Re: Testing autovacuum wraparound (including failsafe) - Mailing list pgsql-hackers

From Anastasia Lubennikova
Subject Re: Testing autovacuum wraparound (including failsafe)
Date
Msg-id CAP4vRV5gEHFLB7NwOE6_dyHAeVfkvqF8Z_g5GaCQZNgBAE0Frw@mail.gmail.com
Whole thread Raw
In response to Testing autovacuum wraparound (including failsafe)  (Andres Freund <andres@anarazel.de>)
Responses Re: Testing autovacuum wraparound (including failsafe)
List pgsql-hackers

On Thu, Jun 10, 2021 at 10:52 AM Andres Freund <andres@anarazel.de> wrote:

I started to write a test for $Subject, which I think we sorely need.

Currently my approach is to:
- start a cluster, create a few tables with test data
- acquire SHARE UPDATE EXCLUSIVE in a prepared transaction, to prevent
  autovacuum from doing anything
- cause dead tuples to exist
- restart
- run pg_resetwal -x 2000027648
- do things like acquiring pins on pages that block vacuum from progressing
- commit prepared transaction
- wait for template0, template1 datfrozenxid to increase
- wait for relfrozenxid for most relations in postgres to increase
- release buffer pin
- wait for postgres datfrozenxid to increase


Cool. Thank you for working on that!
Could you please share a WIP patch for the $subj? I'd be happy to help with it.

So far so good. But I've encountered a few things that stand in the way of
enabling such a test by default:

1) During startup StartupSUBTRANS() zeroes out all pages between
   oldestActiveXID and nextXid. That takes 8s on my workstation, but only
   because I have plenty memory - pg_subtrans ends up 14GB as I currently do
   the test. Clearly not something we could do on the BF.
 ....
3) pg_resetwal -x requires to carefully choose an xid: It needs to be the
   first xid on a clog page. It's not hard to determine which xids are but it
   depends on BLCKSZ and a few constants in clog.c. I've for now hardcoded a
   value appropriate for 8KB, but ...

Maybe we can add new pg_resetwal option?  Something like pg_resetwal --xid-near-wraparound, which will ask pg_resetwal to calculate exact xid value using values from pg_control and clog macros?
I think it might come in handy for manual testing too.


I have 2 1/2 ideas about addressing 1);

- We could exposing functionality to do advance nextXid to a future value at
  runtime, without filling in clog/subtrans pages. Would probably have to live
  in varsup.c and be exposed via regress.so or such?

This option looks scary to me. Several functions rely on the fact that StartupSUBTRANS() have zeroed pages.
And if we will do it conditional just for tests, it means that we won't test the real code path.

- The only reason StartupSUBTRANS() does that work is because of the prepared
  transaction holding back oldestActiveXID. That transaction in turn exists to
  prevent autovacuum from doing anything before we do test setup
  steps.
 

  Perhaps it'd be sufficient to set autovacuum_naptime really high initially,
  perform the test setup, set naptime to something lower, reload config. But
  I'm worried that might not be reliable: If something ends up allocating an
  xid we'd potentially reach the path in GetNewTransaction() that wakes up the
  launcher?  But probably there wouldn't be anything doing so?


  Another aspect that might not make this a good choice is that it actually
  seems relevant to be able to test cases where there are very old still
  running transactions...

Maybe this exact scenario can be covered with a separate long-running test, not included in buildfarm test suite?

--
Best regards,
Lubennikova Anastasia

pgsql-hackers by date:

Previous
From: Andrew Dunstan
Date:
Subject: Re: BF assertion failure on mandrill in walsender, v13
Next
From: Amit Kapila
Date:
Subject: Re: Decoding speculative insert with toast leaks memory