On Wed, 2020-04-08 at 16:00 -0500, Justin Pryzby wrote:
> 90% of the initial goal of this patch was handled by instrumentation
> added by
> "hash spill to disk" (1f39bce02), but this *also* adds:
>
> - separate accounting for tuples vs hashtable;
> - number of hash buckets;
> - handles other agg nodes, and bitmap scan;
>
> Should I continue pursuing this patch?
> Does it still serve any significant purpose?
Those things would be useful for me trying to tune the performance and
cost model. I think we need to put some of these things under "VERBOSE"
or maybe invent a new explain option to provide this level of detail,
though.
Regards,
Jeff Davis