X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Freplay-ost-single.sh;h=036750b6322450af7b053ab95468a040dfca47ba;hp=c2e493b8036539202d53df77eb8a2530b7e37fb1;hb=51206e8cd42134400fa0b6259a92d7138f3dc984;hpb=a7fb215903f6fc9ea93e85b89d1ef9abcff0cdb7

diff --git a/lustre/tests/replay-ost-single.sh b/lustre/tests/replay-ost-single.sh
index c2e493b..036750b 100755
--- a/lustre/tests/replay-ost-single.sh
+++ b/lustre/tests/replay-ost-single.sh
@@ -2,18 +2,20 @@
 
 set -e
 
-PTLDEBUG=${PTLDEBUG:--1}
 LUSTRE=${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
 
-remote_ost_nodsh && skip "remote OST with nodsh" && exit 0
+require_dsh_ost || exit 0
 
 # Tests that fail on uml
 CPU=`awk '/model/ {print $4}' /proc/cpuinfo`
@@ -26,96 +28,142 @@ ALWAYS_EXCEPT="$REPLAY_OST_SINGLE_EXCEPT"
 #					
 [ "$SLOW" = "no" ] && EXCEPT_SLOW="5"
 
-# It is replay-ost-single, after all
-OSTCOUNT=1
-
 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 "lctl set_param 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 || error "error creating $DIR/$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 || error "create $DIR/$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 "grep $DIR/$tfile-$i"
+      grep -q "tag-$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`" ] && skip "iozone missing" && return 0
-    FREE=`df -P $DIR | tail -n 1 | awk '{ print $4/2 }'`
-    GB=1048576  # 1048576KB == 1GB
-    if (( FREE > GB )); then
-        FREE=$GB
+    [ -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
-    IOZONE_OPTS="-i 0 -i 1 -i 2 -+d -r 4 -s $FREE"
-    iozone $IOZONE_OPTS -f $DIR/$tfile &
-    PID=$!
-    
+    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"
 
@@ -126,40 +174,58 @@ kbytesfree() {
 test_6() {
     remote_mds_nodsh && skip "remote MDS with nodsh" && return 0
 
-    f=$DIR/$tfile
+    f=$TDIR/$tfile
     rm -f $f
-    sync && sleep 2 && sync	# wait for delete thread
+    sync && sleep 2 && sync  # wait for delete thread
+
+    # wait till space is returned, following
+    # (( $before > $after_dd)) test counting on that
+    wait_mds_ost_sync || return 4
+    wait_destroy_complete || return 5
+
     before=`kbytesfree`
     dd if=/dev/urandom bs=4096 count=1280 of=$f || return 28
     lfs getstripe $f
+    stripe_index=$(lfs getstripe -i $f)
+
     sync
-    sleep 2					# ensure we have a fresh statfs
+    sleep 2 # ensure we have a fresh statfs
     sync
-#define OBD_FAIL_MDS_REINT_NET_REP       0x119
-    do_facet mds "lctl set_param fail_loc=0x80000119"
+
+    #define OBD_FAIL_MDS_REINT_NET_REP       0x119
+    do_facet $SINGLEMDS "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$((stripe_index + 1)) ||
+        error "OST$((stripe_index + 1)) recovery not completed"
     $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
+    (( $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 > $after_dd)) test counting on that
+    wait_mds_ost_sync || return 4
+    wait_destroy_complete || return 5
+
     before=`kbytesfree`
     dd if=/dev/urandom bs=4096 count=1280 of=$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"
@@ -167,16 +233,78 @@ 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
 }
 run_test 7 "Fail OST before obd_destroy"
 
-equals_msg `basename $0`: test complete, cleaning up
+test_8a() {
+    verify=$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) $SECONDS
 check_and_cleanup_lustre
-[ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG || true
+exit_status