Thread: Reporting from Standby

Good afternoon.

PG 12.12, PGDG ubuntu 18.04 LTS build.

We have a primary with a couple of physical/streaming replicas. One intended for quick lookups and another intended for long-running reports (aka "the reporting replica"). There's also a DR replica in another region but we don't touch that for obvious reasons. All streaming replicas currently make use of replication slots on the primary as well. We have WAL archiving to cloud storage and restore/recovery handled via pgBackrest.

I'm running into the age-old dilemma on the reporting replica where queries are being killed with the "canceling statement due to conflict with recovery" error. We have hot_standby_feedback=on and both max_standby_streaming_delay and max_standby_archive_delay are set to 5 minutes (300s). I'm fine increasing those delay parameters to 15 minutes or more. Like I said this replica is dedicated to long-running reports like month-end reports that you'd see today on the first of the month.

I searched around and I see some suggestions like disabling hot_standby_feedback on replicas like this to minimize impact on the primary, in addition to increasing the delays, or even disabling the delays by setting them to -1. With replication slots in use, I want to make sure that WAL retention doesn't fill up the WAL volume, would it make sense to not use a slot for this replica (and/or not use streaming replication)?

I'm interested to know what people suggest in this space.

--

Don Seiler
www.seiler.us

On Thu, Dec 1, 2022 at 1:31 PM Mahesh Shetty <maheshetty20@gmail.com> wrote:

What is the replication conflict you are facing - Can you share output from  pg_stat_database_conflicts ?

The errors stopped a few hours ago so I'm not sure if this info is meaningful or not, but here it is. The "foo" database (real name changed) is obviously the app database where all the querying takes place. 

postgres=# select * from pg_stat_database_conflicts;
 datid |  datname  | confl_tablespace | confl_lock | confl_snapshot | confl_bufferpin | confl_deadlock
-------+-----------+------------------+------------+----------------+-----------------+----------------
 17156 | postgres  |                0 |          0 |              0 |               0 |              0
 13399 | template0 |                0 |          0 |              0 |               0 |              0
 16400 | template1 |                0 |          0 |              0 |               0 |              0
 16401 | foo       |                0 |          0 |           6736 |              35 |              0

(4 rows) 

--

Don Seiler
www.seiler.us

Hi Sarwar,

> On 08/12/2022 00:51 CET M Sarwar <sarwarmd02@outlook.com> wrote:
>
> I have created a user with my name and unable to verify the same using pgadmin
> or any other way. For many of you this question is a novice. I am seeking the
> help after spending sometime on this. You can point me to the document if you
> or anybody can think of anything or find it.
> Thank you!
> Sarwar

A few questions:

* What Postgres version do you use?

* What error message do you get when connecting with your new user? I assume
  that you get an error when you say "unable to verify".

* How did you create that user? Was it CREATE USER[1] or createuser via command
  line? In case of CREATE ROLE[2] you must also specify attribute LOGIN which
  effectively is the same as CREATE USER. The login privilege must also be
  enabled when creating roles via pgAdmin.[3]

* What does your pg_hba.conf[4] look like? You must configure the right
  databases, usernames, and client addresses before connecting.

[1] https://www.postgresql.org/docs/current/sql-createuser.html
[2] https://www.postgresql.org/docs/current/sql-createrole.html
[3] https://www.pgadmin.org/docs/pgadmin4/latest/role_dialog.html
[4] https://www.postgresql.org/docs/current/auth-pg-hba-conf.html

P.S. Please don't hijack threads.

--
Erik