On Sat, Jan 27, 2018 at 10:45 PM, Andres Freund <andres@anarazel.de> wrote:
> Hi Peter (and others who mucked around with related code),
>
> While testing another patch I found that cancelling a parallel query on
> master segfaults the leader in an interesting manner:
...
> It's clearly not OK to recurse back into the executing query after a
> failure, even less so if resources like the underlying DSM segment have
> already been freed.
>
> I suspect this is largely the fault of
>
> commit 8561e4840c81f7e345be2df170839846814fa004
> Author: Peter Eisentraut <peter_e@gmx.net>
> Date: 2018-01-22 08:30:16 -0500
>
> Transaction control in PL procedures
Yeah, I think this must be the fault of that commit, and I think your
diagnosis is accurate.
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company