Horton
Menu

Re: pg_waldump: support decoding of WAL inside tarfile

Tom Lane <tgl@sss.pgh.pa.us>

View thread
From: Tom Lane <tgl@sss.pgh.pa.us>
To: Andrew Dunstan <andrew@dunslane.net>
Cc: Michael Paquier <michael@paquier.xyz>, Amul Sul <sulamul@gmail.com>, Zsolt Parragi <zsolt.parragi@percona.com>, Robert Haas <robertmhaas@gmail.com>, Chao Li <li.evan.chao@gmail.com>, Jakub Wartak <jakub.wartak@enterprisedb.com>, PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>
Date: 2026-03-23T03:02:20Z
Lists: pgsql-hackers

Attachments

I wrote:
> Everything's pushed, although it looks like it'll be a few hours
> before we see results from batta or hachi.

batta just went green, so I'm feeling optimistic that we've correctly
diagnosed and fixed the problem.

Meanwhile, I noticed one more thing that could use fixing: while
astreamer_zstd.c is careful to use a reasonably-sized output
bbs_buffer (about 256kB looks like), the decompressors in
astreamer_gzip.c and astreamer_lz4.c just leave bbs_buffer's size
at the default 1kB.  This means we're pushing decompressed data
to the next astreamer in very small units in those cases.

I suspect strongly that this difference in buffer sizes has a lot
to do with the observed results that only zstd failed on batta/hachi
while only the other two failed in the test cases that I've been
able to reproduce here.  I'm too lazy to try to run that to ground
though, especially since it's mostly moot now.  But I think that
on performance grounds, we ought to use a reasonable buffer size
for all three decompressors.

I also noticed that astreamer_zstd.c was using ZSTD_DStreamOutSize()
as the target buffer size for both compression and decompression,
although the libzstd API provides a separate function
ZSTD_CStreamOutSize() for the compression case.  Those two functions
produce the same result (256K) on my machine, so this seems to be just
a cosmetic issue, but it still seems pretty tin-eared to not use the
functions as specified.

Proposed patch attached.  There might be an argument for using some
other size than 256K for the other two decompressors, but my
inclination is to try to make all three use roughly the same block
size.  (See also 66ec01dc4.)

			regards, tom lane

Commits

Same data as JSON: GET /api/v1/messages/:b64id/commits the thread's linked commits as JSON, with link sources. API reference →

  1. Use size_t instead of Size in pg_waldump

  2. More tar portability adjustments.

  3. Further harden tests that might use not-so-compatible tar versions.

  4. Harden astreamer tar parsing logic against archives it can't handle.

  5. Fix pg_waldump/t/001_basic.pl with BSD tar on ZFS.

  6. Remove a low-value, high-risk optimization in pg_waldump.

  7. Fix misuse of simplehash.h hash operations in pg_waldump.

  8. Fix file descriptor leakages in pg_waldump.

  9. Fix poorly-sized buffers in astreamer compression modules.

  10. Remove read_archive_file()'s "count" parameter.

  11. Report detailed errors from XLogFindNextRecord() failures.

  12. Fix assorted bugs in archive_waldump.c.

  13. Remove nonfunctional tar file trailer size check.

  14. Fix finalization of decompressor astreamers.

  15. Move tar detection and compression logic to common.

  16. pg_verifybackup: Enable WAL parsing for tar-format backups

  17. pg_waldump: Add support for reading WAL from tar archives

  18. pg_waldump: Preparatory refactoring for tar archive WAL decoding.

  19. pg_waldump: Remove file-level global WalSegSz.

  20. pg_verifybackup: Verify tar-format backups.