X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Ftests%2Freplay-dual.sh;h=820a97b584e1929d31ff00d70b8c9b842c7ae05c;hb=a64f43dfeb7b30b5f94cee813f57845c76feeeac;hp=bd5a748006613cfb000a0253499332ddcb4e6c3e;hpb=fbb7ead129258897f5a5d5c9ce28d31fbbe5bca2;p=fs%2Flustre-release.git

diff --git a/lustre/tests/replay-dual.sh b/lustre/tests/replay-dual.sh
index bd5a748..820a97b 100755
--- a/lustre/tests/replay-dual.sh
+++ b/lustre/tests/replay-dual.sh
@@ -1,117 +1,49 @@
-#!/bin/sh
+#!/bin/bash
 
 set -e
 
+# bug number:  6088 10124 10800
+ALWAYS_EXCEPT="8    15c   17    $REPLAY_DUAL_EXCEPT"
+
+PTLDEBUG=${PTLDEBUG:--1}
 LUSTRE=${LUSTRE:-`dirname $0`/..}
 . $LUSTRE/tests/test-framework.sh
 
 init_test_env $@
 
-. ${CONFIG:=$LUSTRE/tests/cfg/lmv.sh}
+. ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
 
 SETUP=${SETUP:-"setup"}
 CLEANUP=${CLEANUP:-"cleanup"}
 
-gen_config() {
-    rm -f $XMLCONFIG
-    if [ "$MDSCOUNT" -gt 1 ]; then
-        add_lmv lmv1_svc
-        for mds in `mds_list`; do
-            MDSDEV=$TMP/${mds}-`hostname`
-            add_mds $mds --dev $MDSDEV --size $MDSSIZE --lmv lmv1_svc
-        done
-        add_lov_to_lmv lov1 lmv1_svc --stripe_sz $STRIPE_BYTES \
-	    --stripe_cnt $STRIPES_PER_OBJ --stripe_pattern 0
-	MDS=lmv1
-    else
-        add_mds mds1 --dev $MDSDEV --size $MDSSIZE
-        add_lov lov1 mds1 --stripe_sz $STRIPE_BYTES \
-	    --stripe_cnt $STRIPES_PER_OBJ --stripe_pattern 0
-	MDS=mds1_svc
-    fi
-
-    add_ost ost --lov lov1 --dev $OSTDEV --size $OSTSIZE --failover
-    add_ost ost2 --lov lov1 --dev ${OSTDEV}-2 --size $OSTSIZE  --failover
-    add_client client  ${MDS} --lov lov1 --path $MOUNT
-}
-
 build_test_filter
 
-SETUP=${SETUP:-"setup"}
-CLEANUP=${CLEANUP:-"cleanup"}
-
 cleanup() {
-    # make sure we are using the primary MDS, so the config log will
-    # be able to clean up properly.
-    activemds=`facet_active mds1`
-    if [ $activemds != "mds1" ]; then
-        fail mds1
-    fi
-
-    umount $MOUNT2 || true
-    umount $MOUNT || true
-    rmmod llite
-    
-    # b=3941
-    # In mds recovery, the mds will clear orphans in ost by 
-    # mds_lov_clear_orphan, which will sent the request to ost and waiting for
-    # the reply, if we stop mds at this time, we will got the obd_refcount > 1 
-    # errors, because mds_lov_clear_orphan grab a export of mds, 
-    # so the obd_refcount of mds will not be zero. So, wait a while before
-    # stop mds. This bug needs further work.
-    for mds in `mds_list`; do
-	sleep 5
-	stop $mds ${FORCE} $MDSLCONFARGS
-    done
-    stop_lgssd
-    stop_lsvcgssd
-    stop ost2 ${FORCE}
-    stop ost ${FORCE}  --dump cleanup-dual.log
+    stopall
 }
 
 if [ "$ONLY" == "cleanup" ]; then
-    sysctl -w portals.debug=0
+    sysctl -w lnet.debug=0
     cleanup
     exit
 fi
 
