On Wed, Dec 17, 2014 at 2:53 PM, Robert Haas <robertmhaas@gmail.com> wrote:
> In the meantime, I had a good chat with Heikki on IM yesterday which
> gave me some new ideas on how to fix up the transaction handling in
> here, which I am working on implementing. So hopefully I will have
> that by then.
And here is a new version. This version has some real integration
with the transaction system, along the lines of what Heikki suggested
to me, so hopefully tuple visibility calculations in a parallel worker
will now produce the right answers, though I need to integrate this
with code to actually do something-or-other in parallel in order to
really test that. There are still some problems with parallel worker
shutdown. As hard as I tried to fight it, I'm gradually resigning
myself to the fact that we're probably going to have to set things up
so that the worker waits for all of its children to die during abort
processing (and during commit processing, but that's not bothersome).
Otherwise, to take just one example, chaos potentially ensues if you
run a parallel query in a subtransaction and then roll back to a
savepoint. But this version just kills the workers and doesn't
actually wait for them to die; I'll see about fixing that, but wanted
to send this around for comments first.
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company
