On Wed, Aug 8, 2018 at 10:08 PM, Peter Geoghegan <p...@bowt.ie> wrote: >> Hmmm ... maybe we should temporarily stick in an elog(LOG) showing whether >> a parallel build happened or not, so that we can check the buildfarm logs >> next time we see that failure? > > I can do that tomorrow. Of course, it might be easier to just push the > pending fix for the mapped relation bug, and see if that makes the > issue go away. That would take longer, but it's also the only thing > that's likely to be definitive anyway.
I'm going to push my fix for the relmapper.c parallel CREATE INDEX bug tomorrow, and see what happens. I have a hard time imagining how there could be a parallel index build on pg_class, now that Andrew has indicated there is no reason to think that "min_parallel_table_scan_size = 0" could slip in. -- Peter Geoghegan