On Thu, Jun 10, 2010 at 1:25 PM, Aleksey Tsalolikhin
<atsaloli.tech@gmail.com> wrote:
> PostgreSQL 8.4.2 response time slows during pg_dump of a 14 GB database.
>
...
>
> Is there any way I can keep my response fast but still get my pg_dump?
> Or should I move my pg_dump to a Slony-I slave?
Thank you for your kind replies.
More data:
1. We're not dumping across the network. The pg_dump
process runs on the same physical server as PostgreSQL.
2. We're not CPU bound - CPU utilization goes from 95% idle
to 75% idle during the pg_dump.
3. We ARE disk bound. Disk utilization goes to 94% during
the dump. There is an order of magnitude more reads than
writes.
Greg, I didn't see anything in the "I/O" section of
http://wiki.postgresql.org/wiki/Priorities I could use.
We ionice our pg_dump process, but it's a PostgreSQL
process that stresses the disk.
Thanks anyway - glad to hear dumping from a slave is
a popular approach, that's what we'll do. This is good,
it'll allow us to increase our pg_dumps from 1 a day to
as many as we want to feel safe.
Best,
-at