On Wed, Nov 15, 2017 at 12:12 PM, Masahiko Sawada <sawada.mshk@gmail.com> wrote:
> I think we need to check only sessionBackupState and don't need to
> check XLogCtl->Insert.exclusiveBackupState in do_pg_abort_backup(). We
> can quickly return if sessionBackupState !=
> SESSION_BACKUP_NON_EXCLUSIVE. In your suggestion, I think we can still
> get an assertion failure when pg_stop_backup(false) waiting for
> archiving is terminated while concurrent an exclusive backup is in
> progress.
I have just gone through the thread once again, and noticed that it is
actually what I suggested upthread:
https://www.postgresql.org/message-id/CAB7nPqTm5CDrR5Y7yyfKy+PVDZ6dWS_jKG1KStaN5m95gAMTFQ@mail.gmail.com
But your v2 posted here did not do that so it is incorrect from the start:
https://www.postgresql.org/message-id/CAD21AoA+isXYL1_ZXMnk9xJhYEL5h6rxJtTovLi7fumqfmCYgg@mail.gmail.com
We both got a bit confused here. As do_pg_abort_backup() is only used
for non-exclusive backups (including those taken through the
replication protocol), going through the session lock for checks is
fine. Could you update your patch accordingly please?
--
Michael