Re: [HACKERS] intermittent failures in Cygwin from select_parallel tests

On Thu, Jun 15, 2017 at 5:16 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Robert Haas <robertmhaas@gmail.com> writes:
>> On Thu, Jun 15, 2017 at 5:06 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>>> ... nodeGather cannot deem the query done until it's seen EOF on
>>> each tuple queue, which it cannot see until each worker has attached
>>> to and then detached from the associated shm_mq.
>
>> Oh.  That's sad.  It definitely has to wait for any tuple queues that
>> have been attached to be detached, but it would be better if it didn't
>> have to wait for processes that haven't even attached yet.
>
> Hm.  We assume they attach before they start taking any of the query
> work?  Seems reasonable, and this would give us some chance of recovering
> from worker fork failure.

Yeah, something like that.  I'm not sure exactly how to implement it,
though.  I think I intended for it to work that way all along, but the
code's not there.

-- 
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company