Bruce Momjian <pgman@candle.pha.pa.us> writes:
> Tom Lane wrote:
>> [blink] This seems to miss out on the actual point of the thread (hash
>> bucket size shouldn't be a disk page) in favor of an entirely
>> unsupported sub-suggestion.
> Yes, I was unsure of the text myself. I have changed it to:
> * Allow hash buckets to fill disk pages, rather than being
> sparse
OK, though maybe "pack hash index buckets onto disk pages more
efficiently" would be clearer.
> If we sorted the keys, how do we insert new entries efficiently?
That is why I called it "unsupported". I'm not clear what would happen
in buckets that overflow onto multiple pages --- do we try to maintain
ordering across all the pages, or just within a page, or what? How much
does this buy us versus what it costs to maintain? Maybe there's a win
there but I think it's pretty speculative ...
regards, tom lane