 setup() {
-    gen_config
-
-    start_krb5_kdc || exit 1
-    start ost --reformat $OSTLCONFARGS 
-    PINGER=`cat /proc/fs/lustre/pinger`
-
-    if [ "$PINGER" != "on" ]; then
-	echo "ERROR: Lustre must be built with --enable-pinger for replay-dual"
-	stop ost
-	exit 1
-    fi
-
-    start ost2 --reformat $OSTLCONFARGS 
-    start_lsvcgssd || exit 2
-    start_lgssd || exit 3
-    [ "$DAEMONFILE" ] && $LCTL debug_daemon start $DAEMONFILE $DAEMONSIZE
-    for mds in `mds_list`; do
-	start $mds --reformat $MDSLCONFARGS
-    done
-    grep " $MOUNT " /proc/mounts || zconf_mount `hostname` $MOUNT
-    grep " $MOUNT2 " /proc/mounts || zconf_mount `hostname` $MOUNT2
-
-    echo $TIMEOUT > /proc/sys/lustre/timeout
-    echo $UPCALL > /proc/sys/lustre/upcall
+    cleanup
+    formatall
+    setupall
+    mount_client $MOUNT2
 }
 
 $SETUP
 [ "$DAEMONFILE" ] && $LCTL debug_daemon start $DAEMONFILE $DAEMONSIZE
 
-
 test_1() {
     touch $MOUNT1/a
-    replay_barrier mds1
+    replay_barrier mds
     touch $MOUNT2/b
 
-    fail mds1
+    fail mds
     checkstat $MOUNT2/a || return 1
     checkstat $MOUNT1/b || return 2
     rm $MOUNT2/a $MOUNT1/b
@@ -124,24 +56,23 @@ run_test 1 "|X| simple create"
 
 
 test_2() {
-    replay_barrier mds1
+    replay_barrier mds
     mkdir $MOUNT1/adir
 
-    fail mds1
+    fail mds
     checkstat $MOUNT2/adir || return 1
     rmdir $MOUNT2/adir
     checkstat $MOUNT2/adir && return 2
     return 0
 }
-
 run_test 2 "|X| mkdir adir"
 
 test_3() {
-    replay_barrier mds1
+    replay_barrier mds
     mkdir $MOUNT1/adir
     mkdir $MOUNT2/adir/bdir
 
-    fail mds1
+    fail mds
     checkstat $MOUNT2/adir      || return 1
     checkstat $MOUNT1/adir/bdir || return 2
     rmdir $MOUNT2/adir/bdir $MOUNT1/adir
@@ -149,16 +80,15 @@ test_3() {
     checkstat $MOUNT2/adir/bdir && return 4
     return 0
 }
-
 run_test 3 "|X| mkdir adir, mkdir adir/bdir "
 
 test_4() {
     mkdir $MOUNT1/adir
-    replay_barrier mds1
+    replay_barrier mds
     mkdir $MOUNT1/adir  && return 1
     mkdir $MOUNT2/adir/bdir
 
-    fail mds1
+    fail mds
     checkstat $MOUNT2/adir      || return 2
     checkstat $MOUNT1/adir/bdir || return 3
 
@@ -167,7 +97,6 @@ test_4() {
     checkstat $MOUNT2/adir/bdir && return 5
     return 0
 }
-
 run_test 4 "|X| mkdir adir (-EEXIST), mkdir adir/bdir "
 
 
@@ -179,11 +108,11 @@ test_5() {
     # give multiop a chance to open
     sleep 1 
     rm -f $MOUNT1/a
-    replay_barrier mds1
+    replay_barrier mds
     kill -USR1 $pid
     wait $pid || return 1
 
-    fail mds1
+    fail mds
     [ -e $MOUNT2/a ] && return 2
     return 0
 }
@@ -199,11 +128,11 @@ test_6() {
     # give multiop a chance to open
     sleep 1 
     rm -f $MOUNT1/a
-    replay_barrier mds1
+    replay_barrier mds
     kill -USR1 $pid1
     wait $pid1 || return 1
 
-    fail mds1
+    fail mds
     kill -USR1 $pid2
     wait $pid2 || return 1
     [ -e $MOUNT2/a ] && return 2
@@ -211,31 +140,10 @@ test_6() {
 }
 run_test 6 "open1, open2, unlink |X| close1 [fail mds] close2"
 
-test_6b() {
-    mcreate $MOUNT1/a
-    multiop $MOUNT2/a o_c &
-    pid1=$!
-    multiop $MOUNT1/a o_c &
-    pid2=$!
-    # give multiop a chance to open
-    sleep 1
-    rm -f $MOUNT1/a
-    replay_barrier mds1
-    kill -USR1 $pid2
-    wait $pid2 || return 1
-
-    fail mds1
-    kill -USR1 $pid1
-    wait $pid1 || return 1
-    [ -e $MOUNT2/a ] && return 2
-    return 0
-}
-run_test 6b "open1, open2, unlink |X| close2 [fail mds] close1"
-
 test_8() {
-    replay_barrier mds1
+    replay_barrier mds
     drop_reint_reply "mcreate $MOUNT1/$tfile"    || return 1
-    fail mds1
+    fail mds
     checkstat $MOUNT2/$tfile || return 2
     rm $MOUNT1/$tfile || return 3
 
