Re: Race condition in SyncRepGetSyncStandbysPriority

Masahiko Sawada <masahiko.sawada@2ndquadrant.com> writes:
> On Sat, 18 Apr 2020 at 00:31, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> +   /* Quick out if not even configured to be synchronous */
>> +   if (SyncRepConfig == NULL)
>> +       return false;

> I felt strange a bit that we do the above check in
> SyncRepGetSyncRecPtr() because SyncRepReleaseWaiters() which is the
> only caller says the following before calling it:

Notice there was such a test in SyncRepGetSyncRecPtr already --- I just
moved it to be before doing some work instead of after.

> Can we either change it to an assertion, move it to before acquiring
> SyncRepLock in SyncRepReleaseWaiters or just remove it?

I have no objection to that in principle, but it seems like it's a
change in SyncRepGetSyncRecPtr's API that is not necessary to fix
this bug.  So I'd rather leave it to happen along with the larger
API changes (getting rid of am_sync) that are proposed for v14.

            regards, tom lane