recovery-small.sh

lustre 1.6.5 source code

#!/bin/bash

set -e

#         bug  5493
ALWAYS_EXCEPT="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}

# also long tests: 19, 21a, 21e, 21f, 23, 27
#                                   1  2.5  2.5    4    4          (min)"
[ "$SLOW" = "no" ] && EXCEPT_SLOW="17  26a  26b    50   51     57"

build_test_filter

# Allow us to override the setup if we already have a mounted system by
# setting SETUP=" " and CLEANUP=" "
SETUP=${SETUP:-""}
CLEANUP=${CLEANUP:-""}

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

test_1() {
    drop_request "mcreate $MOUNT/1"  || return 1
    drop_reint_reply "mcreate $MOUNT/2"    || return 2
}
run_test 1 "mcreate: drop req, drop rep"

test_2() {
    drop_request "tchmod 111 $MOUNT/2"  || return 1
    drop_reint_reply "tchmod 666 $MOUNT/2"    || return 2
}
run_test 2 "chmod: drop req, drop rep"

test_3() {
    drop_request "statone $MOUNT/2" || return 1
    drop_reply "statone $MOUNT/2"   || return 2
}
run_test 3 "stat: drop req, drop rep"

SAMPLE_NAME=recovery-small.junk
SAMPLE_FILE=$TMP/$SAMPLE_NAME
# make this big, else test 9 doesn't wait for bulk -- bz 5595
dd if=/dev/urandom of=$SAMPLE_FILE bs=1M count=4

test_4() {
    do_facet client "cp $SAMPLE_FILE $MOUNT/$SAMPLE_NAME" || return 1
    drop_request "cat $MOUNT/$SAMPLE_NAME > /dev/null"   || return 2
    drop_reply "cat $MOUNT/$SAMPLE_NAME > /dev/null"     || return 3
}
run_test 4 "open: drop req, drop rep"

test_5() {
    drop_request "mv $MOUNT/$SAMPLE_NAME $MOUNT/renamed" || return 1
    drop_reint_reply "mv $MOUNT/renamed $MOUNT/renamed-again" || return 2
    do_facet client "checkstat -v $MOUNT/renamed-again"  || return 3
}
run_test 5 "rename: drop req, drop rep"

[ ! -e $MOUNT/renamed-again ] && cp $SAMPLE_FILE $MOUNT/renamed-again
test_6() {
    drop_request "mlink $MOUNT/renamed-again $MOUNT/link1" || return 1
    drop_reint_reply "mlink $MOUNT/renamed-again $MOUNT/link2"   || return 2
}
run_test 6 "link: drop req, drop rep"

[ ! -e $MOUNT/link1 ] && mlink $MOUNT/renamed-again $MOUNT/link1
[ ! -e $MOUNT/link2 ] && mlink $MOUNT/renamed-again $MOUNT/link2
test_7() {
    drop_request "munlink $MOUNT/link1"   || return 1
    drop_reint_reply "munlink $MOUNT/link2"     || return 2
}
run_test 7 "unlink: drop req, drop rep"

#bug 1423
test_8() {
    drop_reint_reply "touch $MOUNT/$tfile"    || return 1
}
run_test 8 "touch: drop rep (bug 1423)"

SAMPLE_FILE=$TMP/recovery-small.junk
dd if=/dev/urandom of=$SAMPLE_FILE bs=1M count=4

#bug 1420
test_9() {
    pause_bulk "cp /etc/profile $MOUNT/$tfile"       || return 1
    do_facet client "cp $SAMPLE_FILE $MOUNT/${tfile}.2"  || return 2
    do_facet client "sync"
    do_facet client "rm $MOUNT/$tfile $MOUNT/${tfile}.2" || return 3
}
run_test 9 "pause bulk on OST (bug 1420)"

#bug 1521
test_10() {
    do_facet client mcreate $MOUNT/$tfile        || return 1
    drop_bl_callback "chmod 0777 $MOUNT/$tfile"  || echo "evicted as expected"
    # wait for the mds to evict the client
    #echo "sleep $(($TIMEOUT*2))"
    #sleep $(($TIMEOUT*2))
    do_facet client touch $MOUNT/$tfile || echo "touch failed, evicted"
    do_facet client checkstat -v -p 0777 $MOUNT/$tfile  || return 3
    do_facet client "munlink $MOUNT/$tfile"
}
run_test 10 "finish request on server after client eviction (bug 1521)"

