Original:
I've just opened
this proposal to improve performance of concurrent merges by performing them in parallel (will happen anyway in KV... which just makes this proposal save more time!) and reusing their "failed" metaranges.
It is kind-of an analogue to the lock-free improvement for concurrent commits. I expect it to be quite effective. 🙂
Very interested in your comments (especially if they being "there is no way this could ever work"...)!