@@ -244,12 +152,12 @@ test_8() {
 run_test 8 "replay of resent request"
 
 test_9() {
-    replay_barrier mds1
+    replay_barrier mds
     mcreate $MOUNT1/$tfile-1
     mcreate $MOUNT2/$tfile-2
     # drop first reint reply
     sysctl -w lustre.fail_loc=0x80000119
-    fail mds1
+    fail mds
     sysctl -w lustre.fail_loc=0
 
     rm $MOUNT1/$tfile-[1,2] || return 1
@@ -260,12 +168,12 @@ run_test 9 "resending a replayed create"
 
 test_10() {
     mcreate $MOUNT1/$tfile-1
-    replay_barrier mds1
+    replay_barrier mds
     munlink $MOUNT1/$tfile-1
     mcreate $MOUNT2/$tfile-2
     # drop first reint reply
     sysctl -w lustre.fail_loc=0x80000119
-    fail mds1
+    fail mds
     sysctl -w lustre.fail_loc=0
 
     checkstat $MOUNT1/$tfile-1 && return 1
@@ -277,7 +185,7 @@ test_10() {
 run_test 10 "resending a replayed unlink"
 
 test_11() {
-    replay_barrier mds1
+    replay_barrier mds
     mcreate $MOUNT1/$tfile-1
     mcreate $MOUNT2/$tfile-2
     mcreate $MOUNT1/$tfile-3
@@ -285,11 +193,15 @@ test_11() {
     mcreate $MOUNT1/$tfile-5
     # drop all reint replies for a while
     sysctl -w lustre.fail_loc=0x0119
-    facet_failover mds1
+    facet_failover mds
     #sleep for while, let both clients reconnect and timeout
     sleep $((TIMEOUT * 2))
     sysctl -w lustre.fail_loc=0
-
+    while [ -z "$(ls $MOUNT1/$tfile-[1-5] 2>/dev/null)" ]; do
+	sleep 5
+	echo -n "."
+    done
+    ls $MOUNT1/$tfile-[1-5]
     rm $MOUNT1/$tfile-[1-5] || return 1
 
     return 0
@@ -297,22 +209,22 @@ test_11() {
 run_test 11 "both clients timeout during replay"
 
 test_12() {
-    replay_barrier mds1
+    replay_barrier mds
 
     multiop $DIR/$tfile mo_c &
     MULTIPID=$!
     sleep 5
 
-    # drop first enqueue
+#define OBD_FAIL_LDLM_ENQUEUE            0x302
     sysctl -w lustre.fail_loc=0x80000302
-    facet_failover mds1
+    facet_failover mds
     df $MOUNT || return 1
     sysctl -w lustre.fail_loc=0
 
     ls $DIR/$tfile
-    $CHECKSTAT -t file $DIR/$tfile || return 2
     kill -USR1 $MULTIPID || return 3
     wait $MULTIPID || return 4
+    $CHECKSTAT -t file $DIR/$tfile || return 2
     rm $DIR/$tfile
 
     return 0
@@ -324,14 +236,14 @@ test_13() {
     MULTIPID=$!
     sleep 5
 
-    replay_barrier mds1
+    replay_barrier mds
 
     kill -USR1 $MULTIPID || return 3
     wait $MULTIPID || return 4
 
     # drop close 
     sysctl -w lustre.fail_loc=0x80000115
-    facet_failover mds1
+    facet_failover mds
     df $MOUNT || return 1
     sysctl -w lustre.fail_loc=0
 
@@ -343,22 +255,19 @@ test_13() {
 }
 run_test 13 "close resend timeout"
 
-
 test_14() {
-    replay_barrier mds1
+    replay_barrier mds
     createmany -o $MOUNT1/$tfile- 25
     createmany -o $MOUNT2/$tfile-2- 1
     createmany -o $MOUNT1/$tfile-3- 25
     umount $MOUNT2
 
-    facet_failover mds1
+    facet_failover mds
     # expect failover to fail
     df $MOUNT && return 1
     sleep 1
 
