Thomas B
Thomas B
> rbtree instead of fifo Looks like this would be binary heap, not rb tree. That's only a factor 2 of perf. between the two though.
> To ensure efficient batching and minimize network overhead, it's crucial to group insert requests that will ultimately end up on the same node within ScyllaDB. This necessitates the computation...
@wprzytula > This is also related to #974 where a batch that is token-aware would be much more efficient. Updated the PR. It looks like it indeed does solve #974....
> Why is it not enough? Thanks for the review. I initially thought it was enough as well, however > As [pointed out in the first review](https://github.com/scylladb/scylla-rust-driver/pull/738#discussion_r1243336565), this was not...
> The solution is to batch by token, not by "first" replica. Unless I'm mistaken because the space of tokens is large this would imply that we very rarely are...
> Afaik it doesn't really make sense to perform multi partition batches. This incurs additional work for the coordinator node and you would be better off just using single, non-batched...
> otherwise you start doing multi-partition batches, which incur more work for the coordinator node and are not a good idea afaik The main issue with not batching is that...
> thus making it maximally performant It looks like at the moment the cluster may have to handle some extra networking and individual handling for every statement that might put...
> OK, could you re-iterate what kind of API you need specifically access to in order to get such client-side performance gains? It looks like the discussion here is joining...
That seems to not be the source of the issue. ## nix run When I `nix run` it uses cargo 1.76.0 as pinned in your `nixpkgs`, since `nix derivation show...