Re: pg_waldump: support decoding of WAL inside tarfile - Mailing list pgsql-hackers

From Sami Imseih
Subject Re: pg_waldump: support decoding of WAL inside tarfile
Date
Msg-id CAA5RZ0tt89MgNi4-0F4onH+-TFSsysFjMM-tBc6aXbuQv5xBXw@mail.gmail.com
Whole thread Raw
In response to Re: pg_waldump: support decoding of WAL inside tarfile  (Tom Lane <tgl@sss.pgh.pa.us>)
Responses Re: pg_waldump: support decoding of WAL inside tarfile
List pgsql-hackers
> >> So I think we need something like the attached, in addition
> >> to what I sent before.  This just makes astreamer_tar.c use
> >> the isValidTarHeader function that pg_dump already had.
>
> > LGTM.
>
> Pushed, thanks for reviewing!  In the event I decided to back-patch to
> v18, where these fixes could protect pg_verifybackup against tar files
> it can't handle.

Hi,

I just encountered a regression test failure for pg_waldump due to ebba64c08d9.

````
――――――――――――――――――――――――――――――――――――――――――――――――――――― ✀
――――――――――――――――――――――――――――――――――――――――――――――――――――――
Listing only the last 100 lines from a long log.
#   at /local/home/simseih/pgdev/installations/worktrees/dev/src/bin/pg_waldump/t/001_basic.pl
line 432.
#          got: 'pg_waldump: error: could not find WAL in archive
"pg_wal.tar.gz"
# '
#     expected: ''
```

and regress_log_001_basic shows this:

```
# Running: /usr/bin/tar --format=ustar -cf /tmp/ja26rXZOnb/pg_wal.tar
archive_status 000000010000000000000002 000000010000000000000001
summaries 000000010000000000000003
[22:25:00.525](0.008s) not ok 101
[22:25:00.525](0.000s) #   Failed test at
/local/home/simseih/pgdev/installations/worktrees/dev/src/bin/pg_waldump/t/001_basic.pl
line 350.
[22:25:00.525](0.000s) # ---------- command failed ----------
[22:25:00.526](0.000s) # /usr/bin/tar --format=ustar -cf
/tmp/ja26rXZOnb/pg_wal.tar archive_status 000000010000000000000002
000000010000000000000001 summaries 000000010000000000000003
[22:25:00.526](0.000s) # -------------- stderr --------------
[22:25:00.526](0.000s) # /usr/bin/tar: value 10012663 out of uid_t
range 0..2097151
```

The --format=ustar has a limit of 2^21 (2097151) for UID/GID [1]
and on my machine the UID is 10012663.

So I found that one way to deal with this is to run the tar command with
--owner=0 --group=0. As far as I can tell, the owner and group IDs don't
matter for these tests, so maybe that is OK.

@@ -1333,6 +1333,10 @@ sub tar_portability_options
                == 0)
        {
                push(@tar_p_flags, "--format=ustar");
+               # ustar format supports UIDs only up to 2^21 (2097151).
+               # Override owner/group to avoid failures on systems where
+               # the running user's UID/GID exceeds that limit.
+               push(@tar_p_flags, "--owner=0", "--group=0");
        }

While this fixes the test, I am now not sure what the broader implications are
for --format=ustar for pg_waldump in the broader discussion?

[1] [https://www.gnu.org/software/tar/manual/html_section/Formats.html]

--
Sami Imseih
Amazon Web Services (AWS)



pgsql-hackers by date:

Previous
From: Masahiko Sawada
Date:
Subject: Re: POC: Parallel processing of indexes in autovacuum
Next
From: Thomas Munro
Date:
Subject: Re: pg_waldump: support decoding of WAL inside tarfile