replay-single.sh

lustre 1.6.5 source code

test_25() {
    multiop_bg_pause $DIR/$tfile O_tSc || return 3
    pid=$!
    rm -f $DIR/$tfile

    replay_barrier mds
    fail mds
    kill -USR1 $pid
    wait $pid || return 1
    [ -e $DIR/$tfile ] && return 2
    return 0
}
run_test 25 "open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)"

test_26() {
    replay_barrier mds
    multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
    pid1=$!
    multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
    pid2=$!
    rm -f $DIR/$tfile-1
    rm -f $DIR/$tfile-2
    kill -USR1 $pid2
    wait $pid2 || return 1

    fail mds
    kill -USR1 $pid1
    wait $pid1 || return 2
    [ -e $DIR/$tfile-1 ] && return 3
    [ -e $DIR/$tfile-2 ] && return 4
    return 0
}
run_test 26 "|X| open(O_CREAT), unlink two, close one, replay, close one (test mds_cleanup_orphans)"

test_27() {
    replay_barrier mds
    multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
    pid1=$!
    multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
    pid2=$!
    rm -f $DIR/$tfile-1
    rm -f $DIR/$tfile-2

    fail mds
    kill -USR1 $pid1
    wait $pid1 || return 1
    kill -USR1 $pid2
    wait $pid2 || return 2
    [ -e $DIR/$tfile-1 ] && return 3
    [ -e $DIR/$tfile-2 ] && return 4
    return 0
}
run_test 27 "|X| open(O_CREAT), unlink two, replay, close two (test mds_cleanup_orphans)"

test_28() {
    multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
    pid1=$!
    multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
    pid2=$!
    replay_barrier mds
    rm -f $DIR/$tfile-1
    rm -f $DIR/$tfile-2
    kill -USR1 $pid2
    wait $pid2 || return 1

    fail mds
    kill -USR1 $pid1
    wait $pid1 || return 2
    [ -e $DIR/$tfile-1 ] && return 3
    [ -e $DIR/$tfile-2 ] && return 4
    return 0
}
run_test 28 "open(O_CREAT), |X| unlink two, close one, replay, close one (test mds_cleanup_orphans)"

test_29() {
    multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
    pid1=$!
    multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
    pid2=$!
    replay_barrier mds
    rm -f $DIR/$tfile-1
    rm -f $DIR/$tfile-2

    fail mds
    kill -USR1 $pid1
    wait $pid1 || return 1
    kill -USR1 $pid2
    wait $pid2 || return 2
    [ -e $DIR/$tfile-1 ] && return 3
    [ -e $DIR/$tfile-2 ] && return 4
    return 0
}
run_test 29 "open(O_CREAT), |X| unlink two, replay, close two (test mds_cleanup_orphans)"

test_30() {
    multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
    pid1=$!
    multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
    pid2=$!
    rm -f $DIR/$tfile-1
    rm -f $DIR/$tfile-2

    replay_barrier mds
    fail mds
    kill -USR1 $pid1
    wait $pid1 || return 1
    kill -USR1 $pid2
    wait $pid2 || return 2
    [ -e $DIR/$tfile-1 ] && return 3
    [ -e $DIR/$tfile-2 ] && return 4
    return 0
}
run_test 30 "open(O_CREAT) two, unlink two, replay, close two (test mds_cleanup_orphans)"

test_31() {
    multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
    pid1=$!
    multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
    pid2=$!
    rm -f $DIR/$tfile-1

    replay_barrier mds
    rm -f $DIR/$tfile-2
    fail mds
    kill -USR1 $pid1
    wait $pid1 || return 1
    kill -USR1 $pid2
    wait $pid2 || return 2
    [ -e $DIR/$tfile-1 ] && return 3
    [ -e $DIR/$tfile-2 ] && return 4
    return 0
}
run_test 31 "open(O_CREAT) two, unlink one, |X| unlink one, close two (test mds_cleanup_orphans)"

# tests for bug 2104; completion without crashing is success.  The close is
# stale, but we always return 0 for close, so the app never sees it.
test_32() {
    multiop_bg_pause $DIR/$tfile O_c || return 2
    pid1=$!
    multiop_bg_pause $DIR/$tfile O_c || return 3
    pid2=$!
    mds_evict_client
    df $MOUNT || sleep 1 && df $MOUNT || return 1
    kill -USR1 $pid1
    kill -USR1 $pid2
    sleep 1
    return 0
}
run_test 32 "close() notices client eviction; close() after client eviction"

