recovery-small.sh

lustre 1.6.5 source code

       do_facet mds "sysctl -w lustre.fail_loc=0x80000119"
       touch $DIR/$tdir-2/f &
       sleep 1
       do_facet mds "sysctl -w lustre.fail_loc=0"

       kill -USR1 $pid
       cancel_lru_locks mdc
       wait $pid || return 1

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

test_21f() {
       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=0x80000119"
       touch $DIR/$tdir-2/f &
       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 21f "drop both reply while close and open are both in flight"

test_21g() {
       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=0x80000119"
       touch $DIR/$tdir-2/f &
       sleep 1
       do_facet mds "sysctl -w lustre.fail_loc=0"

       do_facet mds "sysctl -w lustre.fail_loc=0x80000115"
       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 21g "drop open reply and close request while close and open are both in flight"

test_21h() {
       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=0x80000107"
       touch $DIR/$tdir-2/f &
       touch_pid=$!
       sleep 1
       do_facet mds "sysctl -w lustre.fail_loc=0"

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

       wait $touch_pid || return 2

       $CHECKSTAT -t file $DIR/$tdir-1/f || return 3
       $CHECKSTAT -t file $DIR/$tdir-2/f || return 4
       rm -rf $DIR/$tdir-*
}
run_test 21h "drop open request and close reply while close and open are both in flight"

# bug 3462 - multiple MDC requests
test_22() {
    f1=$DIR/${tfile}-1
    f2=$DIR/${tfile}-2
    
    do_facet mds "sysctl -w lustre.fail_loc=0x80000115"
    multiop $f2 Oc &
    close_pid=$!

    sleep 1
    multiop $f1 msu || return 1

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

    wait $close_pid || return 2
    rm -rf $f2 || return 4
}
run_test 22 "drop close request and do mknod"

test_23() { #b=4561
    multiop_bg_pause $DIR/$tfile O_c || return 1
    pid=$!
    # give a chance for open
    sleep 5

    # try the close
    drop_request "kill -USR1 $pid"

    fail mds
    wait $pid || return 1
    return 0
}
run_test 23 "client hang when close a file after mds crash"

test_24() {	# bug 2248 - eviction fails writeback but app doesn't see it
	mkdir -p $DIR/$tdir
	cancel_lru_locks osc
	multiop_bg_pause $DIR/$tdir/$tfile Owy_wyc || return 1
	MULTI_PID=$!
	ost_evict_client
	kill -USR1 $MULTI_PID
	wait $MULTI_PID
	rc=$?
	sysctl -w lustre.fail_loc=0x0
	client_reconnect
	[ $rc -eq 0 ] && error_ignore 5494 "multiop didn't fail fsync: rc $rc" || true
}
run_test 24 "fsync error (should return error)"

test_26a() {      # was test_26 bug 5921 - evict dead exports by pinger
# this test can only run from a client on a separate node.
	remote_ost || { skip "local OST" && return 0; }
	remote_mds || { skip "local MDS" && return 0; }
	OST_FILE=obdfilter.${ost1_svc}.num_exports
        OST_EXP="`do_facet ost1 lctl get_param -n $OST_FILE`"
	OST_NEXP1=`echo $OST_EXP | cut -d' ' -f2`
	echo starting with $OST_NEXP1 OST exports
# OBD_FAIL_PTLRPC_DROP_RPC 0x505
	do_facet client sysctl -w lustre.fail_loc=0x505
	# evictor takes up to 2.25x 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.
	echo Waiting for $(($TIMEOUT * 4)) secs
	sleep $(($TIMEOUT * 4))
        OST_EXP="`do_facet ost1 lctl get_param -n $OST_FILE`"
	OST_NEXP2=`echo $OST_EXP | cut -d' ' -f2`
	echo ending with $OST_NEXP2 OST exports
	do_facet client sysctl -w lustre.fail_loc=0x0
        [ $OST_NEXP1 -le $OST_NEXP2 ] && error "client not evicted"
	return 0
}
run_test 26a "evict dead exports"

