PG10 upgrade issue

Started by Zahir Lalanialmost 7 years ago3 messagesgeneral
Jump to latest
#1Zahir Lalani
ZahirLalani@oliver.agency

Hello

We have done pg10 upgrades from 9.6 on all our environments successfully. We have an automation script that does this for us. One environment is a near replica of our production setup and all works fine.

We have now made 2 attempts at doing this on production - both ending up with rollback

The key issue we have is that all queries were incredibly slow. The PG 10 postmaster workers would all show 100% cpu, especially on the replicas. We even increased CPUs but that made no difference.

I can try and provide logs info etc, but hoping someone has a clue about this

Thx

Zahir

#2Tom Lane
tgl@sss.pgh.pa.us
In reply to: Zahir Lalani (#1)
Re: PG10 upgrade issue

Zahir Lalani <ZahirLalani@oliver.agency> writes:

We have done pg10 upgrades from 9.6 on all our environments successfully. We have an automation script that does this for us. One environment is a near replica of our production setup and all works fine.

We have now made 2 attempts at doing this on production - both ending up with rollback

The key issue we have is that all queries were incredibly slow. The PG 10 postmaster workers would all show 100% cpu, especially on the replicas. We even increased CPUs but that made no difference.

Did you make sure to rebuild statistics (re-ANALYZE) after the upgrade?
pg_upgrade makes a script for that, but I don't believe it runs it for
you.

regards, tom lane

#3Zahir Lalani
ZahirLalani@oliver.agency
In reply to: Tom Lane (#2)
Re: PG10 upgrade issue

Ah no we did not!

Will give that a go next tim Tom

Thank you

Z

On 3 Jun 2019 00:18, Tom Lane <tgl@sss.pgh.pa.us> wrote:
Zahir Lalani <ZahirLalani@oliver.agency> writes:

We have done pg10 upgrades from 9.6 on all our environments successfully. We have an automation script that does this for us. One environment is a near replica of our production setup and all works fine.

We have now made 2 attempts at doing this on production - both ending up with rollback

The key issue we have is that all queries were incredibly slow. The PG 10 postmaster workers would all show 100% cpu, especially on the replicas. We even increased CPUs but that made no difference.

Did you make sure to rebuild statistics (re-ANALYZE) after the upgrade?
pg_upgrade makes a script for that, but I don't believe it runs it for
you.

regards, tom lane