From: | Ron Johnson <ronljohnsonjr(at)gmail(dot)com> |
---|---|
To: | Pgsql-admin <pgsql-admin(at)lists(dot)postgresql(dot)org> |
Subject: | pg_restore and shared_buffers |
Date: | 2024-01-08 20:22:51 |
Message-ID: | CANzqJaBMccmJn8_GiOcDEZ3xrZNG=8ta_L87C83xU7-_VLR1iA@mail.gmail.com |
Views: | Raw Message | Whole Thread | Download mbox | Resend email |
Thread: | |
Lists: | pgsql-admin |
(PG14.10, restoring a pg_dump created from a PG 9.6.24 instance. Restores
take about 10 hours, so testing must be more thoughtful than
see-what-works.)
This 9 year old blog post from Josh Berkus recommends setting
shared_buffers to 1/2 its normal value.
https://www.databasesoup.com/2014/09/settings-for-fast-pgrestore.html
My normal shared_buffers setting is 32GB (on a 132GB VM), so dropped it
down to 16GB (with 1GB mainteance_work_mem). Swap usage is still slowly
creeping up during the COPY phase, while using --jobs=24.
I'm wondering if 16GB is still too high, since the threads don't share
anything. (Or do they???)
Any benefit in dropping shared_buffers as low as 4GB?
From | Date | Subject | |
---|---|---|---|
Next Message | Rakesh Nashine | 2024-01-09 11:59:51 | Re: Ingress to PostgreSQL Migration |
Previous Message | Rambabu V | 2024-01-08 17:26:31 | Re: Db nodes are getting deregistering from consul |