test_26b() {      # bug 10140 - evict dead exports by pinger
	client_df
	zconf_mount `hostname` $MOUNT2 || error "Failed to mount $MOUNT2"
	MDS_FILE=mds.${mds_svc}.num_exports
        MDS_NEXP1="`do_facet mds lctl get_param -n $MDS_FILE | cut -d' ' -f2`"
	OST_FILE=obdfilter.${ost1_svc}.num_exports
        OST_NEXP1="`do_facet ost1 lctl get_param -n $OST_FILE | cut -d' ' -f2`"
	echo starting with $OST_NEXP1 OST and $MDS_NEXP1 MDS exports
	#force umount a client; exports should get evicted
	zconf_umount `hostname` $MOUNT2 -f
	# 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.
	echo Waiting for $(($TIMEOUT * 3)) secs
	sleep $(($TIMEOUT * 3))
        OST_NEXP2="`do_facet ost1 lctl get_param -n $OST_FILE | cut -d' ' -f2`"
        MDS_NEXP2="`do_facet mds lctl get_param -n $MDS_FILE | cut -d' ' -f2`"
	echo ending with $OST_NEXP2 OST and $MDS_NEXP2 MDS exports
        [ $OST_NEXP1 -le $OST_NEXP2 ] && error "client not evicted from OST"
        [ $MDS_NEXP1 -le $MDS_NEXP2 ] && error "client not evicted from MDS"
	return 0
}
run_test 26b "evict dead exports"

test_27() {
	remote_mds && { skip "remote MDS" && return 0; }
	mkdir -p $DIR/$tdir
	writemany -q -a $DIR/$tdir/$tfile 0 5 &
	CLIENT_PID=$!
	sleep 1
	FAILURE_MODE="SOFT"
	facet_failover mds
#define OBD_FAIL_OSC_SHUTDOWN            0x407
	sysctl -w lustre.fail_loc=0x80000407
	# need to wait for reconnect
	echo -n waiting for fail_loc
	while [ `sysctl -n lustre.fail_loc` -eq -2147482617 ]; do
	    sleep 1
	    echo -n .
	done
	facet_failover mds
	#no crashes allowed!
        kill -USR1 $CLIENT_PID
	wait $CLIENT_PID 
	true
}
run_test 27 "fail LOV while using OSC's"

test_28() {      # bug 6086 - error adding new clients
	do_facet client mcreate $MOUNT/$tfile       || return 1
	drop_bl_callback "chmod 0777 $MOUNT/$tfile" ||echo "evicted as expected"
	#define OBD_FAIL_MDS_ADD_CLIENT 0x12f
	do_facet mds sysctl -w lustre.fail_loc=0x8000012f
	# fail once (evicted), reconnect fail (fail_loc), ok
	df || (sleep 1; df) || (sleep 1; df) || error "reconnect failed"
	rm -f $MOUNT/$tfile
	fail mds		# verify MDS last_rcvd can be loaded
}
run_test 28 "handle error adding new clients (bug 6086)"

test_50() {
	mkdir -p $DIR/$tdir
	debugsave
	sysctl -w lnet.debug="-dlmtrace -ha"
	# put a load of file creates/writes/deletes
	writemany -q $DIR/$tdir/$tfile 0 5 &
	CLIENT_PID=$!
	echo writemany pid $CLIENT_PID
	sleep 10
	FAILURE_MODE="SOFT"
	$LCTL mark "$TESTNAME fail mds 1"
	fail mds
	# wait for client to reconnect to MDS
	sleep 60
	$LCTL mark "$TESTNAME fail mds 2"
	fail mds
	sleep 60
	$LCTL mark "$TESTNAME fail mds 3"
	fail mds
	# client process should see no problems even though MDS went down
	sleep $TIMEOUT
        kill -USR1 $CLIENT_PID
	wait $CLIENT_PID 
	rc=$?
	echo writemany returned $rc
	#these may fail because of eviction due to slow AST response.
	debugrestore
	[ $rc -eq 0 ] || error_ignore 13652 "writemany returned rc $rc" || true
}
run_test 50 "failover MDS under load"