-    # first 25 files shouuld have been
-    # replayed
-    sleep 2 
+    # first 25 files should have been replayed 
     unlinkmany $MOUNT1/$tfile- 25 || return 2
 
     zconf_mount `hostname` $MOUNT2
@@ -367,32 +276,113 @@ test_14() {
 run_test 14 "timeouts waiting for lost client during replay"
 
 test_15() {
-    replay_barrier mds1
+    replay_barrier mds
     createmany -o $MOUNT1/$tfile- 25
     createmany -o $MOUNT2/$tfile-2- 1
     umount $MOUNT2
 
-    facet_failover mds1
+    facet_failover mds
     df $MOUNT || return 1
-    sleep 1
 
     unlinkmany $MOUNT1/$tfile- 25 || return 2
+    [ -e $MOUNT1/$tfile-2-0 ] && error "$tfile-2-0 exists"
 
     zconf_mount `hostname` $MOUNT2
     return 0
 }
 run_test 15 "timeout waiting for lost client during replay, 1 client completes"
+
+test_15a() {
+    local ost_last_id=""
+    local osc_last_id=""
+    
+    replay_barrier mds
+    echo "data" > "$MOUNT2/${tfile}-m2"
+
+    umount $MOUNT2
+    facet_failover mds
+    df $MOUNT || return 1
+    
+    ost_last_id=`cat /proc/fs/lustre/obdfilter/*/last_id`
+    mds_last_id=`cat /proc/fs/lustre/osc/*mds*/last_id`
+    
+    echo "Ids after MDS<->OST synchonizing"
+    echo "--------------------------------"
+    echo "MDS last_id:"
+    echo $mds_last_id
+    echo "OST last_id:"
+    echo $ost_last_id
+
+    local i=0
+    echo $ost_last_id | while read id; do
+	ost_ids[$i]=$id
+	((i++))
+    done
+    
+    i=0
+    echo $mds_last_id | while read id; do
+	mds_ids[$i]=$id
+	((i++))
+    done
+    
+    local arr_len=${#mds_ids[*]}
+    for ((i=0;i<$arr_len;i++)); do
+	    mds_id=${mds_ids[i]}
+	    ost_id=${ost_ids[i]}
+	    
+	    test $mds_id -ge $ost_id || {
+		echo "MDS last id ($mds_id) is smaller than OST one ($ost_id)"
+		return 2
+	    }
+    done
+
+    zconf_mount `hostname` $MOUNT2
+    return 0
+}
+#CROW run_test 15a "OST clear orphans - synchronize ids on MDS and OST"
+
+test_15b() {
+    replay_barrier mds
+    echo "data" > "$MOUNT2/${tfile}-m2"
+    umount $MOUNT2
+
+    do_facet ost1 "sysctl -w lustre.fail_loc=0x80000802"
+    facet_failover mds
+
+    df $MOUNT || return 1
+    do_facet ost1 "sysctl -w lustre.fail_loc=0"
+    
+    zconf_mount `hostname` $MOUNT2
+    return 0
+}
+#CROW run_test 15b "multiple delayed OST clear orphans"
+
+test_15c() {
+    replay_barrier mds
+    for ((i = 0; i < 2000; i++)); do
+	echo "data" > "$MOUNT2/${tfile}-$i" || error "create ${tfile}-$i failed"
+    done
+    
+    umount $MOUNT2
+    facet_failover mds
+
+    df $MOUNT || return 1
+    
+    zconf_mount `hostname` $MOUNT2
+    return 0
+}
+run_test 15c "remove multiple OST orphans"
+
 test_16() {
-    replay_barrier mds1
+    replay_barrier mds
     createmany -o $MOUNT1/$tfile- 25
     createmany -o $MOUNT2/$tfile-2- 1
     umount $MOUNT2
 
-    facet_failover mds1
+    facet_failover mds
     sleep $TIMEOUT
-    facet_failover mds1
+    facet_failover mds
     df $MOUNT || return 1
-    sleep 1
 
     unlinkmany $MOUNT1/$tfile- 25 || return 2
 
@@ -400,22 +390,20 @@ test_16() {
     return 0
 
 }
-#run_test 16 "fail MDS during recovery (3571)"
+run_test 16 "fail MDS during recovery (3571)"
 
 test_17() {
     createmany -o $MOUNT1/$tfile- 25
     createmany -o $MOUNT2/$tfile-2- 1
 
