Fu Wei

> Since buffer always have the latest value for other buckets, there should be no problem with overlap between buffer cache and read transacton Hi @siyuanfoundation sorry for late reply....

SGTM. Will keep more eyes on this.

> There is at least an open comment from @fuweid: https://github.com/etcd-io/etcd/pull/15979/files#r1467819098. It's just my two cents for using `%w`. In the whole project, it's using `: %w`. If `err=` prefix...

Hi, @ahrtr. Sorry for late reply. > Is AKS still using 3.4? Yes. And we are also using other versions depending on the cluster. For this issue, it seems reasonable...

Sorry for late reply. This issue was found when I was trying to upgrade grpc-gateway deps. Sorry for the unclear comment. The reproduce step is here ```bash $ git checkout...

I think we can just add tag !cluster_proxy to disable its

Hi @siyuanfoundation you can try to limit cpu by the following command and use patch #17248 ``` EXPECT_DEBUG=true GO_TEST_FLAGS=${GO_TEST_FLAGS} RESULTS_DIR=/tmp/results taskset -c 0,1,2 make test-robustness ``` Hope it can help.

If no process holds lock on snap dir, `ulimit -x` is ulimited? maybe the user holds too many flocks

UPDATED: When server closes the backend, the backend will stop background commit goroutine and reset read transaction. https://github.com/etcd-io/etcd/blob/c8b4b168aff7f84d3b10dc9ddc13a381f15a710c/server/storage/backend/read_tx.go#L133-L138 For `baseReadTx`, it doesn't check the `tx` in UnsafeRange and update `txWg`....

Hi @serathius Sure. Will file pull request later.