Horton

Re: WARNING: relcache reference leak: relation "p1" not closed

Tom Lane <tgl@sss.pgh.pa.us>

View thread View on postgr.es
From: Tom Lane <tgl@sss.pgh.pa.us>
To: Amit Langote <Langote_Amit_f8@lab.ntt.co.jp>
Cc: Michael Paquier <michael.paquier@gmail.com>, Kevin Grittner <kgrittn@gmail.com>, "pgsql-hackers@postgresql.org" <pgsql-hackers@postgresql.org>, Robert Haas <robertmhaas@gmail.com>
Date: 2017-03-07T05:04:42Z
Lists: pgsql-hackers 
Amit Langote <Langote_Amit_f8@lab.ntt.co.jp> writes:
> Also, I found out that alter_table.sql mistakenly forgot to drop
> partitioned table "p1".  Patch 0002 takes care of that.

While that might or might not have been intentional, I think it's an
astoundingly bad idea to not leave any partitioned tables behind in
the final state of the regression database.  Doing so would likely
have meant that this particular bug evaded detection for much longer
than it did.  Moreover, it would mean that the pg_upgrade test would
have exactly no coverage of partitioned cases.

Therefore, there should definitely be a partitioned table, hopefully with
a less generic name than "p1", in the final regression DB state.  Whether
this particular one from alter_table.sql is a good candidate, I dunno.
But let's not drop it without adding a better-thought-out replacement.

			regards, tom lane

Commits

  1. Give partitioned table "p" in regression tests a less generic name.

  2. Fix relcache reference leak.

  3. Combine several DROP variants into generic DropStmt

  4. Reorder the asynchronous libpq calls for replication connection

  5. Don't uselessly rewrite, truncate, VACUUM, or ANALYZE partitioned tables.