Re: Removing dead server from replication ring

[Geoff Kuenning]

I found a simple solution: the CHANGE MASTER TO command. On C, I did CHANGE MASTER TO MASTER_HOST = 'A.example.com'; What that did was somewhat different from what I expected, although in the end it worked. What I expected was that C would start reading A's logs from whereever it had left off on B. Instead, it started reading A's logs from the oldest, which was a bit of a nuisance for two reasons: first, I had never done a PURGE BINARY LOGS so there were a lot of them and so the replay took a while. Second, when it ran into a duplicate INSERT it stopped. I resolved the latter by temporarily changing my.cnf on C to contain slave-skip-errors = 1062, as detailed here: http://www.ducea.com/2008/02/13/mysql-skip-duplicate-replication-errors/

After making that change and waiting for a while, the replay finished, C caught up with A, and I then removed the slave-skip-errors line and restarted the server.

This approach is described in the mysql documents; I didn't save the URL but search for "failover" as the magic term. I should note for others who try it that in my setup we only use INSERT and DELETE commands; we don't tend to use UPDATE. So I can't guarantee that you'll get the same results if you use UPDATEs.