LU-9581 tests: remove recovery-small test from ALWAYS_EXCEPT

[fs/lustre-release.git] / lustre / tests / recovery-small.sh

diff --git a/lustre/tests/recovery-small.sh b/lustre/tests/recovery-small.sh
index 02fc696..125051b 100755 (executable)
--- a/lustre/tests/recovery-small.sh
+++ b/lustre/tests/recovery-small.sh
@@ -2,8 +2,9 @@
 
 set -e
 
-#         bug  5493  LU2034
-ALWAYS_EXCEPT="52 $RECOVERY_SMALL_EXCEPT"
+# bug number for skipped test:
+ALWAYS_EXCEPT="$RECOVERY_SMALL_EXCEPT"
+# UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
 
 export MULTIOP=${MULTIOP:-multiop}
 PTLDEBUG=${PTLDEBUG:--1}
@@ -1756,26 +1757,45 @@ test_100()
 }
 run_test 100 "IR: Make sure normal recovery still works w/o IR"
 
-test_101()
+test_101a()
 {
-        do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
-                { skip "MGS without IR support"; return 0; }
+       do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
+               skip "MGS without IR support"
 
-        set_ir_status full
+       set_ir_status full
 
-        local OST1_IMP=$(get_osc_import_name client ost1)
+       local ost1_imp=$(get_osc_import_name client ost1)
 
-        # disable pinger recovery
-        lctl set_param -n osc.$OST1_IMP.pinger_recov=0
+       # disable pinger recovery
+       lctl set_param -n osc.$ost1_imp.pinger_recov=0
+       stack_trap "$LCTL set_param -n osc.$ost1_imp.pinger_recov=1" EXIT
 
-        fail ost1
+       fail ost1
 
-        target_instance_match ost1 || error "instance mismatch"
-        nidtbl_versions_match || error "version must match"
+       target_instance_match ost1 || error "instance mismatch"
+       nidtbl_versions_match || error "version must match"
+}
+run_test 101a "IR: Make sure IR works w/o normal recovery"
 
-        lctl set_param -n osc.$OST1_IMP.pinger_recov=1
+test_101b()
+{
+       do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
+               skip "MGS without IR support"
+
+       set_ir_status full
+
+       local ost1_imp=$(get_osc_import_name client ost1)
+
+#define OBD_FAIL_OST_PREPARE_DELAY      0x247
+       do_facet ost1 "$LCTL set_param fail_loc=0x247"
+       # disable pinger recovery
+       $LCTL set_param -n osc.$ost1_imp.pinger_recov=0
+       stack_trap "$LCTL set_param -n osc.$ost1_imp.pinger_recov=1" EXIT
+
+#OST may return EAGAIN if it is not configured yet
+       fail ost1
 }
-run_test 101 "IR: Make sure IR works w/o normal recovery"
+run_test 101b "IR: Make sure IR works w/o normal recovery and proceed EAGAIN"
 
 test_102()
 {
@@ -2739,6 +2759,76 @@ test_134() {
 }
 run_test 134 "race between failover and search for reply data free slot"
 
+test_135() {
+       [ $MDS1_VERSION -lt $(version_code 2.12.51) ] &&
+               skip "Need MDS version at least 2.12.51"
+
+       mkdir -p $DIR/$tdir
+       $LFS setstripe -E 1M -L mdt $DIR/$tdir
+       # to have parent dir write lock before open/resend
+       touch $DIR/$tdir/$tfile
+       #define OBD_FAIL_MDS_LDLM_REPLY_NET 0x157
+       do_nodes $(comma_list $(mdts_nodes)) $LCTL set_param fail_loc=0x80000157
+       openfile -f O_RDWR:O_CREAT -m 0755 $DIR/$tdir/$tfile ||
+               error "Failed to open DOM file"
+}
+run_test 135 "DOM: open/create resend to return size"
+
+test_136() {
+       remote_mds_nodsh && skip "remote MDS with nodsh" && return
+       [[ $MDS1_VERSION -ge $(version_code 2.12.52) ]] ||
+               skip "Need MDS version at least 2.12.52"
+
+       local mdts=$(comma_list $(mdts_nodes))
+       local MDT0=$(facet_svc $SINGLEMDS)
+
+       local clog=$(do_facet mds1 $LCTL --device $MDT0 changelog_register -n)
+       [ -n "$clog" ] || error "changelog_register failed"
+       cl_mask=$(do_facet mds1 $LCTL get_param \
+                               mdd.$MDT0.changelog_mask -n)
+       changelog_chmask "ALL"
+
+       # generate some changelog records to accumulate
+       test_mkdir -i 0 -c 0 $DIR/$tdir || error "mkdir $tdir failed"
+       createmany -m $DIR/$tdir/$tfile 10000 ||
+               error "create $DIR/$tdir/$tfile failed"
+
+       local size1=$(do_facet $SINGLEMDS \
+                     $LCTL get_param -n mdd.$MDT0.changelog_size)
+       echo "Changelog size $size1"
+
+       #define OBD_FAIL_LLOG_PURGE_DELAY                   0x1318
+       do_nodes $mdts $LCTL set_param fail_loc=0x1318 fail_val=30
+
+       # launch changelog_deregister in background on MDS
+       do_facet mds1 "nohup $LCTL --device $MDT0 changelog_deregister $clog \
+                       > foo.out 2> foo.err < /dev/null &"
+       # give time to reach fail_loc
+       sleep 15
+
+       # fail_loc will make MDS sleep in the middle of changelog_deregister
+       # take this opportunity to abruptly kill MDS
+       FAILURE_MODE_save=$FAILURE_MODE
+       FAILURE_MODE=HARD
+       fail mds1
+       FAILURE_MODE=$FAILURE_MODE_save
+
+       do_nodes $mdts $LCTL set_param fail_loc=0x0 fail_val=0
+
+       local size2=$(do_facet $SINGLEMDS \
+                     $LCTL get_param -n mdd.$MDT0.changelog_size)
+       echo "Changelog size $size2"
+       local clog2=$(do_facet $SINGLEMDS "$LCTL get_param -n \
+                       mdd.$MDT0.changelog_users | grep $clog")
+       echo "After crash, changelog user $clog2"
+
+       [ -n "$clog2" -o $size2 -lt $size1 ] ||
+               error "changelog record count unchanged"
+
+       do_facet mds1 $LCTL set_param mdd.$MDT0.changelog_mask=\'$cl_mask\' -n
+}
+run_test 136 "changelog_deregister leaving pending records"
+
 complete $SECONDS
 check_and_cleanup_lustre
 exit_status