X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Freplay-ost-single.sh;h=3325fd1f5da1d80f64d4b1c5c4820fd2285e46eb;hp=defcf05e86ebbff3135d70610c9cd9bf420f4907;hb=a6527b28ba2082a445fbf584909a7791ce407ef3;hpb=dcd9b76ceda9a7fcce044f59b141fe1903f26543 diff --git a/lustre/tests/replay-ost-single.sh b/lustre/tests/replay-ost-single.sh index defcf05..3325fd1 100755 --- a/lustre/tests/replay-ost-single.sh +++ b/lustre/tests/replay-ost-single.sh @@ -2,17 +2,26 @@ set -e -PTLDEBUG=${PTLDEBUG:--1} -LUSTRE=${LUSTRE:-`dirname $0`/..} +LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)} SETUP=${SETUP:-""} CLEANUP=${CLEANUP:-""} . $LUSTRE/tests/test-framework.sh init_test_env $@ . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh} +init_logging -ostfailover_HOST=${ostfailover_HOST:-$ost_HOST} +# While we do not use OSTCOUNT=1 setup anymore, +# ost1failover_HOST is used +#ostfailover_HOST=${ostfailover_HOST:-$ost_HOST} #failover= must be defined in OST_MKFS_OPTIONS if ostfailover_HOST != ost_HOST +require_dsh_ost || exit 0 + +if [ "$FAILURE_MODE" = "HARD" ] && mixed_ost_devs; then + skip_env "$0: Several ost services on one ost node are used with FAILURE_MODE=$FAILURE_MODE. " + exit 0 +fi + # Tests that fail on uml CPU=`awk '/model/ {print $4}' /proc/cpuinfo` [ "$CPU" = "UML" ] && EXCEPT="$EXCEPT 6" @@ -22,124 +31,178 @@ CPU=`awk '/model/ {print $4}' /proc/cpuinfo` ALWAYS_EXCEPT="$REPLAY_OST_SINGLE_EXCEPT" # -[ "$SLOW" = "no" ] && EXCEPT_SLOW="" - -# It is replay-ost-single, after all -OSTCOUNT=1 +[ "$SLOW" = "no" ] && EXCEPT_SLOW="5" build_test_filter -REFORMAT=--reformat cleanup_and_setup_lustre +check_and_setup_lustre +assert_DIR rm -rf $DIR/[df][0-9]* +TDIR=$DIR/d0.${TESTSUITE} +mkdir -p $TDIR +$LFS setstripe $TDIR -i 0 -c 1 +$LFS getstripe $TDIR + test_0a() { zconf_umount `hostname` $MOUNT -f # needs to run during initial client->OST connection #define OBD_FAIL_OST_ALL_REPLY_NET 0x211 - do_facet ost "sysctl -w lustre.fail_loc=0x80000211" + do_facet ost1 "lctl set_param fail_loc=0x80000211" zconf_mount `hostname` $MOUNT && df $MOUNT || error "0a mount fail" } run_test 0a "target handle mismatch (bug 5317) `date +%H:%M:%S`" test_0b() { fail ost1 - cp /etc/profile $DIR/$tfile + cp /etc/profile $TDIR/$tfile sync - diff /etc/profile $DIR/$tfile - rm -f $DIR/$tfile + diff /etc/profile $TDIR/$tfile + rm -f $TDIR/$tfile } run_test 0b "empty replay" test_1() { - date > $DIR/$tfile + date > $TDIR/$tfile || error "error creating $TDIR/$tfile" fail ost1 - $CHECKSTAT -t file $DIR/$tfile || return 1 - rm -f $DIR/$tfile + $CHECKSTAT -t file $TDIR/$tfile || return 1 + rm -f $TDIR/$tfile } run_test 1 "touch" test_2() { for i in `seq 10`; do - echo "tag-$i" > $DIR/$tfile-$i + echo "tag-$i" > $TDIR/$tfile-$i || error "create $TDIR/$tfile-$i" done fail ost1 for i in `seq 10`; do - grep -q "tag-$i" $DIR/$tfile-$i || error "f2-$i" + grep -q "tag-$i" $TDIR/$tfile-$i || error "grep $TDIR/$tfile-$i" done - rm -f $DIR/$tfile-* + rm -f $TDIR/$tfile-* } run_test 2 "|x| 10 open(O_CREAT)s" test_3() { verify=$ROOT/tmp/verify-$$ - dd if=/dev/urandom bs=4096 count=1280 | tee $verify > $DIR/$tfile & + dd if=/dev/urandom bs=4096 count=1280 | tee $verify > $TDIR/$tfile & ddpid=$! sync & fail ost1 wait $ddpid || return 1 - cmp $verify $DIR/$tfile || return 2 - rm -f $verify $DIR/$tfile + cmp $verify $TDIR/$tfile || return 2 + rm -f $verify $TDIR/$tfile } run_test 3 "Fail OST during write, with verification" test_4() { verify=$ROOT/tmp/verify-$$ - dd if=/dev/urandom bs=4096 count=1280 | tee $verify > $DIR/$tfile + dd if=/dev/urandom bs=4096 count=1280 | tee $verify > $TDIR/$tfile # invalidate cache, so that we're reading over the wire cancel_lru_locks osc - cmp $verify $DIR/$tfile & + cmp $verify $TDIR/$tfile & cmppid=$! fail ost1 wait $cmppid || return 1 - rm -f $verify $DIR/$tfile + rm -f $verify $TDIR/$tfile } run_test 4 "Fail OST during read, with verification" +iozone_bg () { + local args=$@ + + local tmppipe=$TMP/${TESTSUITE}.