Re: Big UPDATE breaking replication

On Thu, Jun 6, 2013 at 5:19 AM, Kouber Saparev <kouber(at)saparev(dot)com> wrote:

> Do you mean to archive the WALs somewhere and then scp them with a cron
> job? I doubt it would be fast enough neither.
>

I haven't been paying attention to this thread, but this comment caught my
eye. I had a setup at a previous job where the primary and DR datacenters
were at opposite ends of the county making the network latency high enough
that WALs piled up on the master. What we did was to archive the WALs
locally (to the master) in a dedicated volume, then rsync'd them in batches
of 50 or so. It made a big difference for us, as we were generating around
100 WAL segments per minute IIRC. I think the latency was around 46ms and
during bursts we would fall behind quite a bit.