On Thu, Dec 21, 2017 at 10:18 AM, Robert Haas <robertmhaas@gmail.com> wrote:
> If I run the regression tests with force_parallel_mode=on prior to the
> parallel hash join patch, they pass. If I run them now, they fail
> inside the parallel hash join tests here:
>
> create table wide as select generate_series(1, 2) as id, rpad('',
> 320000, 'x') as t;
>
> I'm guessing that test case would have failed before, too, but we
> didn't have it. I'll analyze this further in a bit.
I think this is just a poorly-written assertion. currentCommandIdUsed
is only used to skip redundant increments of the command counter, and
CommandCounterIncrement() is categorically denied under parallelism
anyway. Therefore, it's OK for this to happen in parallel mode; we
just need to be in the leader, not the worker.
Therefore, I proposed the attached patch, which fixes the regression
test crash for me.
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company