fix: Add trap to agent startup script to sleep on failure
#2873
Conversation
32ecb43 to
3a0d162
Compare
provisionersdk/agent.go
Outdated
| @@ -18,6 +18,7 @@ Start-Process -FilePath $env:TEMP\sshd.exe -ArgumentList "agent" -PassThru` | |||
|
|
|||
| linuxScript = `#!/usr/bin/env sh | |||
| set -eux pipefail | |||
| trap '[ $? -ne 0 ] && echo === Agent script exited with non-zero code. Sleeping infinitely to preserve logs... && sleep infinity' EXIT | |||
There was a problem hiding this comment.
I'm a bit worried about sleep infinity compatibility. From a quick test it doesn't seem to work on macOS. It does seem supported in busybox, but e.g. in toolbox it doesn't (more common in Android world I think).
I wonder if it's be enough to sleep just a "long time", say something in the ballpark of 1-24h? I think anything above 15s should spare the Docker container from being erased.
PS. It doesn't hurt, but I don't think we need the non-zero exit check here since the exec at the end replaces this process, the trap should also go away. Up to you which you prefer, though.
There was a problem hiding this comment.
Wonderful catch, I'll make it a static duration (I agree 24h should be plenty).
The non-zero exit code can be removed with exec, another great catch!
The Docker Terraform provider removes containers immediately on exit, making it difficult to debug a failed container start with Coder. This will sleep on exit and output a friendly log, which should assist with debugging failures.
3a0d162 to
b3c8a87
Compare
Co-authored-by: Mathias Fredriksson <mafredri@gmail.com>
Co-authored-by: Mathias Fredriksson <mafredri@gmail.com>
The Docker Terraform provider removes containers immediately on exit, making
it difficult to debug a failed container start with Coder. This will sleep on
exit and output a friendly log, which should assist with debugging failures.