Bug 768144

Summary: quorum_notification_fn_t gives wrong ring_seq
Product: [Retired] Corosync Cluster Engine Reporter: David Teigland <teigland>
Component: quorumAssignee: Fabio Massimo Di Nitto <fdinitto>
Status: CLOSED UPSTREAM QA Contact:
Severity: unspecified Docs Contact:
Priority: unspecified    
Version: 1.4CC: asalkeld, cluster-maint, jfriesse, sdake
Target Milestone: ---   
Target Release: ---   
Hardware: Unspecified   
OS: Unspecified   
Whiteboard:
Fixed In Version: Doc Type: Bug Fix
Doc Text:
Story Points: ---
Clone Of: Environment:
Last Closed: 2013-06-20 14:40:54 UTC Type: ---
Regression: --- Mount Type: ---
Documentation: --- CRM:
Verified Versions: Category: ---
oVirt Team: --- RHEL 7.3 requirements from Atomic Host:
Cloudforms Team: --- Target Upstream Version:
Attachments:
Description Flags
corosync.conf none

Description David Teigland 2011-12-15 21:15:23 UTC 
Description of problem:

After a node is killed, the libquorum callback gives the old ring_seq number, not the new one.  This breaks the in-development version of dlm_controld.

Extracting some messages from dlm_controld log:

- all cluster nodes join in seq 4972
8820 cluster quorum 1 seq 4972 nodes 4
8820 cluster node 1 added seq 4972
8820 cluster node 2 added seq 4972
8820 cluster node 4 added seq 4972
8820 cluster node 5 added seq 4972

- cpg totem callbacks show correct ring seq
8820 dlm:controld ring 1:4972 4 memb 1 2 4 5

- node 4 killed and removed from cluster
  this should obviously be ring seq 4976, not 4972
10233 cluster quorum 1 seq 4972 nodes 3
10233 cluster node 4 removed seq 4972

- cpg totem callbacks show new correct ring seq
10233 dlm:controld ring 1:4976 3 memb 1 2 5

(On a related note, it would be really useful if the corosync
messages in /var/log/messages included the ring seq number so
logs could be correlated among nodes, and with other code.
Similarly, no corosync commands display any ring seq numbers,
which cman_tool did.)

Version-Release number of selected component (if applicable):

corosync-1.2.3-36.el6.x86_64

How reproducible:


Steps to Reproduce:
1.
2.
3.
  
Actual results:


Expected results:


Additional info:
Comment 2 David Teigland 2011-12-15 21:41:38 UTC 
Also happens if a just shut down a node using corosync-cfgtool -H.
Comment 3 David Teigland 2011-12-16 17:05:02 UTC 
Created attachment 547878 [details]
corosync.conf
Comment 4 David Teigland 2011-12-16 17:26:33 UTC 
Fabio says votequorum is irrelevant.  Reassigning to make sure this works properly in the new quorum module.
Comment 5 Fabio Massimo Di Nitto 2011-12-17 07:28:58 UTC 
Either way, there is a bug in votequorum confchg_fn that does not send notifications back to to the quorum module and that would explain why you see different ringids based on when certain things are happening.
Comment 6 Fabio Massimo Di Nitto 2011-12-19 13:46:56 UTC 
This is definitely a bug in votequorum but the API is not supported in RHEL. Moving to upstream.
Comment 7 Fabio Massimo Di Nitto 2012-01-06 04:05:10 UTC 
Fixed in the topic-quorum branch
Comment 8 Jan Friesse 2013-06-20 14:40:54 UTC 
topic-quorum is merged for ages now, so closing as upstream.