On Thu, Feb 24, 2011 at 6:38 PM, Aleksey Tsalolikhin
<atsaloli(dot)tech(at)gmail(dot)com> wrote:
> In practice, if I pg_dump our 100 GB database, our application, which
> is half Web front end and half OLTP, at a certain point, slows to a
> crawl and the Web interface becomes unresponsive. I start getting
> check_postgres complaints about number of locks and query lengths. I
> see locks around for over 5 minutes.
>
I'd venture to say your system does not have enough memory and/or disk
bandwidth, or your Pg is not tuned to make use of enough of your
memory. The most likely thing is that you're saturating your disk
I/O.
Check the various system statistics from iostat and vmstat to see what
your baseline load is, then compare that when pg_dump is running. Are
you dumping over the network or to the local disk as well?