# Abort recovery before client complete
test_33() {
    replay_barrier mds
    createmany -o $DIR/$tfile-%d 100 
    fail_abort mds
    # this file should be gone, because the replay was aborted
    $CHECKSTAT -t file $DIR/$tfile-* && return 3 
    unlinkmany $DIR/$tfile-%d 0 100
    return 0
}
run_test 33 "abort recovery before client does replay"

test_34() {
    multiop_bg_pause $DIR/$tfile O_c || return 2
    pid=$!
    rm -f $DIR/$tfile

    replay_barrier mds
    fail_abort mds
    kill -USR1 $pid
    [ -e $DIR/$tfile ] && return 1
    sync
    return 0
}
run_test 34 "abort recovery before client does replay (test mds_cleanup_orphans)"

# bug 2278 - generate one orphan on OST, then destroy it during recovery from llog 
test_35() {
    touch $DIR/$tfile

#define OBD_FAIL_MDS_REINT_NET_REP       0x119
    do_facet mds "sysctl -w lustre.fail_loc=0x80000119"
    rm -f $DIR/$tfile &
    sleep 1
    sync
    sleep 1
    # give a chance to remove from MDS
    fail_abort mds
    $CHECKSTAT -t file $DIR/$tfile && return 1 || true
}
run_test 35 "test recovery from llog for unlink op"

# b=2432 resent cancel after replay uses wrong cookie,
# so don't resend cancels
test_36() {
    replay_barrier mds
    touch $DIR/$tfile
    checkstat $DIR/$tfile
    facet_failover mds
    cancel_lru_locks mdc
    if dmesg | grep "unknown lock cookie"; then 
	echo "cancel after replay failed"
	return 1
    fi
}
run_test 36 "don't resend cancel"

# b=2368
# directory orphans can't be unlinked from PENDING directory
test_37() {
    rmdir $DIR/$tfile 2>/dev/null
    multiop_bg_pause $DIR/$tfile dD_c || return 2
    pid=$!
    rmdir $DIR/$tfile

    replay_barrier mds
    # clear the dmesg buffer so we only see errors from this recovery
    dmesg -c >/dev/null
    fail_abort mds
    kill -USR1 $pid
    dmesg | grep  "mds_unlink_orphan.*error .* unlinking orphan" && return 1
    sync
    return 0
}
run_test 37 "abort recovery before client does replay (test mds_cleanup_orphans for directories)"

test_38() {
    createmany -o $DIR/$tfile-%d 800
    unlinkmany $DIR/$tfile-%d 0 400
    replay_barrier mds
    fail mds
    unlinkmany $DIR/$tfile-%d 400 400
    sleep 2
    $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
}
run_test 38 "test recovery from unlink llog (test llog_gen_rec) "

test_39() { # bug 4176
    createmany -o $DIR/$tfile-%d 800
    replay_barrier mds
    unlinkmany $DIR/$tfile-%d 0 400
    fail mds
    unlinkmany $DIR/$tfile-%d 400 400
    sleep 2
    ls -1f $DIR/$tfile-*
    $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
}
run_test 39 "test recovery from unlink llog (test llog_gen_rec) "

count_ost_writes() {
    awk -vwrites=0 '/ost_write/ { writes += $2 } END { print writes; }' $LPROC/osc/*/stats
}

#b=2477,2532
test_40(){
    $LCTL mark multiop $MOUNT/$tfile OS_c 
    multiop $MOUNT/$tfile OS_c  &
    PID=$!
    writeme -s $MOUNT/${tfile}-2 &
    WRITE_PID=$!
    sleep 1
    facet_failover mds
#define OBD_FAIL_MDS_CONNECT_NET         0x117
    do_facet mds "sysctl -w lustre.fail_loc=0x80000117"
    kill -USR1 $PID
    stat1=`count_ost_writes`
    sleep $TIMEOUT
    stat2=`count_ost_writes`
    echo "$stat1, $stat2"
    if [ $stat1 -lt $stat2 ]; then 
       echo "writes continuing during recovery"
       RC=0
    else
       echo "writes not continuing during recovery, bug 2477"
       RC=4
    fi
    echo "waiting for writeme $WRITE_PID"
    kill $WRITE_PID
    wait $WRITE_PID 

    echo "waiting for multiop $PID"
    wait $PID || return 2
    do_facet client munlink $MOUNT/$tfile  || return 3
    do_facet client munlink $MOUNT/${tfile}-2  || return 3
    return $RC
}
run_test 40 "cause recovery in ptlrpc, ensure IO continues"


