runner tempdir patch

[dacbd]

• partially fixes: No space left on device creates hung EC2 instance cml#1006
• semi-related: canonical bin path #581

I think the real fix is probably to run the GHA client in a separate slice enforcing resource limits

[0x2b3bfa0]

Launching the GitHub Actions runner in a separate slice is the ersatz version of using containers. 🙃 Still, sounds like a good workaround to have a bit more control over runners without triggering the container detection logic from #146 (comment) and actions/runner#367 (comment).

[0x2b3bfa0]

Silly question: is the default mktemp path being mounted as tmpfs on the images we use? 🤔

[dacbd]

I confess to not completely replicating the original failure brought by @JacksonMaxfield to confirm, but

Silly question: is the default mktemp path being mounted as tmpfs on the images we use? 🤔

No, as I have seen it survive reboots, however, I believe there is a reaping process that hits the dir. Or there was an existing process already in memory that deleted enough files from disk when it got SIGTERM from rebooting my test.

[dacbd]

@0x2b3bfa0 looks like we should just do this: #62 ?

[0x2b3bfa0]

@0x2b3bfa0 looks like we should just do this: #62?

Not until Docker releases the fix? 🤔

[dacbd]

sorry I just saw that the link PR has been merged so I assumed it's been released?

[0x2b3bfa0]

As per the relevant milestone, it has not been released yet

[0x2b3bfa0]

No, as I have seen it survive reboots, however, I believe there is a reaping process that hits the dir. Or there was an existing process already in memory that deleted enough files from disk when it got SIGTERM from rebooting my test.

I guess I don't have enough context. 😅 Looks clear as to me.

[0x2b3bfa0]

If /tmp is not mounted as tmpfs, what does this pull request fix? 🤔