replay-single.sh

lustre 1.6.5 source code

#!/bin/bash

set -e
#set -v

#
# This test needs to be run on the client
#
SAVE_PWD=$PWD
LUSTRE=${LUSTRE:-`dirname $0`/..}
SETUP=${SETUP:-}
CLEANUP=${CLEANUP:-}
. $LUSTRE/tests/test-framework.sh
init_test_env $@
. ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
CHECK_GRANT=${CHECK_GRANT:-"yes"}
GRANT_CHECK_LIST=${GRANT_CHECK_LIST:-""}

# Skip these tests
# bug number: 
ALWAYS_EXCEPT="$REPLAY_SINGLE_EXCEPT"

#                                                     63 min  7 min  AT AT AT AT"
[ "$SLOW" = "no" ] && EXCEPT_SLOW="1 2 3 4 6 6b 12 16 44      44b    65 66 67 68"

build_test_filter

cleanup_and_setup_lustre

mkdir -p $DIR

rm -rf $DIR/[df][0-9]*

test_0() {
    replay_barrier mds
    fail mds
}
run_test 0 "empty replay"

test_0b() {
    # this test attempts to trigger a race in the precreation code, 
    # and must run before any other objects are created on the filesystem
    fail ost1
    createmany -o $DIR/$tfile 20 || return 1
    unlinkmany $DIR/$tfile 20 || return 2
}
run_test 0b "ensure object created after recover exists. (3284)"

test_1() {
    replay_barrier mds
    mcreate $DIR/$tfile
    fail mds
    $CHECKSTAT -t file $DIR/$tfile || return 1
    rm $DIR/$tfile
}
run_test 1 "simple create"

test_1a() {
    do_facet ost1 "sysctl -w lustre.fail_loc=0"

    rm -fr $DIR/$tfile
    local old_last_id=`cat $LPROC/obdfilter/*/last_id`
    touch -o $DIR/$tfile 1
    sync
    local new_last_id=`cat $LPROC/obdfilter/*/last_id`
    
    test "$old_last_id" = "$new_last_id" || {
	echo "OST object create is caused by MDS"
	return 1
    }
    
    old_last_id=`cat $LPROC/obdfilter/*/last_id`
    echo "data" > $DIR/$tfile
    sync
    new_last_id=`cat $LPROC/obdfilter/*/last_id`
    test "$old_last_id" = "$new_last_id "&& {
	echo "CROW does not work on write"
	return 1
    }
    
    rm -fr $DIR/$tfile

#define OBD_FAIL_OST_CROW_EIO | OBD_FAIL_ONCE
    do_facet ost1 "sysctl -w lustre.fail_loc=0x80000801"

    rm -fr $DIR/1a1
    old_last_id=`cat $LPROC/obdfilter/*/last_id`
    echo "data" > $DIR/1a1
    sync
    new_last_id=`cat $LPROC/obdfilter/*/last_id`
    test "$old_last_id" = "$new_last_id" || {
	echo "CROW does work with fail_loc=0x80000801"
	return 1
    }
    
    rm -fr $DIR/1a1
    
    do_facet ost1 "sysctl -w lustre.fail_loc=0"
}
#CROW run_test 1a "CROW object create (check OST last_id)"

test_2a() {
    replay_barrier mds
    touch $DIR/$tfile
    fail mds
    $CHECKSTAT -t file $DIR/$tfile || return 1
    rm $DIR/$tfile
}
run_test 2a "touch"

test_2b() {
    mcreate $DIR/$tfile
    replay_barrier mds
    touch $DIR/$tfile
    fail mds
    $CHECKSTAT -t file $DIR/$tfile || return 1
    rm $DIR/$tfile
}
run_test 2b "touch"

test_3a() {
    replay_barrier mds
    mcreate $DIR/$tfile
    o_directory $DIR/$tfile
    fail mds
    $CHECKSTAT -t file $DIR/$tfile || return 2
    rm $DIR/$tfile
}
run_test 3a "replay failed open(O_DIRECTORY)"

