We've been running PGCluster for about 6 months, with relatively few hiccups.
Our setup is 1 master, 2 slaves, with all 3 nodes writable. We also have 2
tables, listed in the <Not_Replicate_Info> section of the config file, which
exist on all 3 boxes. Writes to these tables are not replicated as expected.
Today, I noticed that the 2 slaves were out of sync with the master (1 of the
tables contained different results than what was shown on the master). As
this is our development cluster, I just brought everything down, started the
master, and brought 1 of the slaves up in recovery mode. While this did
bring the slave back into sync, it also overwrote the 2 tables that were
marked as NOT TO BE REPLICATED. If this is expected behavior, I just want to
be aware of it, so I find some other way to recover from problems we
encounter. Since this is our dev cluster, it isn't that big of a deal. But
once we go live with this setup, losing that much data is not acceptable.
Is there something I might have misconfigured to prevent those 2 tables from
being processed as part of a recovery? If all "-R" does is blindly rsync the
contents of the /data (and associated) directories, I need to know.
|
