X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;ds=inline;f=lustre%2Ftests%2Freplay-ost-single.sh;h=036750b6322450af7b053ab95468a040dfca47ba;hb=51206e8cd42134400fa0b6259a92d7138f3dc984;hp=bd109b938b78aad54527102dafedc773c1ac542c;hpb=ff17cc7e0282d9b1522810e0c5d12171c4d46a2d;p=fs%2Flustre-release.git

diff --git a/lustre/tests/replay-ost-single.sh b/lustre/tests/replay-ost-single.sh
index bd109b9..036750b 100755
--- a/lustre/tests/replay-ost-single.sh
+++ b/lustre/tests/replay-ost-single.sh
@@ -1,182 +1,310 @@
-#!/bin/sh
+#!/bin/bash
 
 set -e
 
-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
 
-. ${CONFIG:=$LUSTRE/tests/cfg/local.sh}
+# 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
 
-ostfailover_HOST=${ostfailover_HOST:-$ost_HOST}
+require_dsh_ost || exit 0
 
-# Skip these tests
-ALWAYS_EXCEPT="5"
-# test 5 needs a larger fs than what local normally has
-
-gen_config() {
-    rm -f $XMLCONFIG
-    add_mds mds --dev $MDSDEV --size $MDSSIZE
-    add_lov lov1 mds --stripe_sz $STRIPE_BYTES\
-	--stripe_cnt $STRIPES_PER_OBJ --stripe_pattern 0
-    add_ost ost --lov lov1 --dev $OSTDEV --size $OSTSIZE --failover
-    if [ ! -z "$ostfailover_HOST" ]; then
-	 add_ostfailover ost --dev $OSTDEV --size $OSTSIZE
-    fi
-    add_client client mds --lov lov1 --path $MOUNT
-}
+# Tests that fail on uml
+CPU=`awk '/model/ {print $4}' /proc/cpuinfo`
+[ "$CPU" = "UML" ] && EXCEPT="$EXCEPT 6"
 
-cleanup() {
-    # make sure we are using the primary MDS, so the config log will
-    # be able to clean up properly.
-    activeost=`facet_active ost`
-    if [ $activeost != "ost" ]; then
-        fail ost
-    fi
-    zconf_umount `hostname` $MOUNT
-    stop mds ${FORCE} $MDSLCONFARGS
-    stop ost ${FORCE} --dump cleanup.log
-}
+# Skip these tests
+# BUG NUMBER: 
+ALWAYS_EXCEPT="$REPLAY_OST_SINGLE_EXCEPT"
 
-if [ "$ONLY" == "cleanup" ]; then
-    sysctl -w portals.debug=0
-    cleanup
-    exit
-fi
+#					
+[ "$SLOW" = "no" ] && EXCEPT_SLOW="5"
 
 build_test_filter
 
-SETUP=${SETUP:-"setup"}
-CLEANUP=${CLEANUP:-"cleanup"}
+check_and_setup_lustre
+assert_DIR
+rm -rf $DIR/[df][0-9]*
 
-setup() {
-    gen_config
+TDIR=$DIR/d0.${TESTSUITE}
+mkdir -p $TDIR
+$LFS setstripe $TDIR -i 0 -c 1
+$LFS getstripe $TDIR
 
-    start ost --reformat $OSTLCONFARGS
-    [ "$DAEMONFILE" ] && $LCTL debug_daemon start $DAEMONFILE $DAEMONSIZE
-    start mds --reformat $MDSLCONFARGS
-    zconf_mount `hostname` $MOUNT
+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 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`"
 
-mkdir -p $DIR
-
-$SETUP
-
-test_0() {
-    fail ost
-    cp /etc/profile  $DIR/$tfile
+test_0b() {
+    fail ost1
+    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 0 "empty replay"
+run_test 0b "empty replay"
 
