LU-1538 tests: fix test cases when OST is full

[fs/lustre-release.git] / lustre / tests / sanity.sh

diff --git a/lustre/tests/sanity.sh b/lustre/tests/sanity.sh
index abc24b3..7fddfe1 100644 (file)
--- a/lustre/tests/sanity.sh
+++ b/lustre/tests/sanity.sh
@@ -12,7 +12,7 @@ ALWAYS_EXCEPT="                27u   42a  42b  42c  42d  45   51d   68b   $SANIT
 # UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
 
 # with LOD/OSP landing
-# bug number for skipped tests: LU2036
+# bug number for skipped tests: LU-2036
 ALWAYS_EXCEPT="                 76     $ALWAYS_EXCEPT"
 
 
@@ -4012,7 +4012,7 @@ run_test 56w "check lfs_migrate -c stripe_count works"
 
 test_57a() {
        # note test will not do anything if MDS is not local
-       if [ "$(facet_type_fstype MDS)" != ldiskfs ]; then
+       if [ "$(facet_fstype $SINGLEMDS)" != ldiskfs ]; then
                skip "Only applicable to ldiskfs-based MDTs"
                return
        fi
@@ -4032,7 +4032,7 @@ test_57a() {
 run_test 57a "verify MDS filesystem created with large inodes =="
 
 test_57b() {
-       if [ "$(facet_type_fstype MDS)" != ldiskfs ]; then
+       if [ "$(facet_fstype $SINGLEMDS)" != ldiskfs ]; then
                skip "Only applicable to ldiskfs-based MDTs"
                return
        fi
@@ -4942,6 +4942,16 @@ test_79() { # bug 12743
 run_test 79 "df report consistency check ======================="
 
 test_80() { # bug 10718
+        # relax strong synchronous semantics for slow backends like ZFS
+        local soc="obdfilter.*.sync_on_lock_cancel"
+        local soc_old=$(do_facet ost1 lctl get_param -n $soc | head -n1)
+        local hosts=
+        if [ "$soc_old" != "never" -a "$FSTYPE" != "ldiskfs" ]; then
+                hosts=$(for host in $(seq -f "ost%g" 1 $OSTCOUNT); do
+                          facet_active_host $host; done | sort -u)
+                do_nodes $hosts lctl set_param $soc=never
+        fi
+
         dd if=/dev/zero of=$DIR/$tfile bs=1M count=1 seek=1M
         sync; sleep 1; sync
         local BEFORE=`date +%s`
@@ -4951,6 +4961,9 @@ test_80() { # bug 10718
         if [ $DIFF -gt 1 ] ; then
                 error "elapsed for 1M@1T = $DIFF"
         fi
+
+        [ -n "$hosts" ] && do_nodes $hosts lctl set_param $soc=$soc_old
+
         true
         rm -f $DIR/$tfile
 }
@@ -4960,7 +4973,7 @@ test_81a() { # LU-456
         remote_ost_nodsh && skip "remote OST with nodsh" && return
         # define OBD_FAIL_OST_MAPBLK_ENOSPC    0x228
         # MUST OR with the OBD_FAIL_ONCE (0x80000000)
-        do_facet ost0 lctl set_param fail_loc=0x80000228
+        do_facet ost1 lctl set_param fail_loc=0x80000228
 
         # write should trigger a retry and success
         $SETSTRIPE -i 0 -c 1 $DIR/$tfile
@@ -4976,7 +4989,7 @@ test_81b() { # LU-456
         remote_ost_nodsh && skip "remote OST with nodsh" && return
         # define OBD_FAIL_OST_MAPBLK_ENOSPC    0x228
         # Don't OR with the OBD_FAIL_ONCE (0x80000000)
-        do_facet ost0 lctl set_param fail_loc=0x228
+        do_facet ost1 lctl set_param fail_loc=0x228
 
         # write should retry several times and return -ENOSPC finally
         $SETSTRIPE -i 0 -c 1 $DIR/$tfile
@@ -5288,7 +5301,7 @@ test_101d() {
         { skip "Need free space ${size}M, have $space" && return; }
 
     echo Creating ${size}M test file $file
-    dd if=/dev/zero of=$file bs=1M count=$size
+    dd if=/dev/zero of=$file bs=1M count=$size || error "dd failed"
     echo Cancel LRU locks on lustre client to flush the client cache
     cancel_lru_locks osc
 
@@ -5950,7 +5963,7 @@ free_min_max () {
        echo Max free space: OST $MAXI: $MAXV
 }
 
-test_116() {
+test_116a() { # was previously test_116()
        [ "$OSTCOUNT" -lt "2" ] && skip_env "$OSTCOUNT < 2 OSTs" && return
 
        echo -n "Free space priority "
@@ -6033,7 +6046,21 @@ test_116() {
 
        rm -rf $DIR/$tdir
 }
-run_test 116 "stripe QOS: free space balance ==================="
+run_test 116a "stripe QOS: free space balance ==================="
+
+test_116b() { # LU-2093
+#define OBD_FAIL_MDS_OSC_CREATE_FAIL     0x147
+       local old_rr
+       old_rr=$(do_facet $SINGLEMDS lctl get_param -n lov.*mdtlov*.qos_threshold_rr)
+       do_facet $SINGLEMDS lctl set_param lov.*mdtlov*.qos_threshold_rr 0
+       mkdir -p $DIR/$tdir
+       do_facet $SINGLEMDS lctl set_param fail_loc=0x147
+       createmany -o $DIR/$tdir/f- 20 || error "can't create"
+       do_facet $SINGLEMDS lctl set_param fail_loc=0
+       rm -rf $DIR/$tdir
+       do_facet $SINGLEMDS lctl set_param lov.*mdtlov*.qos_threshold_rr $old_rr
+}
+run_test 116b "QoS shouldn't LBUG if not enough OSTs found on the 2nd pass"
 
 test_117() # bug 10891
 {
@@ -7066,7 +7093,7 @@ set_dir_limits () {
        done
 }
 test_129() {
-       if [ "$(facet_type_fstype MDS)" != ldiskfs ]; then
+       if [ "$(facet_fstype $SINGLEMDS)" != ldiskfs ]; then
                skip "Only applicable to ldiskfs-based MDTs"
                return
        fi
@@ -9137,7 +9164,7 @@ verify_jobstats() {
 
        # clear old jobstats
        do_facet $SINGLEMDS lctl set_param mdt.*.job_stats="clear"
-       do_facet ost0 lctl set_param obdfilter.*.job_stats="clear"
+       do_facet ost1 lctl set_param obdfilter.*.job_stats="clear"
 
        # use a new JobID for this test, or we might see an old one
        [ "$JOBENV" = "FAKE_JOBID" ] && FAKE_JOBID=test_id.$testnum.$RANDOM
@@ -9158,7 +9185,7 @@ verify_jobstats() {
                        grep $JOBVAL || error "No job stats found on MDT $FACET"
        fi
        if [ "$target" = "ost" -o "$target" = "both" ]; then
-               FACET=ost0
+               FACET=ost1
                do_facet $FACET lctl get_param obdfilter.*.job_stats |
                        grep $JOBVAL || error "No job stats found on OST $FACET"
        fi
@@ -9859,7 +9886,7 @@ test_900() {
 }
 run_test 900 "umount should not race with any mgc requeue thread"
 
-complete $(basename $0) $SECONDS
+complete $SECONDS
 check_and_cleanup_lustre
 if [ "$I_MOUNTED" != "yes" ]; then
        lctl set_param debug="$OLDDEBUG" 2> /dev/null || true