test_3b() {
    replay_barrier mds
#define OBD_FAIL_MDS_OPEN_PACK | OBD_FAIL_ONCE
    do_facet mds "sysctl -w lustre.fail_loc=0x80000114"
    touch $DIR/$tfile
    do_facet mds "sysctl -w lustre.fail_loc=0"
    fail mds
    $CHECKSTAT -t file $DIR/$tfile && return 2
    return 0
}
run_test 3b "replay failed open -ENOMEM"

test_3c() {
    replay_barrier mds
#define OBD_FAIL_MDS_ALLOC_OBDO | OBD_FAIL_ONCE
    do_facet mds "sysctl -w lustre.fail_loc=0x80000128"
    touch $DIR/$tfile
    do_facet mds "sysctl -w lustre.fail_loc=0"
    fail mds

    $CHECKSTAT -t file $DIR/$tfile && return 2
    return 0
}
run_test 3c "replay failed open -ENOMEM"

test_4() {
    replay_barrier mds
    for i in `seq 10`; do
        echo "tag-$i" > $DIR/$tfile-$i
    done 
    fail mds
    for i in `seq 10`; do
      grep -q "tag-$i" $DIR/$tfile-$i || error "$tfile-$i"
    done 
}
run_test 4 "|x| 10 open(O_CREAT)s"

test_4b() {
    replay_barrier mds
    rm -rf $DIR/$tfile-*
    fail mds
    $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
}
run_test 4b "|x| rm 10 files"

# The idea is to get past the first block of precreated files on both 
# osts, and then replay.
test_5() {
    replay_barrier mds
    for i in `seq 220`; do
        echo "tag-$i" > $DIR/$tfile-$i
    done 
    fail mds
    for i in `seq 220`; do
      grep -q "tag-$i" $DIR/$tfile-$i || error "f1c-$i"
    done 
    rm -rf $DIR/$tfile-*
    sleep 3
    # waiting for commitment of removal
}
run_test 5 "|x| 220 open(O_CREAT)"


test_6() {
    replay_barrier mds
    mcreate $DIR/$tdir/$tfile
    fail mds
    $CHECKSTAT -t dir $DIR/$tdir || return 1
    $CHECKSTAT -t file $DIR/$tdir/$tfile || return 2
    sleep 2
    # waiting for log process thread
}
run_test 6 "mkdir + contained create"

test_6b() {
    replay_barrier mds
    rm -rf $DIR/$tdir
    fail mds
    $CHECKSTAT -t dir $DIR/$tdir && return 1 || true 
}
run_test 6b "|X| rmdir"

test_7() {
    replay_barrier mds
    mcreate $DIR/$tdir/$tfile
    fail mds
    $CHECKSTAT -t dir $DIR/$tdir || return 1
    $CHECKSTAT -t file $DIR/$tdir/$tfile || return 2
    rm -fr $DIR/$tdir
}
run_test 7 "mkdir |X| contained create"

test_8() {
    replay_barrier mds
    multiop_bg_pause $DIR/$tfile mo_c || return 4
    MULTIPID=$!
    fail mds
    ls $DIR/$tfile
    $CHECKSTAT -t file $DIR/$tfile || return 1
    kill -USR1 $MULTIPID || return 2
    wait $MULTIPID || return 3
    rm $DIR/$tfile
}
run_test 8 "creat open |X| close"

test_9() {
    replay_barrier mds
    mcreate $DIR/$tfile
    local old_inum=`ls -i $DIR/$tfile | awk '{print $1}'`
    fail mds
    local new_inum=`ls -i $DIR/$tfile | awk '{print $1}'`

    echo " old_inum == $old_inum, new_inum == $new_inum"
    if [ $old_inum -eq $new_inum  ] ;
    then
        echo " old_inum and new_inum match"
    else
        echo "!!!! old_inum and new_inum NOT match"
        return 1
    fi
    rm $DIR/$tfile
}
run_test 9  "|X| create (same inum/gen)"

