MIME-Version: 1.0
References: <CAAJ_b94bqdWN3h2J-PzzzQ2Npbwct5ZQHggn_QoYGhC2rn-=WQ@mail.gmail.com>
 <CAAJ_b97PQjE4kFD8Qk6UvtLrfPMixw1nxBz0OP5Z2WB2B-uMxQ@mail.gmail.com>
 <CAAJ_b97JA8ehy_UDddrnGwDt9HG5NmJq8ATtmeMqo7YD-=tLyQ@mail.gmail.com>
 <CA+TgmoZjhWDG_AR1i+L1yss-wbuWvxrdRwSdVUUUnVPrJV2CnQ@mail.gmail.com>
 <CAAJ_b94Uh+b41LQG45bZFK+i62EVvv972LiGWWWuR64=-64rTQ@mail.gmail.com>
 <CA+TgmobF5c7ZcZHdEhqwNxGDZzWG2bDtpRaDtoVELWX_VHs_1A@mail.gmail.com>
 <CAAJ_b94gK1np8d1h-2c1YoCccGXr4zspTa-FC7X_bfXZNz=-DA@mail.gmail.com>
 <CA+TgmoayDY5b+bP1vRRN7A3xOP-=+tK13B2C1g-Xm1j4WTrT9Q@mail.gmail.com>
 <CAAJ_b97JAF+Zuoh2FBO79hVwLeaBPwsbXw-fY+313a7LfRQ-Bg@mail.gmail.com>
 <CA+Tgmoardk4VuthHc23vov+AVkhq7eT0mFUs-2ctAnP1uiTaog@mail.gmail.com>
 <CAAJ_b959x5VjmLJFmN78r_QohQuuj=fde11VbbAOHn5TzgEzng@mail.gmail.com>
 <CAAJ_b97_N+0sipFyq80n0jX-nKcjcQEMOSTVg8DsqkHR8dW_Sw@mail.gmail.com>
 <CAKZiRmyDk5KqovS9Ez3iFHd+p-TChSt2QTtWkwJ5Ya-+4gg21g@mail.gmail.com>
 <CAAJ_b956a+e8-HNEpeJ60ByFv7XJRqECPu3B0dozv0ChMRTPbQ@mail.gmail.com>
 <CAAJ_b94f6sUDWiZY90O-t7SWWeSK0nMWT7AsydKkpAL90m0oMQ@mail.gmail.com>
 <731ADE6F-01C5-4996-BAEE-5851DFC3F502@gmail.com> <CAAJ_b94SEcBVJcsp0Y1-YvLqZWBHPQH4FhRzSJfaH_ah_eL_FQ@mail.gmail.com>
 <CAAJ_b97nn9denP2SAjuXyjwbd3is-VnZVSkiRMJ-5YNiKfO9MA@mail.gmail.com>
 <CA+Tgmob_DB9QHDOxnP7a5Y0yJdeGqY8YNi+uK_811y7cN4mxYA@mail.gmail.com>
 <CAAJ_b97VUiP-DbLNe-ddq64J_RiB4ZcPgAjHkJH-0dbzgjR++A@mail.gmail.com>
 <CA+TgmoYMtcZBaqy9r59eDapaDy3WOdepkFFURu9MV-x-kxEbKg@mail.gmail.com>
 <CAAJ_b95FOeW38gw-3BLmpdnTWHFimopTvf=eTObYUbTOC0x8qg@mail.gmail.com>
 <CA+Tgmob=3POOO8st-v-fCjKCKREQ=+gs5_PBQnoFeNBdERfuEg@mail.gmail.com>
 <CAAJ_b94=gtCeUKkGPUmPj_2SwHV+PiXQ8Mx-1RqfYj3pP3OwpA@mail.gmail.com> <CA+TgmoYuC-92tMxZHn-KY=ExF-YkHkd4yOfiMhi9upqk3M1DuA@mail.gmail.com>
In-Reply-To: <CA+TgmoYuC-92tMxZHn-KY=ExF-YkHkd4yOfiMhi9upqk3M1DuA@mail.gmail.com>
From: Amul Sul <sulamul@gmail.com>
Date: Wed, 4 Feb 2026 18:39:10 +0530
Message-ID: <CAAJ_b974evwS9+tmKVEV+ctEp1scRnLAMrmx9_pzLFHsPtr0hA@mail.gmail.com>
Subject: Re: pg_waldump: support decoding of WAL inside tarfile
To: Robert Haas <robertmhaas@gmail.com>
Cc: Chao Li <li.evan.chao@gmail.com>, Jakub Wartak <jakub.wartak@enterprisedb.com>, 
	PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Archived-At: <https://www.postgresql.org/message-id/CAAJ_b974evwS9%2BtmKVEV%2BctEp1scRnLAMrmx9_pzLFHsPtr0hA%40mail.gmail.com>
Precedence: bulk

