Bug #20865 Node restart might fail if SQL node with active binlog disconnects
Submitted: 5 Jul 2006 15:31 Modified: 10 Jan 2007 22:09
Reporter: Martin Skold
Status: Analyzing
Category:Server: ClusterRep Severity:S3 (Non-critical)
Version:mysql-5.1-telco-7.0 OS:Any
Assigned to: Martin Skold Target Version:
Triage: Triaged: D3 (Medium) / R6 (Needs Assessment) / E6 (Needs Assessment)

[5 Jul 2006 15:31] Martin Skold
Description:
If a MySQL server (API node) with active subscriptions (for binlog) disconnects
from cluster while a data node restarts, the copying of subscriptions
to this node might fail (causing node restart to fail) due to clean-up of subscription
data done for all subscriptions the belong to the disconnected API node. If the data node
is configured to restart immediately it should succeed the second time. 

How to repeat:
No reproducable test case found yet, this bug report is just a reminder
to continue a bug fix that is more or less in place.

Suggested fix:
Lock subscription data while copying to a restarting node so that
it cannot be removed until copying is completed.