If split brain happens and network merges back, at this point in time there are a few mds events coming to payloads, which are the SVC UP from the other controller; SVC down from services in both controllers due to reboot from split brain detection. In the ticket description, the first partition includes SC1, PL3, the second partition includes SC2, PL4, PL5. The amfnd on PL3 is missing NCSMDS_DOWN with vdest in the below scenario:
- SVC up event from the other amfd (on SC2)
- SVC down event from amfd (SC1), it's the same active adest from
mds-PL3's view, start await_active timer, but no NCSMDS_DOWN with
vdest is sent because the adest on SC2 exists.
- SVC down event from amfd (SC2), it's different active adest.
Because the payloads reside in different partitions so they don't
have the same active adest view at mds level. When both SCs go down
due to split brain detection, the same SVC down events occur and
comes to all payloads, but they have different view so they behave
differently to the payloads in the other partition.
The patch adds an additional condition to send NCSMDS_DOWN if there is
no actual adest existed
---
src/mds/mds_c_api.c | 80 ++++++++++++++++++++++++++++++-----------------------
1 file changed, 46 insertions(+), 34 deletions(-)
diff --git a/src/mds/mds_c_api.c b/src/mds/mds_c_api.c
index f5ba318..73849cc 100644
--- a/src/mds/mds_c_api.c
+++ b/src/mds/mds_c_api.c
@@ -3644,13 +3644,58 @@ uint32_t mds_mcm_svc_down(PW_ENV_ID pwe_id, MDS_SVC_ID
svc_id, V_DEST_RL role,
local_svc_hdl, svc_id, vdest_id,
&active_adest, &tmr_running,
&subtn_result_info, true);
-
+ m_MDS_LOG_INFO("MCM:API: svc_down: "
+ "active_adest:%lu", active_adest);
/* First delete the entry */
mds_subtn_res_tbl_del(
local_svc_hdl, svc_id, vdest_id, adest,
vdest_policy, svc_sub_part_ver,
archword_type);
+ MDS_SUBSCRIPTION_RESULTS_INFO *s_info = NULL;
+ bool adest_exists = false;
+
+ /* if no adest remains for this svc
+ * send MDS_DOWN
+ */
+ status = mds_subtn_res_tbl_getnext_any(
+ local_svc_hdl, svc_id,
+ &s_info);
+
+ while (status != NCSCC_RC_FAILURE) {
+ if (s_info->key.vdest_id !=
+ m_VDEST_ID_FOR_ADEST_ENTRY) {
+ adest_exists = true;
+ break;
+ }
+
+ status = mds_subtn_res_tbl_getnext_any(
+ local_svc_hdl, svc_id, &s_info);
+ }
+
+ if (active_adest != adest
+ && vdest_policy == NCS_VDEST_TYPE_MxN
+ && adest_exists == false) {
+ m_MDS_LOG_INFO("MCM:API: svc_down : "
+ "svc_id = %s(%d) on DEST id =
%d "
+ "got NO_ACTIVE for svc_id =
%s(%d) "
+ "on Vdest id = %d Adest = %s, rem_svc_pvt_ver=%d",
+ get_svc_names(
+
m_MDS_GET_SVC_ID_FROM_SVC_HDL(local_svc_hdl)),
+ m_MDS_GET_SVC_ID_FROM_SVC_HDL(
+ local_svc_hdl),
+ m_MDS_GET_VDEST_ID_FROM_SVC_HDL(
+ local_svc_hdl),
+ get_svc_names(svc_id), svc_id,
+ vdest_id,
+
log_subtn_result_info->sub_adest_details,
+ svc_sub_part_ver);
+ status = mds_mcm_user_event_callback(
+ local_svc_hdl, pwe_id, svc_id,
+ role, vdest_id, 0, NCSMDS_DOWN,
+ svc_sub_part_ver,
archword_type);
+ }
+
if (active_adest == adest) {
if (vdest_policy ==
NCS_VDEST_TYPE_MxN) {
@@ -3698,39 +3743,6 @@ uint32_t mds_mcm_svc_down(PW_ENV_ID pwe_id, MDS_SVC_ID
svc_id, V_DEST_RL role,
->sub_adest_details,
svc_sub_part_ver);
{
-
MDS_SUBSCRIPTION_RESULTS_INFO
- *subtn_result_info =
- NULL;
- bool adest_exists =
- false;
-
- /* if no adest remains
- * for this svc, send
- * MDS_DOWN */
- status =
-
mds_subtn_res_tbl_getnext_any(
- local_svc_hdl,
- svc_id,
-
&subtn_result_info);
-
- while (
- status !=
- NCSCC_RC_FAILURE) {
- if
(subtn_result_info
- ->key
-
.vdest_id !=
-
m_VDEST_ID_FOR_ADEST_ENTRY) {
-
adest_exists =
-
true;
- break;
- }
-
- status =
mds_subtn_res_tbl_getnext_any(
-
local_svc_hdl,
- svc_id,
-
&subtn_result_info);
- }
-
if (adest_exists ==
false) {
/* No other
--
2.7.4
_______________________________________________
Opensaf-devel mailing list
Opensaf-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/opensaf-devel