On Wed, Jan 28, 2026 at 2:41=E2=80=AFAM Robert Haas <robertmhaas@gmail.com>=
 wrote:
>
> On Tue, Jan 27, 2026 at 7:07=E2=80=AFAM Amul Sul <sulamul@gmail.com> wrot=
e:
> > In the attached version, I am using the WAL segment name as the hash
> > key, which is much more straightforward. I have rewritten
> > read_archive_wal_page(), and it looks much cleaner than before. The
> > logic to discard irrelevant WAL files is still within
> > get_archive_wal_entry. I added an explanation for setting cur_wal to
> > NULL, which is now handled in the separate function I mentioned
> > previously.
> >
> > Kindly have a look at the attached version; let me know if you are
> > still not happy with the current approach for filtering/discarding
> > irrelevant WAL segments. It isn't much different from the previous
> > version, but I have tried to keep it in a separate routine for better
> > code readability, with comments to make it easier to understand. I
> > also added a comment for ArchivedWALFile.
>
> I feel like the division of labor between get_archive_wal_entry() and
> read_archive_wal_page() is odd. I noticed this in the last version,
> too, and it still seems to be the case. get_archive_wal_entry() first
> calls ArchivedWAL_lookup(). If that finds an entry, it just returns.
> If it doesn't, it loops until an entry for the requested file shows up
> and then returns it. Then control returns to read_archive_wal_page()
> which loops some more until we have all the data we need for the
> requested file. But it seems odd to me to have two separate loops
> here. I think that the first loop is going to call read_archive_file()
> until we find the beginning of the file that we care about and then
> the second one is going to call read_archive_file() some more until we
> have read enough of it to satisfy the request. It feels odd to me to
> do it that way, as if we told somebody to first wait until 9 o'clock
> and then wait another 30 minutes, instead of just telling them to wait
> until 9:30. I realize it's not quite the same thing, because apart
> from calling read_archive_file(), the two loops do different things,
> but I still think it looks odd.
>
> + /*
> + * Ignore if the timeline is different or the current segment is not
> + * the desired one.
> + */
> + XLogFromFileName(entry->fname, &curSegTimeline, &curSegNo, WalSegSz);
> + if (privateInfo->timeline !=3D curSegTimeline ||
> + privateInfo->startSegNo > curSegNo ||
> + privateInfo->endSegNo < curSegNo ||
> + segno > curSegNo)
> + {
> + free_archive_wal_entry(entry->fname, privateInfo);
> + continue;
> + }
>
> The comment doesn't match the code. If it did, the test would be
> (privateInfo->timeline !=3D curSegTimeline || segno !=3D curSegno). But
> instead the segno test is > rather than !=3D, and the checks against
> startSegNo and endSegNo aren't explained at all. I think I understand
> why the segno test uses > rather than !=3D, but it's the point of the
> comment to explain things like that, rather than leaving the reader to
> guess. And I don't know why we also need to test startSegNo and
> endSegNo.
>
> I also wonder what the point is of doing XLogFromFileName() on the
> fname provided by the caller and then again on entry->fname. Couldn't
> you just compare the strings?
>
> Again, the division of labor is really odd here. It's the job of
> astreamer_waldump_content() to skip things that aren't WAL files at
> all, but it's the job of get_archive_wal_entry() to skip things that
> are WAL files but not the one we want. I disagree with putting those
> checks in completely separate parts of the code.
>

Keeping the timeline and segment start-end range checks inside the
archive streamer creates a circular dependency that cannot be resolved
without a 'dirty hack'. We must read the first available WAL file page
to determine the wal_segment_size before it can calculate the target
segment range. Moving the checks inside the streamer would make it
impossible to process that initial file, as the necessary filtering
parameters --  would still be unknown which would need to be skipped
for the first read somehow. What if later we realized that the first
WAL file which was allowed to be streamed by skipping that check is
irrelevant and doesn't fall under the start-end segment range?

> +static bool
> +member_is_wal_file(astreamer_waldump *mystreamer, astreamer_member *memb=
er,
> +    char **fname)
> +{
> + int pathlen;
> + char    *filename;
> +
> + /* We are only interested in normal files. */
> + if (member->is_directory || member->is_link)
> + return false;
> +
> + pathlen =3D strlen(member->pathname);
> + if (pathlen < XLOG_FNAME_LEN)
> + return false;
> +
> + /* WAL files from the top-level or pg_wal directory will be decoded */
> + if (pathlen > XLOG_FNAME_LEN &&
> + strncmp(member->pathname, XLOGDIR, strlen(XLOGDIR)) !=3D 0)
> + return false;
> +
> + /* WAL file could be with full path */
> + filename =3D member->pathname + (pathlen - XLOG_FNAME_LEN);
> + if (!IsXLogFileName(filename))
> + return false;
> +
> + *fname =3D pnstrdup(filename, XLOG_FNAME_LEN);
> +
> + return true;
> +}
>
> I don't think this code is fully correct. Note that pg_verifybackup's
> member_verify_header() prepends "./" and then applies
> canonicalize_path(). I think the same considerations apply here.
> Consider:
>

Understood, will fix that.

Regards,
Amul