Re: accounting for memory used for BufFile during hash joins

On Mon, Nov 25, 2019 at 05:33:35PM +0900, Michael Paquier wrote:
>On Tue, Sep 10, 2019 at 03:47:51PM +0200, Tomas Vondra wrote:
>> My feeling is that we should get the BNLJ committed first, and then maybe
>> use some of those additional strategies as fallbacks (depending on which
>> issues are still unsolved by the BNLJ).
>
>The glacier is melting more.  Tomas, what's your status here?  The
>patch has been waiting on author for two months now.  If you are not
>planning to work more on this one, then it should be marked as
>returned with feedback?

I'm not planning to do any any immediate work on this, so I agree with
marking it as RWF. I think Melanie is working on the BNL patch, which
seems like the right solution.

Sorry for the delay. I have posted the parallel-aware version BNLJ
(adaptive HJ) of this in the thread which originally had all of the
patches for it [1]. It's not near committable, so I wasn't going to
register it for a commitfest yet, but I would love feedback on the
prototype.