On 7/6/2016 7:10 PM, Patrick B wrote:
I'll archive the wal_files locally on slave01, and then send them to
slave02 every 10 minutes, for example. This will be done by rsync or
something else.
I'll have still the problem with timeline, and slave02 won't be able
to become a master in the future.
the master should archive to a file server (NFS or whatever). all slaves
should use this wal archive as a 'catch up source' when they resume
after downtime. [*]
master /also/ sends streaming replication to any direct attached slaves,
and those slaves can send streaming replication to any cascaded slaves,
repeat.
copying WAL archives around via rsync just seems cray-cray to me.
[*] in the latest version, there are 'replication slots' which provides
a completely different mechanism for this catch-up that doesn't require
WAL archiving
--
john r pierce, recycling bits in santa cruz
