Hi,
In normal situation when I stop Opensaf or reboot the node, I often get
MDS down first, then CLM track callback. [1]
The problem in this ticket is because AMFD has CLM track callback coming
before MDS down. [2]
We need to see it as a race condition, not because of missing the
counters reset, so that the amfd's function to deal with MDS down should
be called. That would be equivalent to [1], which is currently working.
With this patch, we still don't have avd_mds_avnd_down_evh() to be
called, that function processes many things in the event of mds down.
(ie. @nodes_exit_cnt, ...)
I think the node_info.member should be set TRUE or FALSE in the event of
CLM callback since it means CLM membership. We can use both @member and
one of variables indicating MDS down, to trap each other in the events
of MDS down and CLM callback, so that in the end we have
avd_mds_avnd_down_evh() and nodeid is removed out of node_id_db in both
[1] and [2].
Thanks,
Minh
On 06/07/18 20:36, thuan.tran wrote:
There is a case that after AMFD send reboot order due to “out of sync window”.
AMFD receive CLM track callback but node is not member yet and delete node.
Later AMFND MDS down will not reset msg_id counter since it cannot find node.
When node reboot up, AMFD continue use current msg_id counter send to AMFND
cause messasge ID mismatch in AMFND then AMFND order reboot itself node.
---
src/amf/amfd/clm.cc | 7 ++++++-
1 file changed, 6 insertions(+), 1 deletion(-)
diff --git a/src/amf/amfd/clm.cc b/src/amf/amfd/clm.cc
index e113a65f9..4a15d5ad7 100644
--- a/src/amf/amfd/clm.cc
+++ b/src/amf/amfd/clm.cc
@@ -316,9 +316,14 @@ static void clm_track_cb(
__FUNCTION__, node_name.c_str());
goto done;
} else if (node->node_state == AVD_AVND_STATE_ABSENT) {
- LOG_IN("%s: CLM node '%s' is not an AMF cluster member; MDS down
received",
+ LOG_IN("%s: CLM node '%s' is ABSENT; MDS down received",
__FUNCTION__, node_name.c_str());
avd_node_delete_nodeid(node);
+ /* Reset msg_id because AVND MDS down may come later
+ and cannot find node to reset these, cause message ID mismatch.
*/
+ node->rcv_msg_id = 0;
+ node->snd_msg_id = 0;
+ node->node_info.member = SA_FALSE;
goto done;
}
TRACE(" Node Left: rootCauseEntity %s for node %u",
------------------------------------------------------------------------------
Check out the vibrant tech community on one of the world's most
engaging tech sites, Slashdot.org! http://sdm.link/slashdot
_______________________________________________
Opensaf-devel mailing list
Opensaf-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/opensaf-devel