test_10() {
    mcreate $DIR/$tfile
    replay_barrier mds
    mv $DIR/$tfile $DIR/$tfile-2
    rm -f $DIR/$tfile
    fail mds
    $CHECKSTAT $DIR/$tfile && return 1
    $CHECKSTAT $DIR/$tfile-2 ||return 2
    rm $DIR/$tfile-2
    return 0
}
run_test 10 "create |X| rename unlink"

test_11() {
    mcreate $DIR/$tfile
    echo "old" > $DIR/$tfile
    mv $DIR/$tfile $DIR/$tfile-2
    replay_barrier mds
    echo "new" > $DIR/$tfile
    grep new $DIR/$tfile 
    grep old $DIR/$tfile-2
    fail mds
    grep new $DIR/$tfile || return 1
    grep old $DIR/$tfile-2 || return 2
}
run_test 11 "create open write rename |X| create-old-name read"

test_12() {
    mcreate $DIR/$tfile 
    multiop_bg_pause $DIR/$tfile o_tSc || return 3
    pid=$!
    rm -f $DIR/$tfile
    replay_barrier mds
    kill -USR1 $pid
    wait $pid || return 1

    fail mds
    [ -e $DIR/$tfile ] && return 2
    return 0
}
run_test 12 "open, unlink |X| close"


# 1777 - replay open after committed chmod that would make
#        a regular open a failure    
test_13() {
    mcreate $DIR/$tfile 
    multiop_bg_pause $DIR/$tfile O_wc || return 3
    pid=$!
    chmod 0 $DIR/$tfile
    $CHECKSTAT -p 0 $DIR/$tfile
    replay_barrier mds
    fail mds
    kill -USR1 $pid
    wait $pid || return 1

    $CHECKSTAT -s 1 -p 0 $DIR/$tfile || return 2
    return 0
}
run_test 13 "open chmod 0 |x| write close"

test_14() {
    multiop_bg_pause $DIR/$tfile O_tSc || return 4
    pid=$!
    rm -f $DIR/$tfile
    replay_barrier mds
    kill -USR1 $pid || return 1
    wait $pid || return 2

    fail mds
    [ -e $DIR/$tfile ] && return 3
    return 0
}
run_test 14 "open(O_CREAT), unlink |X| close"

test_15() {
    multiop_bg_pause $DIR/$tfile O_tSc || return 5
    pid=$!
    rm -f $DIR/$tfile
    replay_barrier mds
    touch $DIR/g11 || return 1
    kill -USR1 $pid
    wait $pid || return 2

    fail mds
    [ -e $DIR/$tfile ] && return 3
    touch $DIR/h11 || return 4
    return 0
}
run_test 15 "open(O_CREAT), unlink |X|  touch new, close"


test_16() {
    replay_barrier mds
    mcreate $DIR/$tfile
    munlink $DIR/$tfile
    mcreate $DIR/$tfile-2
    fail mds
    [ -e $DIR/$tfile ] && return 1
    [ -e $DIR/$tfile-2 ] || return 2
    munlink $DIR/$tfile-2 || return 3
}
run_test 16 "|X| open(O_CREAT), unlink, touch new,  unlink new"

test_17() {
    replay_barrier mds
    multiop_bg_pause $DIR/$tfile O_c || return 4
    pid=$!
    fail mds
    kill -USR1 $pid || return 1
    wait $pid || return 2
    $CHECKSTAT -t file $DIR/$tfile || return 3
    rm $DIR/$tfile
}
run_test 17 "|X| open(O_CREAT), |replay| close"

test_18() {
    replay_barrier mds
    multiop_bg_pause $DIR/$tfile O_tSc || return 8
    pid=$!
    rm -f $DIR/$tfile
    touch $DIR/$tfile-2 || return 1
    echo "pid: $pid will close"
    kill -USR1 $pid
    wait $pid || return 2

    fail mds
    [ -e $DIR/$tfile ] && return 3
    [ -e $DIR/$tfile-2 ] || return 4
    # this touch frequently fails
    touch $DIR/$tfile-3 || return 5
    munlink $DIR/$tfile-2 || return 6
    munlink $DIR/$tfile-3 || return 7
    return 0
}
run_test 18 "|X| open(O_CREAT), unlink, touch new, close, touch, unlink"

