Galera Cluster discussions at FrOSCon 2012

During and after Henriks great talk about Galera Cluster at the FrOSCon 2012 in St. Augustin we found 2 important things related to Galera Cluster for MySQL:

• The InnoDB double write buffer (innodb_doublewrite) should not be disabled anymore for Galera when using v2.0 and higher!!! The reason for this is: When MySQL crashes InnoDB pages might get corrupted during the crash. They would be fixed by the blocks from the double write buffer during auto-recovery. But if the double write buffer is disabled they are not available. With Galera v1.x that was not a problem because after a crash a SST would have happened and the corrupted InnoDB block are corrected. But now with IST in Galera v2.0 MySQL will start without noticing the corruption (as usual) and only an IST is performed. This leads to a running MySQL database with possibly corrupted InnoDB blocks. And this might cause you later troubles for example if this node is used as a donor. Then the corrupted page is inherited to other nodes (using rsync or Xtrabackup?). And in some bad cased then the whole Cluster could crash at once when hitting the corrupted page. Thanks to Monty W. for bringing this up!
  Recommendation is: Do NOT disable InnoDB double write buffer (innodb_doublewrite) with Galera Cluster >= v2.0 if your care about your data!
  
• The second discussion was about the event sequence in the binary-log (for those who where present: the A-B vs B-A discussion). Codership confirmed that the binary-log sequence on 2 different Galera nodes of the same Galera Cluster should be the same (everything else is considered to be a bug). As a result this leads to 2 different consequences:
  a) The binary-log of node B can be used for a PiTR of node A in case we need it. Finding the right position is a bit tricky and it needs some manual work on this (finding XID with binlog-pos of node B, then finding binlog-pos of node A with XID). But Codership told me they are planning a tool for automatizing this.
  b) The binary-log of node B can be used for a Channel failover in case we have 2 different Galera Clusters in 2 different data centers connect to each other through MySQL asynchronous replication... For more on this topic see also MySQL Cluster and channel failover...