memdb: Implement btree.Item on item directly, in addition to *item

[ValarDragon]

Currently we force every btree item to be *item, not just item. This in turn causes every set to require new heap allocations, which turns out to be a significant overhead. These allocations cause lots of memory to need to be garbage collected, and this ends up often being the bottleneck, not the actual btree searching operations.

In Osmosis, we get that we have ~no item spent in the btree operations, its all spent in allocating the newPair onto Heap. Heres a screenshot of a relevant flamegraph output here:

[tac0turtle]

Hey, I can help in getting a release out to fix this, if you are willing to make a PR. I can push a branch for you to target as master has diverged too much to be used for this

[ValarDragon]

smh stale bot

[ValarDragon]

Can we please remove stale bot issue closing =/ Its not even that long of a wait atm

[tac0turtle]

yea ill fix

[faddat]

What's the status of this one?

I guess I want to include it in the notinal tracking issue

@ValarDragon do you think this is needed still?

[ValarDragon]

Yeah this is still useful

[faddat]

great!

[faddat]

So, fair to say this one is more important than #188 ?

that’s good because ah, wow

Pain is a teacher I guess?

[faddat]

Seems set :)