# bug 1855 (a simpler form of test_11 above)
test_19() {
    replay_barrier mds
    mcreate $DIR/$tfile
    echo "old" > $DIR/$tfile
    mv $DIR/$tfile $DIR/$tfile-2
    grep old $DIR/$tfile-2
    fail mds
    grep old $DIR/$tfile-2 || return 2
}
run_test 19 "|X| mcreate, open, write, rename "

test_20() {
    replay_barrier mds
    multiop_bg_pause $DIR/$tfile O_tSc || return 3
    pid=$!
    rm -f $DIR/$tfile

    fail mds
    kill -USR1 $pid
    wait $pid || return 1
    [ -e $DIR/$tfile ] && return 2
    return 0
}
run_test 20 "|X| open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)"

test_20b() { # bug 10480
    BEFOREUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`

    dd if=/dev/zero of=$DIR/$tfile bs=4k count=10000 &
    pid=$!
    while [ ! -e $DIR/$tfile ] ; do
        sleep 0.060s                           # give dd a chance to start
    done

    lfs getstripe $DIR/$tfile || return 1
    rm -f $DIR/$tfile || return 2       # make it an orphan
    mds_evict_client
    df -P $DIR || df -P $DIR || true    # reconnect

    fail mds                            # start orphan recovery
    df -P $DIR || df -P $DIR || true    # reconnect
    wait_mds_recovery_done || error "MDS recovery not done"

    AFTERUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
    log "before $BEFOREUSED, after $AFTERUSED"
    [ $AFTERUSED -gt $((BEFOREUSED + 20)) ] && \
        error "after $AFTERUSED > before $BEFOREUSED" && return 5
    return 0
}
run_test 20b "write, unlink, eviction, replay, (test mds_cleanup_orphans)"

test_20c() { # bug 10480
    multiop $DIR/$tfile Ow_c &
    pid=$!
    # give multiop a chance to open
    sleep 1

    ls -la $DIR/$tfile

    mds_evict_client

    df -P $DIR || df -P $DIR || true    # reconnect

    kill -USR1 $pid
    test -s $DIR/$tfile || error "File was truncated"

    return 0
}
run_test 20c "check that client eviction does not affect file content"

test_21() {
    replay_barrier mds
    multiop_bg_pause $DIR/$tfile O_tSc || return 5
    pid=$!
    rm -f $DIR/$tfile
    touch $DIR/g11 || return 1

    fail mds
    kill -USR1 $pid
    wait $pid || return 2
    [ -e $DIR/$tfile ] && return 3
    touch $DIR/h11 || return 4
    return 0
}
run_test 21 "|X| open(O_CREAT), unlink touch new, replay, close (test mds_cleanup_orphans)"

test_22() {
    multiop_bg_pause $DIR/$tfile O_tSc || return 3
    pid=$!

    replay_barrier mds
    rm -f $DIR/$tfile

    fail mds
    kill -USR1 $pid
    wait $pid || return 1
    [ -e $DIR/$tfile ] && return 2
    return 0
}
run_test 22 "open(O_CREAT), |X| unlink, replay, close (test mds_cleanup_orphans)"

test_23() {
    multiop_bg_pause $DIR/$tfile O_tSc || return 5
    pid=$!

    replay_barrier mds
    rm -f $DIR/$tfile
    touch $DIR/g11 || return 1

    fail mds
    kill -USR1 $pid
    wait $pid || return 2
    [ -e $DIR/$tfile ] && return 3
    touch $DIR/h11 || return 4
    return 0
}
run_test 23 "open(O_CREAT), |X| unlink touch new, replay, close (test mds_cleanup_orphans)"

test_24() {
    multiop_bg_pause $DIR/$tfile O_tSc || return 3
    pid=$!

    replay_barrier mds
    fail mds
    rm -f $DIR/$tfile
    kill -USR1 $pid
    wait $pid || return 1
    [ -e $DIR/$tfile ] && return 2
    return 0
}
run_test 24 "open(O_CREAT), replay, unlink, close (test mds_cleanup_orphans)"