Elastic Agent on Windows cannot be stopped or removed if --delay-enroll is retrying

[strawgate]

per elastic/elastic-stack-installers#277 and likely caused by #4800

The Elastic Agent on Windows is unable to be stopped after the service is started post a --delay-enroll installation if the enrollment cannot be completed.

To reproduce:

1. Install the agent with --delay-enroll and a fake fleet server or bad key
2. Restart the system or start the service in task manager
3. Notice in the Agent log that the agent is failing to enroll
4. Attempt to stop the agent via service manager or uninstall the agent via elastic-agent uninstall

The agent service should respond to stop requests even if enrollment is failing, and uninstall should not be blocked by bad enrollment information.

[elasticmachine]

Pinging @elastic/elastic-agent-control-plane (Team:Elastic-Agent-Control-Plane)

[pierrehilbert]

Unfortunately as I don't have a Windows setup I didn't test it but is it working if you are doing elastic-agent uninstall --force to force the uninstall?

[blakerouse]

Interesting because the loop clearly watches for the context to be cancelled and doesn't continue. That context also gets cancelled when the signal is given to the process. This is strange behaviour:

elastic-agent/internal/pkg/agent/cmd/run.go

Lines 572 to 580 in 00d7268

	for {
	if ctx.Err() != nil {
	return nil, ctx.Err()
	}
	err = c.Execute(ctx, cli.NewIOStreams())
	if err == nil {
	// enrollment was successful
	break
	}

[pierrehilbert]

@blakerouse did you try to reproduce or not yet? (just for me to understand where we are with the investigations)

[blakerouse]

I have not tried to reproduce, just providing code context. Some times these issues are easy to fix by just looking at the code and noticing an error.

[cmacknz]

As part of the original PR I did try this on a Windows machine and it let me uninstall: #4727 (comment)

However it is possible something has changed since I tested it originally.