On Fri, May 7, 2021 at 3:34 PM Greg Stark wrote:
> We've talked before about buffering inserts even just for disk-based
> indexes. Much like how GIN buffers inserts and periodically flushes
> them out. We talked about doing a local buffer in each session since
> no other session even needs to see
On Fri, 9 Apr 2021 at 16:58, Tomas Vondra wrote:
>
>
>
> On 4/9/21 7:33 PM, James Coleman wrote:
> > A specific area where this is particularly painful is btree index reads.
> > Walking the tree to leaf pages isn't naturally prefetchable, and so for
> > each level you pay the random page cost. Of
On Fri, Apr 9, 2021 at 4:57 PM Tomas Vondra
wrote:
>
>
>
> On 4/9/21 7:33 PM, James Coleman wrote:
> > $SUBJECT is still a very loosely formed idea, so forgive lack of detail
> > or things I've likely missed, but I wanted to get it out there to see if
> > it sounded at all intriguing to people.
>
On 4/9/21 7:33 PM, James Coleman wrote:
> $SUBJECT is still a very loosely formed idea, so forgive lack of detail
> or things I've likely missed, but I wanted to get it out there to see if
> it sounded at all intriguing to people.
>
> Background: One of the big problems with non-local storage
