Обсуждение: Max tables in a cluster?

Hi admins,

Is there a limit to number of tables that a cluster can support?

At the moment, one of my clusters has about 250k user tables (plus 150k toast tables) across a hundred schemas within a single database.

My application team wants to double the number of schemas (and with it the tables)

I am looking for both pros and cons of doing this and what tuning can help this cluster perform acceptably.

My questions are

1. Is there a limit to number of tables a cluster can support effectively?

2. My guess is autovacuum would have to be tuned. Current gucs for this are

max_autovacuum_workers = 6

maintenance_work_mem = 1GB

autovacuum_vacuum_scale_factor = 0.2

autovacuum_vacuum_cost_delay = 20ms

Are there any potential problems that I need to think about ?

My machine specs are as below.

RAM 256GB

32-core CPU

7TB SSD

thanks

karthik

On Wed, Apr 05, 2023 at 12:37:41PM +0530, Karthik Krishnakumar wrote:
> Is there a limit to number of tables that a cluster can support? 

Well, yes.

For starters: https://www.postgresql.org/docs/current/limits.html

But more important is: with large number of tables certain operations
become VERY slow. Specifically getting schema dump. Which is used, for
example, by pg_upgrade.

In my experience hitting 1 million rows in pg_class made it
realistically impossible to use pg_dump. But there have been
many improvements to pg_dump over the years
(https://why-upgrade.depesz.com/show?from=9.5&to=15.2&keywords=pg_dump)
to it's possible that the problem is gone now.

Try taking `pg_dump -s` of the db, and see how long it takes...

Best regards,

depesz