Tom Lane wrote:
> Heikki Linnakangas <heikki@enterprisedb.com> writes:
>
>> A big question is, do we need to implement spilling to disk?
>>
>
> My thought is no, at least not in the first cut ... this is something
> that can be added later if it proves critical, and right at the moment
> my guess is that it never will. The data structure design sounds fine.
>
I thought so too.
We could also limit the size of the hash table, which takes up most of
the memory, and only keep the latest phantom cids there. Presumably, if
current command id is 1000, you're not likely to set cmax to 500 on any
tuple in that transaction anymore.
-- Heikki Linnakangas EnterpriseDB http://www.enterprisedb.com