Markus Thömmes

From the logs and following a bit more of what the shim does etc, it looks like the shim thought it had actually killed the correct PID and it even...

Here some more information. This time, we didn't get the umount issues (which I believe are only a followup problem on the "kill not being effective" anyway) but I got...

Just seen another occurrence of containerd talking about an exit code but the respective process still be running. ``` Dec 13 18:58:56 fancy-machine containerd[602]: time="2023-12-13T18:58:56.486265600Z" level=info msg="CreateContainer within sandbox \"18517e9ae953629c515c5381bccb63cf7d66536fc97dfeabd5d1d4a792340b21\"...

Here's the stacktrace of the sandbox in such a case. The 68min goroutines are when the kill signal was sent. Superficially, this looks to me as if the termination signal...

@avagin thanks for chiming in. After your comments, I made a few more tests around SIGTERM handling to see if there's any regression here, but no: Even if I handle...

With a bit more digging, I think what I'm looking at is that the **container** has been successfully removed since its processes are gone (including its state file as mentioned...

@manninglucas any chance that the two rather recent commits https://github.com/google/gvisor/commit/3ab01aedb8741e09f90ec2858fbd80077757347b and https://github.com/google/gvisor/commit/6a112c60a257dadac59962e0bc9e9b5aee70b5b6 have anything to do with this?

@avagin here we go. Sadly, as noted above, the "inner" container correctly stops and so there's no processes left that I could check there. All the python fds on the...

@ayushr2 this is running with `directfs=false` already.

Sadly, I don't have a good reproducer for this so I can't easily confirm. I'll have to get a new version into the pipeline and let it sit for a...