test_51() {
	mkdir -p $DIR/$tdir
	# put a load of file creates/writes/deletes
	writemany -q $DIR/$tdir/$tfile 0 5 &
	CLIENT_PID=$!
	sleep 1
	FAILURE_MODE="SOFT"
	facet_failover mds
	# failover at various points during recovery
	SEQ="1 5 10 $(seq $TIMEOUT 5 $(($TIMEOUT+10)))"
        echo will failover at $SEQ
        for i in $SEQ
          do
          echo failover in $i sec
          sleep $i
	  $LCTL mark "$TESTNAME fail mds $i"
          facet_failover mds
        done
	# client process should see no problems even though MDS went down
	# and recovery was interrupted
	sleep $TIMEOUT
        kill -USR1 $CLIENT_PID
	wait $CLIENT_PID 
	rc=$?
	echo writemany returned $rc
	[ $rc -eq 0 ] || error_ignore 13652 "writemany returned rc $rc" || true
}
run_test 51 "failover MDS during recovery"

test_52_guts() {
	do_facet client "writemany -q -a $DIR/$tdir/$tfile 300 5" &
	CLIENT_PID=$!
	echo writemany pid $CLIENT_PID
	sleep 10
	FAILURE_MODE="SOFT"
	$LCTL mark "$TESTNAME fail ost $1"
	fail ost1
	rc=0
	wait $CLIENT_PID || rc=$?
	# active client process should see an EIO for down OST
	[ $rc -eq 5 ] && { echo "writemany correctly failed $rc" && return 0; }
	# but timing or failover setup may allow success
	[ $rc -eq 0 ] && { echo "writemany succeeded" && return 0; }
	echo "writemany returned $rc"
	return $rc
}

test_52() {
	mkdir -p $DIR/$tdir
	test_52_guts 1
	rc=$?
	[ $rc -ne 0 ] && { return $rc; }
	# wait for client to reconnect to OST
	sleep 30
	test_52_guts 2
	rc=$?
	[ $rc -ne 0 ] && { return $rc; }
	sleep 30
	test_52_guts 3
	rc=$?
	client_reconnect
	#return $rc
}
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" ||\
		return 2
}
run_test 53 "touch: drop rep"

test_54() {
	zconf_mount `hostname` $MOUNT2
        touch $DIR/$tfile
        touch $DIR2/$tfile.1
        sleep 10
        cat $DIR2/$tfile.missing # save transno = 0, rc != 0 into last_rcvd
        fail mds
        umount $MOUNT2
        ERROR=`dmesg | egrep "(test 54|went back in time)" | tail -n1 | grep "went back in time"`
        [ x"$ERROR" == x ] || error "back in time occured"
}
run_test 54 "back in time"

# bug 11330 - liblustre application death during I/O locks up OST
test_55() {
	remote_ost && { skip "remote OST" && return 0; }

	mkdir -p $DIR/$tdir

	# first dd should be finished quickly
	dd if=/dev/zero of=$DIR/$tdir/$tfile-1 bs=32M count=4  &
	DDPID=$!
	count=0
	echo  "step1: testing ......"
	while [ true ]; do
	    if [ -z `ps x | awk '$1 == '$DDPID' { print $5 }'` ]; then break; fi
	    count=$[count+1]
	    if [ $count -gt 64 ]; then
		error "dd should be finished!"
	    fi
	    sleep 1
	done	
	echo "(dd_pid=$DDPID, time=$count)successful"