#bug 2460
# wake up a thread waiting for completion after eviction
test_11(){
    do_facet client multiop $MOUNT/$tfile Ow  || return 1
    do_facet client multiop $MOUNT/$tfile or  || return 2

    cancel_lru_locks osc

    do_facet client multiop $MOUNT/$tfile or  || return 3
    drop_bl_callback multiop $MOUNT/$tfile Ow || echo "evicted as expected"

    do_facet client munlink $MOUNT/$tfile  || return 4
}
run_test 11 "wake up a thread waiting for completion after eviction (b=2460)"

#b=2494
test_12(){
    $LCTL mark multiop $MOUNT/$tfile OS_c 
    do_facet mds "sysctl -w lustre.fail_loc=0x115"
    clear_failloc mds $((TIMEOUT * 2)) &
    multiop_bg_pause $MOUNT/$tfile OS_c || return 1
    PID=$!
#define OBD_FAIL_MDS_CLOSE_NET           0x115
    kill -USR1 $PID
    echo "waiting for multiop $PID"
    wait $PID || return 2
    do_facet client munlink $MOUNT/$tfile  || return 3
}
run_test 12 "recover from timed out resend in ptlrpcd (b=2494)"

# Bug 113, check that readdir lost recv timeout works.
test_13() {
    mkdir $MOUNT/readdir || return 1
    touch $MOUNT/readdir/newentry || return
# OBD_FAIL_MDS_READPAGE_NET|OBD_FAIL_ONCE
    do_facet mds "sysctl -w lustre.fail_loc=0x80000104"
    ls $MOUNT/readdir || return 3
    do_facet mds "sysctl -w lustre.fail_loc=0"
    rm -rf $MOUNT/readdir || return 4
}
run_test 13 "mdc_readpage restart test (bug 1138)"

# Bug 113, check that readdir lost send timeout works.
test_14() {
    mkdir $MOUNT/readdir
    touch $MOUNT/readdir/newentry
# OBD_FAIL_MDS_SENDPAGE|OBD_FAIL_ONCE
    do_facet mds "sysctl -w lustre.fail_loc=0x80000106"
    ls $MOUNT/readdir || return 1
    do_facet mds "sysctl -w lustre.fail_loc=0"
}
run_test 14 "mdc_readpage resend test (bug 1138)"

test_15() {
    do_facet mds "sysctl -w lustre.fail_loc=0x80000128"
    touch $DIR/$tfile && return 1
    return 0
}
run_test 15 "failed open (-ENOMEM)"

READ_AHEAD=`lctl get_param -n llite.*.max_read_ahead_mb | head -n 1`
stop_read_ahead() {
   lctl set_param -n llite.*.max_read_ahead_mb 0
}

start_read_ahead() {
   lctl set_param -n llite.*.max_read_ahead_mb $READ_AHEAD
}

test_16() {
    do_facet client cp $SAMPLE_FILE $MOUNT
    sync
    stop_read_ahead

#define OBD_FAIL_PTLRPC_BULK_PUT_NET 0x504 | OBD_FAIL_ONCE
    do_facet ost1 sysctl -w lustre.fail_loc=0x80000504
    cancel_lru_locks osc
    # OST bulk will time out here, client resends
    do_facet client "cmp $SAMPLE_FILE $MOUNT/${SAMPLE_FILE##*/}" || return 1
    do_facet ost1 sysctl -w lustre.fail_loc=0
    # give recovery a chance to finish (shouldn't take long)
    sleep $TIMEOUT
    do_facet client "cmp $SAMPLE_FILE $MOUNT/${SAMPLE_FILE##*/}" || return 2
    start_read_ahead
}
run_test 16 "timeout bulk put, don't evict client (2732)"

test_17() {
    local at_max_saved=0

    # With adaptive timeouts, bulk_get won't expire until adaptive_timeout_max
    if at_is_valid && at_is_enabled; then
        at_max_saved=$(at_max_get ost1)
        at_max_set $TIMEOUT ost1
    fi

