Re: adding status for COPY progress report
От | Zhihong Yu |
---|---|
Тема | Re: adding status for COPY progress report |
Дата | |
Msg-id | CALNJ-vQbYuOR-NJOYP8OH4+Bn5qd_QnueoTYZVqq1TFrkGpu9A@mail.gmail.com обсуждение исходный текст |
Ответ на | Re: adding status for COPY progress report (Matthias van de Meent <boekewurm+postgres@gmail.com>) |
Ответы |
Re: adding status for COPY progress report
|
Список | pgsql-hackers |
On Tue, May 24, 2022 at 12:37 PM Matthias van de Meent <boekewurm+postgres@gmail.com> wrote:
On Tue, 24 May 2022 at 19:13, Zhihong Yu <zyu@yugabyte.com> wrote:
>
> Hi,
> Please see attached for enhancement to COPY command progress.
>
> The added status column would allow users to get the status of the most recent COPY command.
I fail to see the merit of retaining completed progress reporting
commands in their views after completion, other than making the
behaviour of the pg_stat_progress-views much more complicated and
adding overhead in places where we want the system to have as little
overhead as possible.
Trying to get the status of a COPY command after it finished on a
different connection seems weird, as that other connection is likely
to have already disconnected / started another task. To be certain
that a backend can see the return status of the COPY command, you'd
have to be certain that the connection doesn't run any other
_progress-able commands in the following seconds / minutes, which
implies control over the connection, which means you already have
access to the resulting status of your COPY command.
Regarding the patch: I really do not like that this leaks entries into
all _progress views: I get garbage data from e.g. the _create_index
and _copy views when VACUUM is running, etc, because you removed the
filter on cmdtype.
Also, the added fields in CopyToStateData / CopyFromStateData seem
useless when a pgstat_progress_update_param in the right place should
suffice.
Kind regards,
Matthias van de Meent
Hi,
For #2 above, can you let me know where the pgstat_progress_update_param() call(s) should be added ?
In my patch, pgstat_progress_update_param() is called from error callback and EndCopy().
For #1, if I use param18 (which is not used by other views), would that be better ?
Thanks
В списке pgsql-hackers по дате отправления: