Re: Running pg_dump from a slave server

2016-08-17 15:31 GMT+12:00 Sameer Kumar <sameer.kumar@ashnik.com>:

On Wed, Aug 17, 2016 at 10:34 AM Patrick B <patrickbakerbr@gmail.com> wrote:

Hi guys,

I'm using PostgreSQL 9.2 and I got one master and one slave with streaming replication.

Currently, I got a backup script that runs daily from the master, it generates a dump file with 30GB of data.

I changed the script to start running from the slave instead the master, and I'm getting this errors now:

pg_dump: Dumping the contents of table "invoices" failed: PQgetResult() failed.
pg_dump: Error message from server: ERROR:  canceling statement due to conflict with recovery
DETAIL:  User was holding a relation lock for too long.

Looks like while your pg_dump sessions were trying to fetch the data, someone fired a DDL or REINDEX or VACUUM FULL on the master database. 

Isn't that possible? I can't run pg_dump from a slave?

Well you can do that, but it has some limitation. If you do this quite often, it would be rather better to have a dedicated standby for taking backups/pg_dumps. Then you can set max_standby_streaming_delay and max_standby_archiving_delay to -1. But I would not recommend doing this if you use your standby for other read queries or for high availability.

Another option would be avoid such queries which causes Exclusive Lock on the master database during pg_dump.

Sameer, 

yeah I was just reading this thread: https://www.postgresql.org/message-id/AANLkTinLg%2BbpzcjzdndsnGGNFC%3DD1OsVh%2BhKb85A-s%3Dn%40mail.gmail.com

Well.. I thought it was possible, but as the DB is big, this dump takes a long time and it won't work.

I also could increase those parameters you showed, but won't do that as I only have one slave.

cheers