $remote_dir" || - error "lfs mkdir failed" + drop_update_reply $mdtidx "$LFS mkdir -i $mdtidx -c2 $remote_dir" || + error "lfs mkdir failed" - diridx=$($GETSTRIPE -M $remote_dir) - [ $diridx -eq $MDTIDX ] || error "$diridx != $MDTIDX" + diridx=$($GETSTRIPE -m $remote_dir) + [ $diridx -eq $mdtidx ] || error "$diridx != $mdtidx" rm -rf $DIR/$tdir || error "rmdir failed" } @@ -2117,27 +2151,27 @@ test_110f () { run_test 110f "remove remote directory: drop slave rep" test_110g () { - [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.6.57) ]] || - { skip "Need MDS version at least 2.6.57"; return 0; } + [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.11.0) ]] || + { skip "Need MDS version at least 2.11.0"; return 0; } [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0 - local remote_dir=$DIR/$tdir/remote_dir - local MDTIDX=1 - - mkdir -p $remote_dir - createmany -o $remote_dir/f 100 + mkdir -p $DIR/$tdir + touch $DIR/$tdir/$tfile - #define OBD_FAIL_MIGRATE_NET_REP 0x1800 - do_facet mds$MDTIDX lctl set_param fail_loc=0x1800 - $LFS migrate -m $MDTIDX $remote_dir || error "migrate failed" - do_facet mds$MDTIDX lctl set_param fail_loc=0x0 + # OBD_FAIL_MDS_REINT_NET_REP 0x119 + do_facet mds1 $LCTL set_param fail_loc=0x119 + $LFS migrate -m 1 $DIR/$tdir & + migrate_pid=$! + sleep 5 + do_facet mds1 $LCTL set_param fail_loc=0 + wait $migrate_pid - for file in $(find $remote_dir); do - mdt_index=$($LFS getstripe -M $file) - [ $mdt_index == $MDTIDX ] || - error "$file is not on MDT${MDTIDX}" - done + local mdt_index + mdt_index=$($LFS getstripe -m $DIR/$tdir) + [ $mdt_index == 1 ] || error "$tdir is not on MDT1" + mdt_index=$($LFS getstripe -m $DIR/$tdir/$tfile) + [ $mdt_index == 1 ] || error "$tfile is not on MDT1" rm -rf $DIR/$tdir || error "rmdir failed" } @@ -2145,6 +2179,10 @@ run_test 110g "drop reply during migration" test_110h () { [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0 + local server_version=$(lustre_version_code mds1) + [[ $server_version -ge $(version_code 2.7.56) ]] || + { skip "Need MDS version at least 2.7.56"; return 0; } + local src_dir=$DIR/$tdir/source_dir local tgt_dir=$DIR/$tdir/target_dir local MDTIDX=1 @@ -2169,6 +2207,10 @@ run_test 110h "drop update reply during cross-MDT file rename" test_110i () { [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0 + local server_version=$(lustre_version_code mds1) + [[ $server_version -ge $(version_code 2.7.56) ]] || + { skip "Need MDS version at least 2.7.56"; return 0; } + local src_dir=$DIR/$tdir/source_dir local tgt_dir=$DIR/$tdir/target_dir local MDTIDX=1 @@ -2196,6 +2238,10 @@ run_test 110i "drop update reply during cross-MDT dir rename" test_110j () { [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0 + local server_version=$(lustre_version_code mds1) + [[ $server_version -ge $(version_code 2.7.56) ]] || + { skip "Need MDS version at least 2.7.56"; return 0; } + local remote_dir=$DIR/$tdir/remote_dir local local_dir=$DIR/$tdir/local_dir local MDTIDX=1 @@ -2220,13 +2266,14 @@ test_111 () [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.62) ]] || { skip "Need MDS version at least 2.3.62"; return 0; } - local mdsdev=$(mdsdevname ${SINGLEMDS//mds/}) #define OBD_FAIL_MDS_CHANGELOG_INIT 0x151 do_facet $SINGLEMDS lctl set_param fail_loc=0x151 stop $SINGLEMDS || error "stop MDS failed" - start $SINGLEMDS $mdsdev && error "start MDS should fail" + start $SINGLEMDS $(mdsdevname ${SINGLEMDS//mds/}) && + error "start MDS should fail" do_facet $SINGLEMDS lctl set_param fail_loc=0 - start $SINGLEMDS $mdsdev || error "start MDS failed" + start $SINGLEMDS $(mdsdevname ${SINGLEMDS//mds/}) || + error "start MDS failed" } run_test 111 "mdd setup fail should not cause umount oops" @@ -2481,7 +2528,7 @@ test_113() { # drop 1 reply with UPDATE lock, # resend should not create 2nd lock on server mcreate $DIR/$tfile || error "mcreate failed: $?" - drop_ldlm_reply_once "stat $DIR/$tfile" || error "stat failed: $?" + drop_mdt_ldlm_reply_once "stat $DIR/$tfile" || error "stat failed: $?" # 2 BL AST will be sent to client, both must find the same lock, # race them to not get EINVAL for 2nd BL AST @@ -2673,8 +2720,8 @@ test_133() { multiop_bg_pause $DIR/$tfile O_jc || return 1 PID=$! - #define OBD_FAIL_LDLM_REPLY 0x30c - do_nodes $list $LCTL set_param fail_loc=0x8000030c + #define OBD_FAIL_MDS_LDLM_REPLY_NET 0x157 + do_nodes $list $LCTL set_param fail_loc=0x80000157 kill -USR1 $PID echo "waiting for multiop $PID" wait $PID || return 2 @@ -2685,6 +2732,106 @@ test_133() { } run_test 133 "don't fail on flock resend" +test_134() { + [ -z "$CLIENTS" ] && skip "Need two or more clients" && return + [ $CLIENTCOUNT -lt 2 ] && + { skip "Need 2+ clients, have $CLIENTCOUNT" && return; } + + mkdir -p $MOUNT/$tdir/1 $MOUNT/$tdir/2 || error "mkdir failed" + touch $MOUNT/$tdir/1/$tfile $MOUNT/$tdir/2/$tfile || + error "touch failed" + zconf_umount_clients $CLIENTS $MOUNT + zconf_mount_clients $CLIENTS $MOUNT + +#define OBD_FAIL_TGT_REPLY_DATA_RACE 0x722 + # assume commit interval is 5 + do_facet mds1 "$LCTL set_param fail_loc=0x722 fail_val=5" + + local -a clients=(${CLIENTS//,/ }) + local client1=${clients[0]} + local client2=${clients[1]} + + do_node $client1 rm $MOUNT/$tdir/1/$tfile & + rmpid=$! + do_node $client2 mv $MOUNT/$tdir/2/$tfile $MOUNT/$tdir/2/${tfile}_2 & + mvpid=$! + fail mds1 + wait $rmpid || error "rm failed" + wait $mvpid || error "mv failed" + return 0 +} +run_test 134 "race between failover and search for reply data free slot" + +test_135() { + [ $MDS1_VERSION -lt $(version_code 2.12.51) ] && + skip "Need MDS version at least 2.12.51" + + mkdir -p $DIR/$tdir + $LFS setstripe -E 1M -L mdt $DIR/$tdir + # to have parent dir write lock before open/resend + touch $DIR/$tdir/$tfile + #define OBD_FAIL_MDS_LDLM_REPLY_NET 0x157 + do_nodes $(comma_list $(mdts_nodes)) $LCTL set_param fail_loc=0x80000157 + openfile -f O_RDWR:O_CREAT -m 0755 $DIR/$tdir/$tfile || + error "Failed to open DOM file" +} +run_test 135 "DOM: open/create resend to return size" + +test_136() { + remote_mds_nodsh && skip "remote MDS with nodsh" && return + [[ $MDS1_VERSION -ge $(version_code 2.12.52) ]] || + skip "Need MDS version at least 2.12.52" + + local mdts=$(comma_list $(mdts_nodes)) + local MDT0=$(facet_svc $SINGLEMDS) + + local clog=$(do_facet mds1 $LCTL --device $MDT0 changelog_register -n) + [ -n "$clog" ] || error "changelog_register failed" + cl_mask=$(do_facet mds1 $LCTL get_param \ + mdd.$MDT0.changelog_mask -n) + changelog_chmask "ALL" + + # generate some changelog records to accumulate + test_mkdir -i 0 -c 0 $DIR/$tdir || error "mkdir $tdir failed" + createmany -m $DIR/$tdir/$tfile 10000 || + error "create $DIR/$tdir/$tfile failed" + + local size1=$(do_facet $SINGLEMDS \ + $LCTL get_param -n mdd.$MDT0.changelog_size) + echo "Changelog size $size1" + + #define OBD_FAIL_LLOG_PURGE_DELAY 0x1318 + do_nodes $mdts $LCTL set_param fail_loc=0x1318 fail_val=30 + + # launch changelog_deregister in background on MDS + do_facet mds1 "nohup $LCTL --device $MDT0 changelog_deregister $clog \ + > foo.out 2> foo.err < /dev/null &" + # give time to reach fail_loc + sleep 15 + + # fail_loc will make MDS sleep in the middle of changelog_deregister + # take this opportunity to abruptly kill MDS + FAILURE_MODE_save=$FAILURE_MODE + FAILURE_MODE=HARD + fail mds1 + FAILURE_MODE=$FAILURE_MODE_save + + do_nodes $mdts $LCTL set_param fail_loc=0x0 fail_val=0 + + local size2=$(do_facet $SINGLEMDS \ + $LCTL get_param -n mdd.$MDT0.changelog_size) + echo "Changelog size $size2" + local clog2=$(do_facet $SINGLEMDS "$LCTL get_param -n \ + mdd.$MDT0.changelog_users | grep $clog") + echo "After crash, changelog user $clog2" + + [ -n "$clog2" -o $size2 -lt $size1 ] || + error "changelog record count unchanged" + + do_facet mds1 $LCTL set_param mdd.$MDT0.changelog_mask=\'$cl_mask\' -n +} +run_test 136 "changelog_deregister leaving pending records" + complete $SECONDS check_and_cleanup_lustre exit_status