Horton

Re: BUG #17434: CREATE/DROP DATABASE can be executed in the same transaction with other commands

Peter Eisentraut <peter.eisentraut@enterprisedb.com>

View thread View on postgr.es
From: Peter Eisentraut <peter.eisentraut@enterprisedb.com>
To: Yugo NAGATA <nagata@sraoss.co.jp>, Tom Lane <tgl@sss.pgh.pa.us>, pgsql-hackers@postgresql.org
Cc: "David G. Johnston" <david.g.johnston@gmail.com>, Bruce Momjian <bruce@momjian.us>, PostgreSQL mailing lists <pgsql-bugs@lists.postgresql.org>
Date: 2022-11-09T16:24:43Z
Lists: pgsql-bugs, pgsql-hackers 
On 08.08.22 17:21, Yugo NAGATA wrote:
>>> In fact, the result of IsInTransactionBlock does not make senses at
>>> all in pipe-line mode regardless to the fix. ANALYZE could commit all
>>> previous commands in pipelining, and this may not be user expected
>>> behaviour.
>> This seems pretty much isomorphic to the fact that CREATE DATABASE
>> will commit preceding steps in the pipeline.
> I am not sure if we can think CREATE DATABASE case and ANLALYZE case
> similarly. First, CREATE DATABASE is one of the commands that cannot be
> executed inside a transaction block, but ANALYZE can be. So, users would
> not be able to know ANALYZE in a pipeline causes a commit from the
> documentation. Second, ANALYZE issues a commit internally in an early
> stage not only after it finished successfully. For example, even if
> ANALYZE is failing because a not-existing column name is specified, it
> issues a commit before checking the column name. This makes more hard
> to know which statements will be committed and which statements not
> committed in a pipeline. Also, as you know, there are other commands
> that issue internal commits.

This has broken the following use:

parse: create temporary table t1 (a int) on commit drop
bind
execute
parse: analyze t1
bind
execute
parse: select * from t1
bind
execute
sync

I think the behavior of IsInTransactionBlock() needs to be further 
refined to support this.  If we are worried about external callers, 
maybe we need to provide a separate version.  AFAICT, all the callers of 
IsInTransactionBlock() over time have been in vacuum/analyze-related 
code, so perhaps in master we should just move it there.

Commits

  1. Rethink handling of [Prevent|Is]InTransactionBlock in pipeline mode.

  2. Doc: add comments about PreventInTransactionBlock/IsInTransactionBlock.

  3. Force immediate commit after CREATE DATABASE etc in extended protocol.