| From: | yuansong <yyuansong(at)126(dot)com> |
|---|---|
| To: | pgsql-hackers(at)lists(dot)postgresql(dot)org |
| Cc: | josef(dot)simanek(at)gmail(dot)com |
| Subject: | Re: [PATCH v1] Add pg_stat_progress overview view |
| Date: | 2026-11-19 15:50:11 |
| Message-ID: | 1b33a84a.7f77.1a1f4804e77.Coremail.yyuansong@126.com |
| Views: | Whole Thread | Raw Message | Download mbox | Resend email |
| Thread: | |
| Lists: | pgsql-hackers |
Hi Josef,
I took a look at v1. Overall this looks like a good, low-risk patch -- a
small SQL-only view that reuses pg_stat_get_progress_info() with no backend
changes. I checked it against current master and found no functional
problems. A few comments:
1. The patch bumps CATALOG_VERSION_NO from 202611021 to 202611071, but it is
based on an older master (current master is already at 202611121), so the
hunk will conflict on rebase. Per the usual practice, catversion bumps
are handled by the committer at commit time rather than in submitted
patches -- repeatedly bumping it just creates avoidable rebases. Could
you please rebase and drop the catversion change? If you'd like to flag
it, a note in the commit message that a catversion bump is needed is the
recommended way to do so.
2. (Optional nit) The VALUES list ties the displayed command names to the
strings accepted by pg_stat_get_progress_info(). I verified all seven
match the pg_strcasecmp() mapping in pgstatfuncs.c, and that omitting
CLUSTER is correct since it is only a compatibility view over REPACK.
This mirrors the existing per-command progress views, so it's not a new
pattern; a short comment noting the list must stay in sync would still
help whoever adds the next command type, but I don't consider it blocking.
3. CROSS JOIN LATERAL calls pg_stat_get_progress_info() once per command
(7 times), each call walking all backends and taking a separate stats
snapshot. This is the same trade-off the existing progress views make,
so I'm not asking for a code change -- but as the wiki asks for a note on
performance impact, it may be worth mentioning in the cover email.
4. In the TAP test, progress_gate() is declared IMMUTABLE while taking an
advisory lock. I understand it has to be IMMUTABLE to be usable in the
expression index that drives the test -- worth a comment to that effect
so nobody "fixes" it later.
Nits: consider noting in the docs how to get the database dimension (join
pg_stat_activity on pid), since datname was deliberately left out.
Tests and docs look thorough. Aside from the rebase in (1), this looks
ready for committer from my side.
Best regards,
Yuansong
| From | Date | Subject | |
|---|---|---|---|
| Next Message | Alexander Korotkov | 2026-11-19 16:15:13 | Re: test: avoid redundant standby catchup in 049_wait_for_lsn |
| Previous Message | Zhou, Zhiguo | 2026-11-19 15:23:22 | Re: Increase NUM_XLOGINSERT_LOCKS |