${TESTNAME}.pipe + mkfifo $tmppipe + + echo "+ iozone $args" + iozone $args > $tmppipe & + + local pid=$! + + echo "tmppipe=$tmppipe" + echo iozone pid=$pid + + # iozone exit code is 0 even if iozone is not completed + # need to check iozone output on "complete" + local iozonelog=$TMP/${TESTSUITE}.iozone.log + rm -f $iozonelog + cat $tmppipe | while read line ; do + echo "$line" + echo "$line" >>$iozonelog + done; + + local rc=0 + wait $pid + rc=$? + if ! $(tail -1 $iozonelog | grep -q complete); then + echo iozone failed! + rc=1 + fi + rm -f $tmppipe + rm -f $iozonelog + return $rc +} + test_5() { - [ -z "`which iozone 2> /dev/null`" ] && log "iozone missing" && return - FREE=`df -P -h $DIR | tail -n 1 | awk '{ print $3 }'` - case $FREE in - *T|*G) FREE=1G;; - esac - IOZONE_OPTS="-i 0 -i 1 -i 2 -+d -r 4 -s $FREE" - iozone $IOZONE_OPTS -f $DIR/$tfile & - PID=$! - + [ -z "`which iozone 2> /dev/null`" ] && skip_env "iozone missing" && return 0 + + # striping is -c 1, get min of available + local minavail=$(lctl get_param -n osc.*[oO][sS][cC][-_]*.kbytesavail | sort -n | head -1) + local size=$(( minavail * 3/4 )) + local GB=1048576 # 1048576KB == 1GB + + if (( size > GB )); then + size=$GB + fi + local iozone_opts="-i 0 -i 1 -i 2 -+d -r 4 -s $size -f $TDIR/$tfile" + + iozone_bg $iozone_opts & + local pid=$! + + echo iozone bg pid=$pid + sleep 8 fail ost1 - wait $PID - RC=$? - log "iozone rc=$RC" - rm -f $DIR/$tfile - [ $RC -ne 0 ] && return $RC || true + local rc=0 + wait $pid + rc=$? + log "iozone rc=$rc" + rm -f $TDIR/$tfile + [ $rc -eq 0 ] || error "iozone failed" + return $rc } run_test 5 "Fail OST during iozone" kbytesfree() { - lctl get_param -n osc.*-osc-*.kbytesfree | awk '{total+=$1} END {print total}' + calc_osc_kbytes kbytesfree } test_6() { - f=$DIR/$tfile + remote_mds_nodsh && skip "remote MDS with nodsh" && return 0 + + f=$TDIR/$tfile rm -f $f sync && sleep 2 && sync # wait for delete thread before=`kbytesfree` dd if=/dev/urandom bs=4096 count=1280 of=$f || return 28 lfs getstripe $f + get_stripe_info client $f + sync - sleep 2 # ensure we have a fresh statfs + sleep 2 # ensure we have a fresh statfs sync #define OBD_FAIL_MDS_REINT_NET_REP 0x119 - do_facet mds "sysctl -w lustre.fail_loc=0x80000119" + do_facet mds "lctl set_param fail_loc=0x80000119" after_dd=`kbytesfree` log "before: $before after_dd: $after_dd" (( $before > $after_dd )) || return 1 rm -f $f - fail ost1 + fail ost$((stripe_index + 1)) + wait_recovery_complete ost$((stripe_index + 1)) || error "OST recovery not done" $CHECKSTAT -t file $f && return 2 || true sync # let the delete happen - sleep 5 + wait_mds_ost_sync || return 4 + wait_destroy_complete || return 5 after=`kbytesfree` log "before: $before after: $after" (( $before <= $after + 40 )) || return 3 # take OST logs into account @@ -147,13 +210,13 @@ test_6() { run_test 6 "Fail OST before obd_destroy" test_7() { - f=$DIR/$tfile + f=$TDIR/$tfile rm -f $f sync && sleep 5 && sync # wait for delete thread before=`kbytesfree` dd if=/dev/urandom bs=4096 count=1280 of=$f || return 4 sync - sleep 2 # ensure we have a fresh statfs + sleep 2 # ensure we have a fresh statfs sync after_dd=`kbytesfree` log "before: $before after_dd: $after_dd" @@ -161,10 +224,12 @@ test_7() { replay_barrier ost1 rm -f $f fail ost1 + wait_recovery_complete ost1 || error "OST recovery not done" $CHECKSTAT -t file $f && return 2 || true sync # let the delete happen - sleep 5 + wait_mds_ost_sync || return 4 + wait_destroy_complete || return 5 after=`kbytesfree` log "before: $before after: $after" (( $before <= $after + 40 )) || return 3 # take OST logs into account @@ -173,4 +238,4 @@ run_test 7 "Fail OST before obd_destroy" equals_msg `basename $0`: test complete, cleaning up check_and_cleanup_lustre -[ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG || true +[ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG && grep -q FAIL $TESTSUITELOG && exit 1 || true