        #define OBD_FAIL_OST_DROP_REQ            0x21d
	do_facet ost sysctl -w lustre.fail_loc=0x0000021d
	# second dd will be never finished
	dd if=/dev/zero of=$DIR/$tdir/$tfile-2 bs=32M count=4  &	
	DDPID=$!
	count=0
	echo  "step2: testing ......"
	while [ $count -le 64 ]; do
	    dd_name="`ps x | awk '$1 == '$DDPID' { print $5 }'`"	    
	    if [ -z  $dd_name ]; then 
                ls -l $DIR/$tdir
		echo  "debug: (dd_name=$dd_name, dd_pid=$DDPID, time=$count)"
		error "dd shouldn't be finished!"
	    fi
	    count=$[count+1]
	    sleep 1
	done	
	echo "(dd_pid=$DDPID, time=$count)successful"

	#Recover fail_loc and dd will finish soon
	do_facet ost sysctl -w lustre.fail_loc=0
	count=0
	echo  "step3: testing ......"
	while [ true ]; do
	    if [ -z `ps x | awk '$1 == '$DDPID' { print $5 }'` ]; then break; fi
	    count=$[count+1]
	    if [ $count -gt 500 ]; then
		error "dd should be finished!"
	    fi
	    sleep 1
	done	
	echo "(dd_pid=$DDPID, time=$count)successful"

        rm -rf $DIR/$tdir
}
run_test 55 "ost_brw_read/write drops timed-out read/write request"

test_56() { # b=11277
#define OBD_FAIL_MDS_RESEND      0x136
        touch $DIR/$tfile
        do_facet mds sysctl -w lustre.fail_loc=0x80000136
        stat $DIR/$tfile
        do_facet mds sysctl -w lustre.fail_loc=0
        rm -f $DIR/$tfile
}
run_test 56 "do not allow reconnect to busy exports"

test_57_helper() {
        # no oscs means no client or mdt 
        while lctl get_param osc.*.* > /dev/null 2>&1; do
                : # loop until proc file is removed
        done
}

test_57() { # bug 10866
        test_57_helper &
        pid=$!
        sleep 1
#define OBD_FAIL_LPROC_REMOVE            0xB00
        sysctl -w lustre.fail_loc=0x80000B00
        zconf_umount `hostname` $DIR
        sysctl -w lustre.fail_loc=0x80000B00
        fail_abort mds
        kill -9 $pid
        sysctl -w lustre.fail_loc=0
        mount_client $DIR
        do_facet client "df $DIR"
}
run_test 57 "read procfs entries causes kernel crash"

test_58() { # bug 11546
#define OBD_FAIL_MDC_ENQUEUE_PAUSE        0x801
        touch $MOUNT/$tfile
        ls -la $MOUNT/$tfile
        sysctl -w lustre.fail_loc=0x80000801
        cp $MOUNT/$tfile /dev/null &
        pid=$!
        sleep 1
        sysctl -w lustre.fail_loc=0
        drop_bl_callback rm -f $MOUNT/$tfile
        wait $pid
        do_facet client "df $DIR"
}
run_test 58 "Eviction in the middle of open RPC reply processing"

test_59() { # bug 10589
	zconf_mount `hostname` $MOUNT2 || error "Failed to mount $MOUNT2"
	echo $DIR2 | grep -q $MOUNT2 || error "DIR2 is not set properly: $DIR2"
	sysctl -w lustre.fail_loc=0x311
	writes=$(LANG=C dd if=/dev/zero of=$DIR2/$tfile count=1 2>&1)
	[ $? = 0 ] || error "dd write failed"
	writes=$(echo $writes | awk  -F '+' '/out/ {print $1}')
	sysctl -w lustre.fail_loc=0
	sync
	zconf_umount `hostname` $MOUNT2 -f
	reads=$(LANG=C dd if=$DIR/$tfile of=/dev/null 2>&1)
	[ $? = 0 ] || error "dd read failed"
	reads=$(echo $reads | awk -F '+' '/in/ {print $1}')
	[ "$reads" -eq "$writes" ] || error "read" $reads "blocks, must be" $writes
}
run_test 59 "Read cancel race on client eviction"

equals_msg `basename $0`: test complete, cleaning up
check_and_cleanup_lustre
[ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG || true