Fix resource leak

[nicolai86]

We ran into issues with *terraform.Apply where individual executions are interrupted and retried.

If an execution is interrupted without sending SIGINT terraform losses changes to the backend, breaking subsequent retries.

Some context when this happens:

• we use the eks module eks
• we use the pg backend for terraform

This change fixes the issue for us.

I'll try to add a minimal reproducable setup to show case in a bit.

[radeksimko]

Hello @nicolai86
Thanks for the report and the contribution.

I believe #512 which leverages more of the native os/exec API, should address the mentioned issues, including the one from your TODO. The default timeout after cancellation (which now sends SIGINT instead of SIGKILL) is 60s after which SIGKILL is sent. This is also adjustable as per v0.23.0 Changelog.

Feel free to upgrade to the most recent version and let us know if this is still an issue.