Horton

Re: [BUG] temporary file usage report with extended protocol and unnamed portals

Frédéric Yhuel <frederic.yhuel@dalibo.com>

View thread View on postgr.es
From: Frédéric Yhuel <frederic.yhuel@dalibo.com>
To: Sami Imseih <samimseih@gmail.com>
Cc: Tom Lane <tgl@sss.pgh.pa.us>, PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>, Benoit Lobréau <benoit.lobreau@dalibo.com>, Guillaume Lelarge <guillaume.lelarge@dalibo.com>, Pierrick Chovelon <pierrick.chovelon@dalibo.com>
Date: 2025-04-23T09:27:05Z
Lists: pgsql-hackers 

On 4/23/25 09:41, Frédéric Yhuel wrote:
> 
> 
> On 4/22/25 19:37, Sami Imseih wrote:
>> the patch relies on looking up queryDesc->sourceText inside DropPortal,
>> which Tom raised concerns about earlier in the thread [0]
> 
> Yes, I think I had misunderstood what Tom said. Thank you for pointing 
> that out.
> 
> However, is it really unsafe?
> 
> In exec_bind_message, the portal's query string comes from a duplicate 
> of the original string (see CreateCachedPlan). So we are safe in this case.
> 
> In exec_simple_query, the portal is dropped towards the end of this 
> function, so we are safe here too.
> 
> Am I missing something?

Note: the patch doesn't work well with server-side prepared statements: 
the PREPARE query is blamed instead of the EXECUTE one. But this is 
maybe something that can be fixed easily.

Commits

  1. Drop unnamed portal immediately after execution to completion

  2. Override log_error_verbosity to "default" in test 009_log_temp_files

  3. Add tests for logging of temporary file removal and statement