 test_1() {
-    date > $DIR/$tfile
-    fail ost
-    $CHECKSTAT -t file $DIR/$tfile || return 1
-    rm -f $DIR/$tfile
+    date > $TDIR/$tfile || error "error creating $TDIR/$tfile"
+    fail ost1
+    $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-$i"
     done 
-    fail ost
+    fail ost1
     for i in `seq 10`; do
-      grep -q "tag-$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=1024 count=5120 | tee $verify > $DIR/$tfile &
+    dd if=/dev/urandom bs=4096 count=1280 | tee $verify > $TDIR/$tfile &
     ddpid=$!
     sync &
-    fail ost
+    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=1024 count=5120 | 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
-    for i in /proc/fs/lustre/ldlm/namespaces/OSC_*MNT*; do
-        echo -n clear > $i/lru_size
-    done
-    cmp $verify $DIR/$tfile &
+    cancel_lru_locks osc
+    cmp $verify $TDIR/$tfile &
     cmppid=$!
-    fail ost
+    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() {
-    IOZONE_OPTS="-i 0 -i 1 -i 2 -+d -r 64 -s 1g"
-    iozone $IOZONE_OPTS -f $DIR/$tfile &
-    PID=$!
-    
-    sleep 10
-    fail ost
-    wait $PID || return 1
-    rm -f $DIR/$tfile
+    [ -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
+    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() {
-   cat /proc/fs/lustre/osc/OSC_*MNT*/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
+
+    # wait till space is returned, following
+    # (( $before > $after_dd)) test counting on that
+    wait_mds_ost_sync || return 4
+    wait_destroy_complete || return 5
+
     before=`kbytesfree`
-    dd if=/dev/urandom bs=1024 count=5120 of=$f
-#define OBD_FAIL_MDS_REINT_NET_REP       0x119
-    do_facet mds "sysctl -w lustre.fail_loc=0x80000119"
+    dd if=/dev/urandom bs=4096 count=1280 of=$f || return 28
+    lfs getstripe $f
+    stripe_index=$(lfs getstripe -i $f)
+
+    sync
+    sleep 2 # ensure we have a fresh statfs
     sync
+
+    #define OBD_FAIL_MDS_REINT_NET_REP       0x119
+    do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
     after_dd=`kbytesfree`
-    echo "before: $before after_dd: $after_dd"
-    (( before > after_dd )) || return 1
+    log "before: $before after_dd: $after_dd"
+    (( $before > $after_dd )) || return 1
     rm -f $f
-    fail ost
+    fail ost$((stripe_index + 1))
+    wait_recovery_complete ost$((stripe_index + 1)) ||
+        error "OST$((stripe_index + 1)) recovery not completed"
     $CHECKSTAT -t file $f && return 2 || true
     sync
     # let the delete happen
-    sleep 2
+    wait_mds_ost_sync || return 4
+    wait_destroy_complete || return 5
     after=`kbytesfree`
-    echo "before: $before after: $after"
-    (( before == after ))  || return 3
+    log "before: $before after: $after"
+    (( $before <= $after + 40 )) || return 3   # take OST logs into account
 }
 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
+
+    # wait till space is returned, following
+    # (( $before > $after_dd)) test counting on that
+    wait_mds_ost_sync || return 4
+    wait_destroy_complete || return 5
+
     before=`kbytesfree`
-    dd if=/dev/urandom bs=1024 count=5120 of=$f
+    dd if=/dev/urandom bs=4096 count=1280 of=$f || return 4
+    sync
+    sleep 2 # ensure we have a fresh statfs
     sync
     after_dd=`kbytesfree`
-    echo "before: $before after_dd: $after_dd"
-    (( before > after_dd )) || return 1
-    replay_barrier ost
+    log "before: $before after_dd: $after_dd"
+    (( $before > $after_dd )) || return 1
+    replay_barrier ost1
     rm -f $f
-    fail ost
+    fail ost1
+    wait_recovery_complete ost1 || error "OST recovery not done"
     $CHECKSTAT -t file $f && return 2 || true
     sync
     # let the delete happen
-    sleep 2
+    wait_mds_ost_sync || return 4
+    wait_destroy_complete || return 5
     after=`kbytesfree`
-    echo "before: $before after: $after"
-    (( before == after )) || return 3
+    log "before: $before after: $after"
+    (( $before <= $after + 40 )) || return 3	# take OST logs into account
 }
 run_test 7 "Fail OST before obd_destroy"
 
-equals_msg test complete, cleaning up
-$CLEANUP
+test_8a() {
+    verify=$ROOT/tmp/verify-$$
+    dd if=/dev/urandom of=$verify bs=4096 count=1280 ||
+        error "Create verify file failed"
+#define OBD_FAIL_OST_DQACQ_NET           0x230
+    do_facet ost1 "lctl set_param fail_loc=0x230"
+    dd if=$verify of=$TDIR/$tfile bs=4096 count=1280 oflag=sync &
+    ddpid=$!
+    sleep $TIMEOUT  # wait for the io to become redo io
+    if ! ps -p $ddpid  > /dev/null 2>&1; then
+            error "redo io finished incorrectly"
+            return 1
+    fi
+    do_facet ost1 "lctl set_param fail_loc=0"
+    wait $ddpid || return 1
+    cancel_lru_locks osc
+    cmp $verify $TDIR/$tfile || return 2
+    rm -f $verify $TDIR/$tfile
+}
+run_test 8a "Verify redo io: redo io when get -EINPROGRESS error"
+
+test_8b() {
+    verify=$ROOT/tmp/verify-$$
+    dd if=/dev/urandom of=$verify bs=4096 count=1280 ||
+        error "Create verify file failed"
+#define OBD_FAIL_OST_DQACQ_NET           0x230
+    do_facet ost1 "lctl set_param fail_loc=0x230"
+    dd if=$verify of=$TDIR/$tfile bs=4096 count=1280 oflag=sync &
+    ddpid=$!
+    sleep $TIMEOUT  # wait for the io to become redo io
+    fail ost1
+    do_facet ost1 "lctl set_param fail_loc=0"
+    wait $ddpid || return 1
+    cancel_lru_locks osc
+    cmp $verify $TDIR/$tfile || return 2
+    rm -f $verify $TDIR/$tfile
+}
+run_test 8b "Verify redo io: redo io should success after recovery"
+
+test_8c() {
+    verify=$ROOT/tmp/verify-$$
+    dd if=/dev/urandom of=$verify bs=4096 count=1280 ||
+        error "Create verify file failed"
+#define OBD_FAIL_OST_DQACQ_NET           0x230
+    do_facet ost1 "lctl set_param fail_loc=0x230"
+    dd if=$verify of=$TDIR/$tfile bs=4096 count=1280 oflag=sync &
+    ddpid=$!
+    sleep $TIMEOUT  # wait for the io to become redo io
+    ost_evict_client
+    # allow recovery to complete
+    sleep $((TIMEOUT + 2))
+    do_facet ost1 "lctl set_param fail_loc=0"
+    wait $ddpid
+    cancel_lru_locks osc
+    cmp $verify $TDIR/$tfile && return 2
+    rm -f $verify $TDIR/$tfile
+}
+run_test 8c "Verify redo io: redo io should fail after eviction"
+
+
+complete $(basename $0) $SECONDS
+check_and_cleanup_lustre
+exit_status