     # Make sure the disconnect is lost
-    replay_barrier ost
+    replay_barrier ost1
     umount $MOUNT2
 
-    echo -1 > /proc/sys/portals/debug
-    facet_failover ost
+    facet_failover ost1
     sleep $TIMEOUT
-    facet_failover ost
+    facet_failover ost1
     df $MOUNT || return 1
-    sleep 1
 
     unlinkmany $MOUNT1/$tfile- 25 || return 2
 
@@ -423,55 +411,47 @@ test_17() {
     return 0
 
 }
-#Still not support ost fail over
-#run_test 17 "fail OST during recovery (3571)"
+run_test 17 "fail OST during recovery (3571)"
 
-test_18 () {
-    replay_barrier mds1
-    multiop $MOUNT2/$tfile O_c &
-    pid2=$!
-    multiop $MOUNT1/$tfile O_c &
-    pid1=$!
-    # give multiop a chance to open
-    sleep 1 
-    kill -USR1 $pid2
-    kill -USR1 $pid1
+# cleanup with blocked enqueue fails until timer elapses (MDS busy), wait for it
+export NOW=0
+
+test_18() { # bug 3822 - evicting client with enqueued lock
+    #set -vx
+    mkdir -p $MOUNT1/$tdir
+    touch $MOUNT1/$tdir/f0
+#define OBD_FAIL_LDLM_ENQUEUE_BLOCKED    0x30b
+    statmany -s $MOUNT1/$tdir/f 1 500 &
+    OPENPID=$!
+    NOW=`date +%s`
+    do_facet mds sysctl -w lustre.fail_loc=0x8000030b  # hold enqueue
     sleep 1
-    umount $MOUNT2
-    facet_failover mds1
-    df || df ||  return 1
-    zconf_mount `hostname` $MOUNT2
+#define OBD_FAIL_LDLM_BL_CALLBACK        0x305
+    do_facet client sysctl -w lustre.fail_loc=0x80000305  # drop cb, evict
+    cancel_lru_locks mdc
+    usleep 500 # wait to ensure first client is one that will be evicted
+    openfile -f O_RDONLY $MOUNT2/$tdir/f0
+    wait $OPENPID
+    dmesg | grep "entering recovery in server" && \
+        error "client not evicted" || true
 }
-run_test 18 "replay open, Abort recovery, don't assert (3892)"
+run_test 18 "ldlm_handle_enqueue succeeds on evicted export (3822)"
 
-# cleanup with blocked enqueue fails until timer elapses (MDS busy), wait for
-# itexport NOW=0
+test_19() { # Bug 10991 - resend of open request does not fail assertion.
+    replay_barrier mds
+    drop_ldlm_reply "createmany -o $DIR/$tfile 1" || return 1
+    fail mds
+    checkstat $DIR2/${tfile}0 || return 2
+    rm $DIR/${tfile}0 || return 3
 
-test_20() {     # bug 3822 - evicting client with enqueued lock
-       mkdir -p $MOUNT1/$tdir
-       touch $MOUNT1/$tdir/f0
-#define OBD_FAIL_LDLM_ENQUEUE_BLOCKED    0x30b
-       statmany -s $MOUNT1/$tdir/f 500 &
-       OPENPID=$!
-       NOW=`date +%s`
-       do_facet mds1 sysctl -w lustre.fail_loc=0x8000030b  # hold enqueue
-       sleep 1
-#define OBD_FAIL_LDLM_BL_CALLBACK        0x305
-       do_facet client sysctl -w lustre.fail_loc=0x80000305  # drop cb, evict
-       cancel_lru_locks MDC
-       usleep 500 # wait to ensure first client is one that will be evicted
-       openfile -f O_RDONLY $MOUNT2/$tdir/f0
-       wait $OPENPID
-       dmesg | grep "entering recovery in server" && \
-               error "client not evicted" || true
+    return 0
 }
-run_test 20 "ldlm_handle_enqueue succeeds on evicted export (3822)"
+run_test 19 "resend of open request"
 
 if [ "$ONLY" != "setup" ]; then
-	equals_msg test complete, cleaning up
-	if [ $NOW ]; then
-            SLEEP=$((`date +%s` - $NOW))
-            [ $SLEEP -lt $TIMEOUT ] && sleep $SLEEP
-	fi
-	$CLEANUP
+   equals_msg test complete, cleaning up
+   SLEEP=$((`date +%s` - $NOW))
+   [ $SLEEP -lt $TIMEOUT ] && sleep $SLEEP
+   $CLEANUP
 fi
+echo "$0: completed"