    # OBD_FAIL_PTLRPC_BULK_GET_NET 0x0503 | OBD_FAIL_ONCE
    # OST bulk will time out here, client retries
    do_facet ost1 sysctl -w lustre.fail_loc=0x80000503
    # need to ensure we send an RPC
    do_facet client cp $SAMPLE_FILE $DIR/$tfile
    sync

    # with AT, client will wait adaptive_max*factor+net_latency before
    # expiring the req, hopefully timeout*2 is enough
    sleep $(($TIMEOUT*2))

    do_facet ost1 sysctl -w lustre.fail_loc=0
    do_facet client "df $DIR"
    # expect cmp to succeed, client resent bulk
    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)
    return 0
}
run_test 17 "timeout bulk get, don't evict client (2732)"

test_18a() {
    [ -z ${ost2_svc} ] && skip "needs 2 osts" && return 0

    do_facet client mkdir -p $MOUNT/$tdir
    f=$MOUNT/$tdir/$tfile

    cancel_lru_locks osc
    pgcache_empty || return 1

    # 1 stripe on ost2
    lfs setstripe $f -s $((128 * 1024)) -i 1 -c 1

    do_facet client cp $SAMPLE_FILE $f
    sync
    local osc2dev=`lctl get_param -n devices | grep ${ost2_svc}-osc- | awk '{print $1}'`
    $LCTL --device $osc2dev deactivate || return 3
    # my understanding is that there should be nothing in the page
    # cache after the client reconnects?     
    rc=0
    pgcache_empty || rc=2
    $LCTL --device $osc2dev activate
    rm -f $f
    return $rc
}
run_test 18a "manual ost invalidate clears page cache immediately"

test_18b() {
    do_facet client mkdir -p $MOUNT/$tdir
    f=$MOUNT/$tdir/$tfile
    f2=$MOUNT/$tdir/${tfile}-2

    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
    lfs setstripe $f2 -s $((128 * 1024)) -i 0 -c 1

    do_facet client cp $SAMPLE_FILE $f
    sync
    ost_evict_client

    # force reconnect
    df $MOUNT > /dev/null 2>&1
    sleep 2
    # my understanding is that there should be nothing in the page
    # cache after the client reconnects?     
    rc=0
    pgcache_empty || rc=2
    rm -f $f $f2
    return $rc
}
run_test 18b "eviction and reconnect clears page cache (2766)"

test_18c() {
    do_facet client mkdir -p $MOUNT/$tdir
    f=$MOUNT/$tdir/$tfile
    f2=$MOUNT/$tdir/${tfile}-2

    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
    lfs setstripe $f2 -s $((128 * 1024)) -i 0 -c 1

    do_facet client cp $SAMPLE_FILE $f
    sync
    ost_evict_client

    # OBD_FAIL_OST_CONNECT_NET2
    # lost reply to connect request
    do_facet ost1 sysctl -w lustre.fail_loc=0x80000225
    # force reconnect
    df $MOUNT > /dev/null 2>&1
    sleep 2
    # my understanding is that there should be nothing in the page
    # cache after the client reconnects?     
    rc=0
    pgcache_empty || rc=2
    rm -f $f $f2
    return $rc
}
run_test 18c "Dropped connect reply after eviction handing (14755)"

test_19a() {
    f=$MOUNT/$tfile
    do_facet client mcreate $f        || return 1
    drop_ldlm_cancel "chmod 0777 $f"  || echo "evicted as expected"

    do_facet client checkstat -v -p 0777 $f  || echo evicted
    # let the client reconnect
    sleep 5
    do_facet client "munlink $f"
}
run_test 19a "test expired_lock_main on mds (2867)"

test_19b() {
    f=$MOUNT/$tfile
    do_facet client multiop $f Ow  || return 1
    do_facet client multiop $f or  || return 2

    cancel_lru_locks osc

    do_facet client multiop $f or  || return 3
    drop_ldlm_cancel multiop $f Ow  || echo "client evicted, as expected"

    do_facet client munlink $f  || return 4
}
run_test 19b "test expired_lock_main on ost (2867)"

test_20a() {	# bug 2983 - ldlm_handle_enqueue cleanup
	mkdir -p $DIR/$tdir
	lfs setstripe $DIR/$tdir/${tfile} -i 0 -c 1
	multiop_bg_pause $DIR/$tdir/${tfile} O_wc || return 1
	MULTI_PID=$!
	cancel_lru_locks osc
#define OBD_FAIL_LDLM_ENQUEUE_EXTENT_ERR 0x308
	do_facet ost1 sysctl -w lustre.fail_loc=0x80000308
	kill -USR1 $MULTI_PID
	wait $MULTI_PID
	rc=$?
	[ $rc -eq 0 ] && error "multiop didn't fail enqueue: rc $rc" || true
}
run_test 20a "ldlm_handle_enqueue error (should return error)" 

test_20b() {	# bug 2986 - ldlm_handle_enqueue error during open
	mkdir -p $DIR/$tdir
	lfs setstripe $DIR/$tdir/${tfile} -i 0 -c 1
	cancel_lru_locks osc
#define OBD_FAIL_LDLM_ENQUEUE_EXTENT_ERR 0x308
	do_facet ost1 sysctl -w lustre.fail_loc=0x80000308
	dd if=/etc/hosts of=$DIR/$tdir/$tfile && \
		error "didn't fail open enqueue" || true
}
run_test 20b "ldlm_handle_enqueue error (should return error)"

test_21a() {
       mkdir -p $DIR/$tdir-1
       mkdir -p $DIR/$tdir-2
       multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
       close_pid=$!

       do_facet mds "sysctl -w lustre.fail_loc=0x80000129"
       multiop $DIR/$tdir-2/f Oc &
       open_pid=$!
       sleep 1
       do_facet mds "sysctl -w lustre.fail_loc=0"

       do_facet mds "sysctl -w lustre.fail_loc=0x80000115"
       kill -USR1 $close_pid
       cancel_lru_locks mdc
       wait $close_pid || return 1
       wait $open_pid || return 2
       do_facet mds "sysctl -w lustre.fail_loc=0"

       $CHECKSTAT -t file $DIR/$tdir-1/f || return 3
       $CHECKSTAT -t file $DIR/$tdir-2/f || return 4

       rm -rf $DIR/$tdir-*
}
run_test 21a "drop close request while close and open are both in flight"

test_21b() {
       mkdir -p $DIR/$tdir-1
       mkdir -p $DIR/$tdir-2
       multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
       close_pid=$!

       do_facet mds "sysctl -w lustre.fail_loc=0x80000107"
       mcreate $DIR/$tdir-2/f &
       open_pid=$!
       sleep 1
       do_facet mds "sysctl -w lustre.fail_loc=0"

       kill -USR1 $close_pid
       cancel_lru_locks mdc
       wait $close_pid || return 1
       wait $open_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 21b "drop open request while close and open are both in flight"

test_21c() {
       mkdir -p $DIR/$tdir-1
       mkdir -p $DIR/$tdir-2
       multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
       close_pid=$!

       do_facet mds "sysctl -w lustre.fail_loc=0x80000107"
       mcreate $DIR/$tdir-2/f &
       open_pid=$!
       sleep 3
       do_facet mds "sysctl -w lustre.fail_loc=0"

       do_facet mds "sysctl -w lustre.fail_loc=0x80000115"
       kill -USR1 $close_pid
       cancel_lru_locks mdc
       wait $close_pid || return 1
       wait $open_pid || return 2

       do_facet mds "sysctl -w lustre.fail_loc=0"

       $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
       $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
       rm -rf $DIR/$tdir-*
}
run_test 21c "drop both request while close and open are both in flight"

test_21d() {
       mkdir -p $DIR/$tdir-1
       mkdir -p $DIR/$tdir-2
       multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
       pid=$!

       do_facet mds "sysctl -w lustre.fail_loc=0x80000129"
       multiop $DIR/$tdir-2/f Oc &
       sleep 1
       do_facet mds "sysctl -w lustre.fail_loc=0"

       do_facet mds "sysctl -w lustre.fail_loc=0x80000122"
       kill -USR1 $pid
       cancel_lru_locks mdc
       wait $pid || return 1
       do_facet mds "sysctl -w lustre.fail_loc=0"

       $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
       $CHECKSTAT -t file $DIR/$tdir-2/f || return 3

       rm -rf $DIR/$tdir-*
}
run_test 21d "drop close reply while close and open are both in flight"

test_21e() {
       mkdir -p $DIR/$tdir-1
       mkdir -p $DIR/$tdir-2
       multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
       pid=$!