Comment 9 for bug 1123233

From the logs and core file, it is very clear what is the immediate reason for the crash. There is a transaction which has not logged any modified row key values for the replication provider, and from provider perspective this transaction does not exists. However, at commit time, the transaction cache contains some replication events and node decides to do emergency shutdown to avoid potential inconsistency.

Attempts to reproduce this issue have not been successful so far, and it is still a mystery why transaction processing ended into this troubled state.

Pushed new revision: http://bazaar.launchpad.net/~codership/codership-mysql/5.5-23/revision/3857, which logs a warning in error log and prints the excessive RBR buffer in GRA_* file for later inspection.