#b=2814
# make sure that a read to one osc doesn't try to double-unlock its page just
# because another osc is invalid.  trigger_group_io used to mistakenly return
# an error if any oscs were invalid even after having successfully put rpcs
# on valid oscs.  This was fatal if the caller was ll_readpage who unlocked
# the page, guarnateeing that the unlock from the RPC completion would
# assert on trying to unlock the unlocked page.
test_41() {
    [ $OSTCOUNT -lt 2 ] && \
	skip "skipping test 41: we don't have a second OST to test with" && \
	return

    local f=$MOUNT/$tfile
    # make sure the start of the file is ost1
    lfs setstripe $f -s $((128 * 1024)) -i 0 
    do_facet client dd if=/dev/zero of=$f bs=4k count=1 || return 3
    cancel_lru_locks osc
    # fail ost2 and read from ost1
    local osc2dev=`grep ${ost2_svc}-osc- $LPROC/devices | awk '{print $1}'`
    [ "$osc2dev" ] || return 4
    $LCTL --device $osc2dev deactivate || return 1
    do_facet client dd if=$f of=/dev/null bs=4k count=1 || return 3
    $LCTL --device $osc2dev activate || return 2
    return 0
}
run_test 41 "read from a valid osc while other oscs are invalid"

# test MDS recovery after ost failure
test_42() {
    blocks=`df -P $MOUNT | tail -n 1 | awk '{ print $2 }'`
    createmany -o $DIR/$tfile-%d 800
    replay_barrier ost1
    unlinkmany $DIR/$tfile-%d 0 400
    debugsave
    sysctl -w lnet.debug=-1
    facet_failover ost1
    
    # osc is evicted, fs is smaller (but only with failout OSTs (bug 7287)
    #blocks_after=`df -P $MOUNT | tail -n 1 | awk '{ print $2 }'`
    #[ $blocks_after -lt $blocks ] || return 1
    echo wait for MDS to timeout and recover
    sleep $((TIMEOUT * 2))
    debugrestore
    unlinkmany $DIR/$tfile-%d 400 400
    $CHECKSTAT -t file $DIR/$tfile-* && return 2 || true
}
run_test 42 "recovery after ost failure"

# timeout in MDS/OST recovery RPC will LBUG MDS
test_43() { # bug 2530
    replay_barrier mds

    # OBD_FAIL_OST_CREATE_NET 0x204
    do_facet ost1 "sysctl -w lustre.fail_loc=0x80000204"
    fail mds
    sleep 10
    do_facet ost1 "sysctl -w lustre.fail_loc=0"

    return 0
}
run_test 43 "mds osc import failure during recovery; don't LBUG"

test_44() {
    local at_max_saved=0

    mdcdev=`awk '/-mdc-/ {print $1}' $LPROC/devices`
    [ "$mdcdev" ] || exit 2

    # adaptive timeouts slow this way down
    if at_is_valid && at_is_enabled; then
        at_max_saved=$(at_max_get mds)
        at_max_set 40 mds
    fi

    for i in `seq 1 10`; do
	echo "$i of 10 ($(date +%s))"
	do_facet mds "grep service $LPROC/mdt/MDS/mds/timeouts"
	#define OBD_FAIL_TGT_CONN_RACE     0x701
	do_facet mds "sysctl -w lustre.fail_loc=0x80000701"
	$LCTL --device $mdcdev recover
	df $MOUNT
    done

    do_facet mds "sysctl -w lustre.fail_loc=0"
    [ $at_max_saved -ne 0 ] && at_max_set $at_max_saved mds
    return 0
}
run_test 44 "race in target handle connect"

test_44b() {
    mdcdev=`awk '/-mdc-/ {print $1}' $LPROC/devices`
    [ "$mdcdev" ] || exit 2
    for i in `seq 1 10`; do
	echo "$i of 10 ($(date +%s))"
	do_facet mds "grep service $LPROC/mdt/MDS/mds/timeouts"
	#define OBD_FAIL_TGT_DELAY_RECONNECT 0x704
	do_facet mds "sysctl -w lustre.fail_loc=0x80000704"
	$LCTL --device $mdcdev recover
	df $MOUNT
    done
    do_facet mds "sysctl -w lustre.fail_loc=0"
    return 0
}
run_test 44b "race in target handle connect"

# Handle failed close
test_45() {
    mdcdev=`awk '/-mdc-/ {print $1}' $LPROC/devices`
    [ "$mdcdev" ] || exit 2
    $LCTL --device $mdcdev recover

    multiop_bg_pause $DIR/$tfile O_c || return 1
    pid=$!

