public inbox for [email protected]  

help / color / mirror / Atom feed

From: Fujii Masao <[email protected]>
To: Masahiko Sawada <[email protected]>
Cc: Pgsql Hackers <[email protected]>
Subject: Re: Enhance file_fdw to report processed and skipped tuples in COPY progress
Date: Fri, 11 Oct 2024 10:53:10 +0900
Message-ID: <[email protected]> (raw)
In-Reply-To: <CAD21AoDDJiJRZapB+mbyvetPp_jmYh3bbhu-ntn+911cmWdiyA@mail.gmail.com>
References: <[email protected]>
	<CAD21AoDDJiJRZapB+mbyvetPp_jmYh3bbhu-ntn+911cmWdiyA@mail.gmail.com>



On 2024/10/04 2:12, Masahiko Sawada wrote:
> Hi,
> 
> On Thu, Oct 3, 2024 at 2:23 AM Fujii Masao <[email protected]> wrote:
>>
>> Hi,
>>
>> Currently, file_fdw updates several columns in the pg_stat_progress_copy view,
>> like relid and bytes_processed, but it doesn't track tuples_processed or
>> tuples_skipped. Monitoring these would be particularly useful when handling
>> large data sets via file_fdw, as it helps track the progress of scan.
>>
>> The attached patch updates file_fdw to add support for reporting
>> the number of tuples processed and skipped (due to on_error = 'ignore')
>> in the pg_stat_progress_copy view. What are your thoughts?
> 
> While the patch works fine and looks good to me, in the first place,
> it seems to me that the fact that file_fdw uses the COPY progress
> itself doesn't work properly. For example, unlike COPY command,
> queries could have multiple scans on one or more flie_fdw foreign
> tables when joining tables. I found the discussion for that[1]: there
> was a proposal of disabling COPY progress for file_fdw but the votes
> are split. I think it would be better to consider if we really want to
> support COPY progress for file_fdw before supporting more progress
> information.

Yes, you're right. We need to address how to handle multiple commands
that trigger progress reporting when executed concurrently, at first.

The current progress reporting mechanism assumes only one command
triggering progress is running at a time, as each backend has just
one memory area for progress reporting. If multiple commands run simultaneously,
the progress data would be incorrect. As you mentioned, this could happen
when querying multiple file_fdw foreign tables, where multiple COPY commands
could execute concurrently.

However, this issue already exists without the proposed patch.
Since file_fdw already reports progress partially, querying multiple
file_fdw tables can lead to inaccurate or confusing progress reports.
You can even observe this when analyzing a file_fdw table and also
when copying to the table with a trigger that executes progress-reporting
commands.

So, I don’t think this issue should block the proposed patch.
In fact, progress reporting is already flawed in these scenarios,
regardless of whether the patch is applied.

On the other hand, in many cases where a single file_fdw table is scanned,
COPY progress reporting works correctly for file_fdw and is useful.
Therefore, I believe it's still worth improving file_fdw’s progress reporting.

To prevent misleading reports when multiple commands are run concurrently,
just idea, we could consider displaying NULL columns in the progress report
if this situation is detected, as a separate patch.

Regards,

-- 
Fujii Masao
Advanced Computing Technology Center
Research and Development Headquarters
NTT DATA CORPORATION

view thread (4+ messages)  latest in thread

reply

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Reply to all the recipients using the --to and --cc options:
  reply via email

  To: [email protected]
  Cc: [email protected], [email protected]
  Subject: Re: Enhance file_fdw to report processed and skipped tuples in COPY progress
  In-Reply-To: <[email protected]>

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

This inbox is served by agora; see mirroring instructions
for how to clone and mirror all data and code used for this inbox