On Tue, Aug 20, 2013 at 2:37 AM, Heikki Linnakangas <hlinnakangas@vmware.com> wrote: > Throttling in the client seems much better to me. TCP is designed to handle > a slow client.
Other people have already offered some good points in this area, but let me just add one thought that I don't think has been mentioned yet. We have a *general* need to be able to throttle server-side resource utilization, particularly I/O. This is a problem not only for pg_basebackup, but for COPY, CLUSTER, VACUUM, and even things like UPDATE. Of all of those, the only one for which we currently have any kind of a solution is VACUUM. Now, maybe pg_basebackup also needs its own special-purpose solution, but I think we'd do well to consider a general I/O rate-limiting strategy and then consider particular needs in the light of that framework. In that context, server-side seems better to me, because something like CLUSTER isn't going to produce anything that the client can effectively limit.
+1 it is very easy at the moment to for example run a manual vacuum full/cluster against a big table and generate WAL so quickly that the hot standby disconnects because it gets "too far behind".