BUG #17565: Extended query protocol incorrectly sets TBLOCK_STARTED even though multiple queries can be executed

From: PG Bug reporting form <noreply(at)postgresql(dot)org>
To: pgsql-bugs(at)lists(dot)postgresql(dot)org
Cc: matt(dot)jibson(at)gmail(dot)com
Subject: BUG #17565: Extended query protocol incorrectly sets TBLOCK_STARTED even though multiple queries can be executed
Date: 2022-08-01 20:12:00
Message-ID: 17565-156f3bd8f262150f@postgresql.org
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-bugs

The following bug has been logged on the website:

Bug reference: 17565
Logged by: Matt Jibson
Email address: matt(dot)jibson(at)gmail(dot)com
PostgreSQL version: 14.4
Operating system: linux
Description:

The extended query protocol is most often used with the 4 messages in the
sequence: Parse, Bind, Execute, Sync. Parse, Bind, and Execute all call
start_xact_command, which will change blockState to TBLOCK_STARTED. However
it is within the protocol spec to instead do: Parse, Bind, Execute, Parse,
Bind, Execute, Sync. That is: issue two query executions and then a Sync.
The Npgsql driver does exactly this at its startup.

The doc comment for TBLOCK_STARTED says "running single-query transaction",
but this is a case where a driver can escape that restriction which is
assumed by the rest of the code. For example, there are various things
differentiate between STARTED and INPROGRESS or IMPLICIT_INPROGRESS:
PreventInTransactionBlock, IsInTransactionBlock. Some statements (VACUUM,
DISCARD ALL) use those functions to prevent their execution in a multi
statement context, but those can be escaped using the extended protocol.

The protocol docs have an excellent section on how to handle multiple
statements in a simple query. Assuming that this is indeed a bug and is
fixed somehow, the extended protocol may need its docs changed to similarly
describe what happens in this case.

Responses

Browse pgsql-bugs by date

  From Date Subject
Next Message Tom Lane 2022-08-01 22:17:11 Re: BUG #17564: Planner bug in combination of generate_series(), unnest() and ORDER BY
Previous Message Tom Lane 2022-08-01 17:22:35 Re: Statistics updates is delayed when using `commit and chain`