X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Frecovery-small.sh;h=5b09a2e413b846d86b63ceda5d59d2b93d7fc236;hp=87b87afe485218279266d2da05fd656e7acf41ff;hb=db7c55c3441d058ca3f9c055b46fef18330674fc;hpb=3a25a5e0f4d00e9dd31eb3257861dc18a12f02ad

diff --git a/lustre/tests/recovery-small.sh b/lustre/tests/recovery-small.sh
index 87b87af..5b09a2e 100755
--- a/lustre/tests/recovery-small.sh
+++ b/lustre/tests/recovery-small.sh
@@ -218,7 +218,7 @@ test_17() {
     remote_ost_nodsh && skip "remote OST with nodsh" && return 0
 
     # With adaptive timeouts, bulk_get won't expire until adaptive_timeout_max
-    if at_is_valid && at_is_enabled; then
+    if at_is_enabled; then
         at_max_saved=$(at_max_get ost1)
         at_max_set $TIMEOUT ost1
     fi
@@ -239,7 +239,7 @@ test_17() {
     # expect cmp to succeed, client resent bulk
     do_facet client "cmp /etc/termcap $DIR/$tfile" || return 3
     do_facet client "rm $DIR/$tfile" || return 4
-    [ $at_max_saved -ne 0 ] && $(at_max_set $at_max_saved ost1)
+    [ $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)"
@@ -255,6 +255,11 @@ test_18a() {
 
     # 1 stripe on ost2
     lfs setstripe $f -s $((128 * 1024)) -i 1 -c 1
+    get_stripe_info client $f
+    if [ $stripe_index -ne 1 ]; then
+        lfs getstripe $f
+        error "$f: different stripe offset ($stripe_index)" && return
+    fi
 
     do_facet client cp $SAMPLE_FILE $f
     sync
@@ -275,14 +280,17 @@ test_18b() {
 
     do_facet client mkdir -p $DIR/$tdir
     f=$DIR/$tdir/$tfile
-    f2=$DIR/$tdir/${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
+    get_stripe_info client $f
+    if [ $stripe_index -ne 0 ]; then
+        lfs getstripe $f
+        error "$f: different stripe offset ($stripe_index)" && return
+    fi
 
     do_facet client cp $SAMPLE_FILE $f
     sync
@@ -293,7 +301,7 @@ test_18b() {
     # cache after the client reconnects?     
     rc=0
     pgcache_empty || rc=2
-    rm -f $f $f2
+    rm -f $f
     return $rc
 }
 run_test 18b "eviction and reconnect clears page cache (2766)"
@@ -303,14 +311,17 @@ test_18c() {
 
     do_facet client mkdir -p $DIR/$tdir
     f=$DIR/$tdir/$tfile
-    f2=$DIR/$tdir/${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
+    get_stripe_info client $f
+    if [ $stripe_index -ne 0 ]; then
+        lfs getstripe $f
+        error "$f: different stripe offset ($stripe_index)" && return
+    fi
 
     do_facet client cp $SAMPLE_FILE $f
     sync
@@ -326,7 +337,7 @@ test_18c() {
     # cache after the client reconnects?     
     rc=0
     pgcache_empty || rc=2
-    rm -f $f $f2
+    rm -f $f
     return $rc
 }
 run_test 18c "Dropped connect reply after eviction handing (14755)"
@@ -633,54 +644,65 @@ test_24() { # bug 11710 details correct fsync() behavior
 }
 run_test 24 "fsync error (should return error)"
 
+wait_client_evicted () {
+	local facet=$1
+	local exports=$2
+	local varsvc=${facet}_svc
+
+	wait_update $(facet_host $facet) "lctl get_param -n *.${!varsvc}.num_exports | cut -d' ' -f2" $((exports - 1)) $3
+}
+
 test_26a() {      # was test_26 bug 5921 - evict dead exports by pinger
 # this test can only run from a client on a separate node.
 	remote_ost || { skip "local OST" && return 0; }
 	remote_ost_nodsh && skip "remote OST with nodsh" && return 0
 	remote_mds || { skip "local MDS" && return 0; }
-	OST_FILE=obdfilter.${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 $OST_NEXP1 OST exports
+
+	check_timeout || return 1
+
+	local OST_NEXP=$(do_facet ost1 lctl get_param -n obdfilter.${ost1_svc}.num_exports | cut -d' ' -f2)
+
+	echo starting with $OST_NEXP OST exports
 # OBD_FAIL_PTLRPC_DROP_RPC 0x505
 	do_facet client lctl set_param fail_loc=0x505
         # 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 * 8)) secs
-	sleep $(($TIMEOUT * 8))
-        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
+
+	local rc=0
+	wait_client_evicted ost1 $OST_NEXP $((TIMEOUT * 2 + TIMEOUT * 3 / 4))
+	rc=$?
 	do_facet client lctl set_param fail_loc=0x0
-        [ $OST_NEXP1 -le $OST_NEXP2 ] && error "client not evicted"
-	return 0
+        [ $rc -eq 0 ] || error "client not evicted from OST"
 }
 run_test 26a "evict dead exports"
 
 test_26b() {      # bug 10140 - evict dead exports by pinger
 	remote_ost_nodsh && skip "remote OST with nodsh" && return 0
 
