On Tue, Jun 19, 2018 at 03:56:15PM +0800, hemin wrote:
> I do not checkpoint before running pg_rewind, because the checkpoint
> of rewound primary node is obviously slow than the promoted standby,
> But I will try it later.
You need to run the checkpoint on the promoted standby. That's actually
a critical thing to do to ensure consistency.. Of course this depends
on how you set max_wal_size.
--
Michael