We have found a bug that pods restarted after containerd restarted, especially on the condition that the container IO is very frequent. The parent process of the container processm that is our async containerd
-shim process is not very careful to handle the pipe of container process IO. Thus when containerd restarts, some of containers, especially that has huge IO requests, would receive EPIPE and exited unexpectly.
We have found a bug that pods restarted after containerd restarted, especially on the condition that the container IO is very frequent. The parent process of the container processm that is our async containerd -shim process is not very careful to handle the pipe of container process IO. Thus when containerd restarts, some of containers, especially that has huge IO requests, would receive EPIPE and exited unexpectly.