@gabriel-samfira Hi,I don't have a windows environment to reproduce it, do you have time to help look at it. Thanks

Hi @Iceber

Yup. I'll have a look!

I have not forgotten about this. While reproducing this, it seems that the container exits normally, but there's nothing in the stdout buffer. Will debug further and see why that is.

@aznashwan would you mind taking this one? I've been sidetracked a bit and this deserves some attention. Thanks!

While I am unsure of the cause yet, I've run the workflow with only the flaky test on repeat a dozen times or so to see how it behaves.

TL;DR:

• more likely on 2022 than 2019, but affects both at less than 2% from my testing.
• does not seem to be a race condition or conflict with other tests. (I ran TestContainerPTY by itself)
• seems to occur per go test command (i.e. running one go test -count 1000 command will either see it failing instantly or working all 1k runs), so might be a test setup issue?
• sandboxed 2019 seems immune but may be just "good luck" on it not flaking.

The runs on my fork, done over this containerd commit, can be broken down as follows:

1. single-command runs with go test -run TestContainerPTY -count 1000 (run numbers 60-65)

• all tests happened to run fine (a couple were killed for exceeding the time limit on GH actions though)

2. re-executing go test -run TestContainerPTY every time (run numbers 66-75) yielded these results:

At this stage I suspect it may be related to either how the tests are set up, or it really is a random I/O failure in HCS, but will be investigating further...