mittalrishabh
mittalrishabh
please merge these changes.
please merge this change.
please merge this change.
A potential approach is to include a flag in the read index response indicating the presence of a memory lock within the region. If there is no memory lock detected...
stale reads retrying with follower reads overwhelm raft state machine causing split/scatter to slow.
I have two proposals 1. Proposal 1: Retry leader reads on follower with stale read if original request is stale reads and it tikv returns deadline exceeded or server busy...
we are doing a unnecessarily backoff of 2 seconds. https://github.com/tikv/client-go/blob/53cb1b65b8ccba665d342187cf23d0d51d6e4ac5/internal/locate/region_request.go#L2142. Our kv timeout is 500 ms.
I want to understand how it impacts tikv if there are lot of pending scatter operations. Do you have a write up on it
Lightning with parallel import can cause split operations on non empty regions. I want to understand the implication of that.
I would like to discuss the possible solutions to tackle above problems Solution 1 : Reusing the result for the request with the same ts and region id if there...