Andrew Fitzgerald
Andrew Fitzgerald
Was bench-tps run to compare before/after this change?
Moving my points from slack to here for more visibility. Assuming you have a snapshot, and the append vecs should be consistent with that snapshot: 1. Unpack the snapshot only...
On point 1: - We remap because a full and incremental snapshot could come from different sources. That can lead to a collision in the AppendVecIds. - We could potentially...
> There will be no more threads. It will be like the one thread in 1.c which does one pass on all accountvec files. is this to avoid the remapping,...
> I thought multiple threads were at the unpacking side. checked again, yes, the rebuilding is also multi-threaded. OK, I can make it the same. But looking at the code,...
> Maybe per-appendvec hash computation can also be done here? Yeah, seems like a pretty natural place to do it if we're adding per-file hash verfication. Just return (Self, Hash)...
Are you on an intel mac? I'm not seeing these tests fail on my arm mac, FYI. Just ran off of c04747dd
> Can you link to a place in the code base where this proc macro would be useful, esp. compared to `measure!()`? Draft PR: #26976 (an example, but also ones...
>I think I'm still in the boat where without a specific example, the scope-escaping potential seems like too much of a footgun to me. Well that's my fault, I referenced...
I have been primarily running against 4 main benchmark cases: 1. unrelated: mostly default settings, with a large number of accounts so tx are mostly unrelated ``` cargo run --release...