    # This will cause the CLOSE to fail before even 
    # allocating a reply buffer
    $LCTL --device $mdcdev deactivate || return 4

    # try the close
    kill -USR1 $pid
    wait $pid || return 1

    $LCTL --device $mdcdev activate || return 5
    sleep 1

    $CHECKSTAT -t file $DIR/$tfile || return 2
    return 0
}
run_test 45 "Handle failed close"

test_46() {
    dmesg -c >/dev/null
    drop_reply "touch $DIR/$tfile"
    fail mds
    # ironically, the previous test, 45, will cause a real forced close,
    # so just look for one for this test
    dmesg | grep -i "force closing client file handle for $tfile" && return 1
    return 0
}
run_test 46 "Don't leak file handle after open resend (3325)"

test_47() { # bug 2824
    # create some files to make sure precreate has been done on all 
    # OSTs. (just in case this test is run independently)
    createmany -o $DIR/$tfile 20  || return 1

    # OBD_FAIL_OST_CREATE_NET 0x204
    fail ost1
    do_facet ost1 "sysctl -w lustre.fail_loc=0x80000204"
    df $MOUNT || return 2

    # let the MDS discover the OST failure, attempt to recover, fail
    # and recover again.  
    sleep $((3 * TIMEOUT))

    # Without 2824, this createmany would hang 
    createmany -o $DIR/$tfile 20 || return 3
    unlinkmany $DIR/$tfile 20 || return 4

    do_facet ost1 "sysctl -w lustre.fail_loc=0"
    return 0
}
run_test 47 "MDS->OSC failure during precreate cleanup (2824)"

test_48() {
    replay_barrier mds
    createmany -o $DIR/$tfile 20  || return 1
    # OBD_FAIL_OST_EROFS 0x216
    fail mds
    do_facet ost1 "sysctl -w lustre.fail_loc=0x80000216"
    df $MOUNT || return 2

    createmany -o $DIR/$tfile 20 20 || return 2
    unlinkmany $DIR/$tfile 40 || return 3

    do_facet ost1 "sysctl -w lustre.fail_loc=0"
    return 0
}
run_test 48 "MDS->OSC failure during precreate cleanup (2824)"

test_50() {
    local oscdev=`do_facet mds grep \'${ost1_svc}-osc \' $LPROC/devices | awk '{print $1}' | head -1`
    [ "$oscdev" ] || return 1
    do_facet mds $LCTL --device $oscdev recover || return 2 
    do_facet mds $LCTL --device $oscdev recover || return 3 
    # give the mds_lov_sync threads a chance to run
    sleep 5
}
run_test 50 "Double OSC recovery, don't LASSERT (3812)"

# b3764 timed out lock replay
test_52() {
    touch $DIR/$tfile
    cancel_lru_locks mdc

    multiop $DIR/$tfile s || return 1
    replay_barrier mds
#define OBD_FAIL_LDLM_REPLY              0x30c
    do_facet mds "sysctl -w lustre.fail_loc=0x8000030c"
    fail mds || return 2
    do_facet mds "sysctl -w lustre.fail_loc=0x0"

    $CHECKSTAT -t file $DIR/$tfile-* && return 3 || true
}
run_test 52 "time out lock replay (3764)"

# bug 3462 - simultaneous MDC requests
test_53a() {
    mkdir -p $DIR/${tdir}-1
    mkdir -p $DIR/${tdir}-2
    multiop $DIR/${tdir}-1/f O_c &
    close_pid=$!
    # give multiop a chance to open
    sleep 1

    #define OBD_FAIL_MDS_CLOSE_NET 0x115
    do_facet mds "sysctl -w lustre.fail_loc=0x80000115"
    kill -USR1 $close_pid
    cancel_lru_locks MDC # force the close
    do_facet mds "sysctl -w lustre.fail_loc=0"
    mcreate $DIR/${tdir}-2/f || return 1
    
    # close should still be here
    [ -d /proc/$close_pid ] || return 2
    replay_barrier_nodf mds
    fail mds
    wait $close_pid || return 3

    $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
    $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
    rm -rf $DIR/${tdir}-*
}
run_test 53a "|X| close request while two MDC requests in flight"