Fu Wei

@ambarve ~~would you mind share your CRI setting? If you're using shim-v1, it seems expected.~~ sorry for misunderstand it. Checking.

https://github.com/containerd/containerd/blob/f5e7fe0cb6286333a597a407029fe7c760626669/core/runtime/v2/manager.go#L221 For main branch, v2.0, the sandbox mode won't start shim to get address. It connects to shim and let the hook-on-close empty. It won't call shim delete anymore. I'm...

I think we need to reconsider this because the default mode is using pod sandbox and shim author doesn't notice this change. Or highlight this change in 2.0

> was because of a bug. I don't think so. Please check out that shim v2 command line design doc. The shim v2 introduces binary delete to ensure resource can...

> I am not saying that Containerd issuing shim delete is a bug. What I am saying is that shim delete should be issued only once per shim/sandbox. Sorry for...

Hi @lengrongfu is it test issue or component issue? cc @klihub

I will setup my arm64 env to test it as well. Updated it later.

closed because it's related to cAdvisor

> but it can not resolve the problem of forwarding IO failed. If the exec-cmd exits before cid-stdout/cid-stderr fifos opened by containerd, forwarding IO will be failed and some goroutines...

Hi @gaohuatao-1 sorry for late reply. I think go_threads number doesn't mean goroutine leak. If existing threads are stuck in kernel space, go runtime will clone new threads to handle...