On Friday, May 27, 2016 05:32:08 PM Melvin Davidson wrote:
> Well, Slony certainly will do the trick.
> Keep in mind you will need to do schema only first to the slave.
> You set up replication from the old server with the db on the new server as
> the slave. Then you initiate replication. It will probably take a long time
> to
> replicate, but then you have the option to promote the slave at your time
> preference (IE: your 2 hr window). It should only take a few minutes for
> Slony to do the switchover, but the best thing to do is a dry run first.
> IOW, you'll have to do the whole thing twice to get an accurate switch time,
> but you won't need to change your network until you are ready to go live.
Slony doesn't do BLOBs, afaik, unless he's using BYTEA fields.
Otherwise I believe dump/reload is OP's only choice. He should be able to do
90GB in 2 hours on fast enough hardware; just pipe it over the network to do
the restore simultaneous with the dump.
Also remove as many indexes as possible beforehand and use create concurrently
manually afterwards to add them back in.