+	check_timeout || return 1
 	client_df
-        zconf_mount `hostname` $MOUNT2 || error "Failed to mount $MOUNT2"
-        sleep 1 # wait connections being established
-	MDS_FILE=mdt.${mds1_svc}.num_exports
-        MDS_NEXP1="`do_facet $SINGLEMDS 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
-        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 $SINGLEMDS 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
+	zconf_mount `hostname` $MOUNT2 || error "Failed to mount $MOUNT2"
+	sleep 1 # wait connections being established
+
+	local MDS_NEXP=$(do_facet $SINGLEMDS lctl get_param -n mdt.${mds1_svc}.num_exports | cut -d' ' -f2)
+	local OST_NEXP=$(do_facet ost1 lctl get_param -n obdfilter.${ost1_svc}.num_exports | cut -d' ' -f2)
+
+	echo starting with $OST_NEXP OST and $MDS_NEXP MDS exports
+
+	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.
+	# PING_INTERVAL max(obd_timeout / 4, 1U)
+	# sleep (2*PING_INTERVAL) 
+
+        local rc=0
+        wait_client_evicted ost1 $OST_NEXP $((TIMEOUT * 2 + TIMEOUT * 3 / 4)) || \
+		error "Client was not evicted by ost" rc=1
+	wait_client_evicted $SINGLEMDS $MDS_NEXP $((TIMEOUT * 2 + TIMEOUT * 3 / 4)) || \
+		error "Client was not evicted by mds"
 }
 run_test 26b "evict dead exports"
 
@@ -960,6 +982,92 @@ test_59() { # bug 10589
 }
 run_test 59 "Read cancel race on client eviction"
 
+err17935 () {
+    # we assume that all md changes are in the MDT0 changelog
+    if [ $MDSCOUNT -gt 1 ]; then
+	error_ignore 17935 $*
+    else
+	error $*
+    fi
+}
+
+test_60() {
+        MDT0=$($LCTL get_param -n mdc.*.mds_server_uuid | \
+	    awk '{gsub(/_UUID/,""); print $1}' | head -1)
+
+	NUM_FILES=15000
+	mkdir -p $DIR/$tdir
+
+	# Register (and start) changelog
+	USER=$(do_facet $SINGLEMDS lctl --device $MDT0 changelog_register -n)
+	echo "Registered as $MDT0 changelog user $USER"
+
+	# Generate a large number of changelog entries
+	createmany -o $DIR/$tdir/$tfile $NUM_FILES
+	sync
+	sleep 5
+
+	# Unlink files in the background
+	unlinkmany $DIR/$tdir/$tfile $NUM_FILES	&
+	CLIENT_PID=$!
+	sleep 1
+
+	# Failover the MDS while unlinks are happening
+	facet_failover $SINGLEMDS
+
+	# Wait for unlinkmany to finish
+	wait $CLIENT_PID
+
+	# Check if all the create/unlink events were recorded
+	# in the changelog
+	$LFS changelog $MDT0 >> $DIR/$tdir/changelog
+	local cl_count=$(grep UNLNK $DIR/$tdir/changelog | wc -l)
+	echo "$cl_count unlinks in $MDT0 changelog"
+
+	do_facet $SINGLEMDS lctl --device $MDT0 changelog_deregister $USER
+	USERS=$(( $(do_facet $SINGLEMDS lctl get_param -n \
+	    mdd.$MDT0.changelog_users | wc -l) - 2 ))
+	if [ $USERS -eq 0 ]; then
+	    [ $cl_count -eq $NUM_FILES ] || \
+		err17935 "Recorded ${cl_count} unlinks out of $NUM_FILES"
+	    # Also make sure we can clear large changelogs
+	    cl_count=$($LFS changelog $FSNAME | wc -l)
+	    [ $cl_count -le 2 ] || \
+		error "Changelog not empty: $cl_count entries"
+	else
+	    # If there are other users, there may be other unlinks in the log
+	    [ $cl_count -ge $NUM_FILES ] || \
+		err17935 "Recorded ${cl_count} unlinks out of $NUM_FILES"
+	    echo "$USERS other changelog users; can't verify clear"
+	fi
+}
+run_test 60 "Add Changelog entries during MDS failover"
+
+test_61()
+{
+	local cflags='osc.*-OST0000-osc-MDT*.connect_flags'
+	do_facet $SINGLEMDS "lctl get_param -n $cflags" |grep -q skip_orphan
+	[ $? -ne 0 ] && skip "don't have skip orphan feature" && return
+
+	mkdir -p $DIR/$tdir || error "mkdir dir $DIR/$tdir failed"
+	# Set the default stripe of $DIR/$tdir to put the files to ost1
+	$LFS setstripe -c 1 --index 0 $DIR/$tdir
+
+	replay_barrier $SINGLEMDS
+	createmany -o $DIR/$tdir/$tfile-%d 10 
+	local oid=`do_facet ost1 "lctl get_param -n obdfilter.*OST0000.last_id"`
+
+	fail_abort $SINGLEMDS
+	
+	touch $DIR/$tdir/$tfile
+	local id=`$LFS getstripe $DIR/$tdir/$tfile |awk '$2 ~ /^[1-9]+/ {print $2}'`
+	[ $id -le $oid ] && error "the orphan objid was reused, failed"
+
+	# Cleanup
+	rm -rf $DIR/$tdir
+}
+run_test 61 "Verify to not reuse orphan objects - bug 17025"
+
 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