Bug #30937 endless loop with ring-spoke topology
Submitted: 10 Sep 2007 19:07 Modified: 2 Aug 2008 1:12
Reporter: Sloan Childers Email Updates:
Status: Closed Impact on me:
None 
Category:MySQL Enterprise Monitor: Server Severity:S2 (Serious)
Version:2.0 OS:Any
Assigned to: Darren Oldag CPU Architecture:Any
Tags: mer 120

[10 Sep 2007 19:07] Sloan Childers
Description:
Endless loop trying to consolodate group names kills the server.

I fired up replmxj and used the RingSpoke topology.

I first encountered this bug:
http://bugs.mysql.com/bug.php?id=30922

And as a result of the order of discovery we ended up in an endless loop trying to consolodate groups.

How to repeat:
Oldag has created a test case that reproduces the race condition/order of topology discovery condition.

Suggested fix:
Fix infinite loop so we can recover from the FK error.
[10 Sep 2007 19:12] Darren Oldag
if a ring/spoke topology discovery didn't complete the first time,
for any of various reasons, the server could not recover as it would
enter an infinite loop update the ring replication group.

the fix is to remember the originator of the update, and use that as
the end-condition for terminating the update of the ring.

--

committed to trunk, awaiting review for push into 1.2 release branch.
[10 Sep 2007 20:14] Sloan Childers
r7559,r7560 trunk
r7561 br_6559_1.2.0
[10 Sep 2007 21:59] Sloan Childers
r7562 development-1.2.1
[17 Nov 2007 18:47] Sloan Childers
Fixed in 1.2.1, reopening to fix in 2.0
[2 Aug 2008 1:12] Bill Weber
doesn't occur in build 2.0.0.7017