[HACKERS] Assertion failure when the non-exclusive pg_stop_backup aborted.

Hi,

I got an assert failure when executing pg_terminate_backend to the
process that waiting for WAL to be archived in non-exclusive backup
mode.

TRAP: FailedAssertion("!(XLogCtl->Insert.nonExclusiveBackups > 0)",
File: "xlog.c", Line: 11123)

The one reproducing procedure is,
1. Start postgres with enabling the WAL archive
2. Execute pg_start_backup()
3. Revoke the access privileges of archive directory. (e.g., chown
root:root /path/to/archive)
4. Execute pg_stop_backup() and hangs
5. Execute pg_terminate_backend() to the process that is waiting for
WAL to be archived.
Got the assertion failure.

Perhaps we can reproduce it using pg_basebackup as well.

I think the cause of this is that do_pg_abort_backup can be called
even after we decremented XLogCtl->Insert.nonExclusiveBackups in
do_pg_stop_backup(). That is, do_pg_abort_backup can be called with
XLogCtl->Insert.nonExclusiveBackups = 0 when the transaction aborts
after processed the nonExclusiveBackups (e.g, during waiting for WAL
to be archived)
The bug can happen in PostgreSQL 9.1 or higher that non-exclusive
backup has been introduced, so we should back-patch to the all
supported versions.

I changed do_pg_abort_backup() so that it decrements
nonExclusiveBackups only if > 0. Feedback is very welcome.

Regards,

--
Masahiko Sawada
NIPPON TELEGRAPH AND TELEPHONE CORPORATION
NTT Open Source Software Center

-- 
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers