Re: BUG #17156: pg_restore: [custom archiver] WARNING: ftell mismatch with expected position -- ftell used

On Tue, Aug 24, 2021 at 5:42 PM Kyotaro Horiguchi
<horikyota.ntt@gmail.com> wrote:
>
> At Tue, 24 Aug 2021 14:25:50 +0900, Masahiko Sawada <sawada.mshk@gmail.com> wrote in
> > On Mon, Aug 23, 2021 at 7:31 PM PG Bug reporting form
> > <noreply@postgresql.org> wrote:
> > >
> > > The following bug has been logged on the website:
> > >
> > > Bug reference:      17156
> > > Logged by:          Ashutosh Kumar
> > > Email address:      ashutosh.kumar@morningstar.com
> > > PostgreSQL version: 10.18
> > > Operating system:   RHEL
> > > Description:
> > >
> > > Getting below warning when I am restoring the backup file :
> > >
> > > Warning Message : pg_restore: [custom archiver] WARNING: ftell mismatch with
> > > expected position -- ftell used
> > >
> > > Backup was taken with below command :
> > >
> > > pg_dump -h'127.0.0.1'  -p 5432 -U (username)  -Fc  databasename >
> > > backupfile.sql
> > >
> > > I am restoring it with pg_restore when i got the above warning.
> >
> > Could you share the exact pg_restore command with the arguments you executed?
> >
> > What's is the size of the backup? Are both Postgres servers that you
> > took/restored the backup from/to are RHEL?
> >
> > Could you try to execute pg_dump without a pipe and restore it and see
> > the same error happens? That is, please execute "pg_dump -h'127.0.0.1'
> >  -p 5432 -U (username)  -Fc  databasename -f backupfile.sql".
>
> That error means when reading a seekable source, the file pointer
> ftello returns doesn't consistent with how many bytes pg_restore has
> read, or ftello returned an error.  Considering that the path is very
> ancient, the latter likely to happen.

If ftello returned an error, i.g., -1, wouldn't pg_restore die instead
of emitting the warning? As Tom mentioned[1], we have fseeko() calls
in some code paths but don't correct ctx->filePos afterward. This is
also likely to cause this warning.

FIWI I'd like to note that this warning is meant to be for debugging
purposes as the comment in the source code mentioned. Since if the
input source is seekable we rely on ftello() to get the current
position in the file the restoration should be performed even if we
got this warning. Actually, we removed manual tracking of file
position by commit a8d0732ac, which backpatched to v12.

Regards,

[1] https://www.postgresql.org/message-id/3172545.1594845659%40sss.pgh.pa.us

--
Masahiko Sawada
EDB:  https://www.enterprisedb.com/