Handle OS interrupts in lokoctl to fix leaking terraform process.

[ipochi]

Fixes #329

Registers a Signal handler for terraform process and components installation/deletion.

Before running the lokoctl cluster apply

ps ef | grep terraform | wc -l
0

module.packet-liger.data.ct_config.controller-install-ignitions[0]: Refreshing state...
module.worker-madrid.data.ct_config.install-ignitions: Refreshing state...
module.worker-madrid.packet_device.nodes[0]: Creating...
module.packet-liger.packet_device.controllers[0]: Creating...
module.worker-madrid.packet_device.nodes[0]: Still creating... [10s elapsed]
module.packet-liger.packet_device.controllers[0]: Still creating... [10s elapsed]
^CInterrupt received.
Please wait for Terraform to exit or data loss may occur.
Gracefully shutting down...
Stopping operation...

After Ctrl + C

ps ef | grep terraform | wc -l
0

[invidian]

Some thoughts, please have a look.

[iaguis]

Some partial review I did the other day.

[ipochi]

After a call with @invidian we found out that terraform apply handles the interrupts but terraform init doesn't and this behavior is reflected the same in lokoctl as well.

An upstream issue has been filed for the same.

hashicorp/terraform#25118

@invidian @iaguis please re-review.

please re-review

[iaguis]

I have some nits and a general comment.

From my testing it seems when I interrupt lokoctl at, say, terraform init this doesn't stop lokoctl. I'm aware terraform init doesn't handle interrupts but I would expect lokoctl to stop after terraform init is finished instead of continuing with the next steps (terraform plan and so on). This works fine if I do two interrupts for example when terraform plan is running, I guess because the process ends with exit status 1 and we propagate the error.

Should we add a channel to send interrupt information from the signal handler to the Executor so we can os.Exit() or error out after the command is finished if we get an interrupt?

[ipochi]

I have some nits and a general comment.

From my testing it seems when I interrupt lokoctl at, say, terraform init this doesn't stop lokoctl. I'm aware terraform init doesn't handle interrupts but I would expect lokoctl to stop after terraform init is finished instead of continuing with the next steps (terraform plan and so on). This works fine if I do two interrupts for example when terraform plan is running, I guess because the process ends with exit status 1 and we propagate the error.

Should we add a channel to send interrupt information from the signal handler to the Executor so we can os.Exit() or error out after the command is finished if we get an interrupt?

It would become quite messy, besides there would be two branches where it would be handle the differently if the current process is terraform version/init vs terraform plan/apply/refresh

Nevertheless, I am opening up a follow up issue that tracks this limitation and and will also link the upstream issue in the follow up issue.

[ipochi]

Follow up issue created #555

[invidian]

LGTM

Addressed your feedback, follow up issue created, please re-review changes.

[iaguis]

LGTM. Thanks!