Allow JIT killing of Raven.Embedded.EmbeddedServer by tearing it down as part of stop sequence

[ramonsmits]

Teardown Raven.Embedded.EmbeddedServer as part of stop sequence so that we kill the instance if needed as adhering to cancellation is not possible in host builder root container dispose.

[johnsimons]

@ramonsmits I am trying to understand the reason for this code.
I thought we wanted to give a chance for RavenDB server to end on its own, and then if it takes a long time, then we resort to killing it, or am I misunderstanding?

[ramonsmits]

I thought we wanted to give a chance for RavenDB server to end on its own, and then if it takes a long time, then we resort to killing it, or am I misunderstanding?

@johnsimons that is exactly what this code should do:

• Launch the dispose as a task, and the timer
• Wait for either to complete
• If its the timer task it ran out of time and we kill the child process

[johnsimons]

I may be wrong here @ramonsmits, but wouldn't in that case there would be an issue with passing the same cancellationToken token to both tasks.
For example, the cancellation token is triggered, so both tasks (the Delay and the Raven process will receive that cancellation), the Delay task will throw an exception (CanceallationException, I can't really remember), but the Raven process may take 30 secs to finish off, so the chances of the Delay being first are quite high IMO, right?

[johnsimons]

I was just looking at what EmbededServer is, and it seems to be just a wrapper for the RavenDB server process.
Which makes me wonder if we could do a better job by just implementing that code ourselves.
Has this been considered at all?

[ramonsmits]

I was just looking at what EmbededServer is, and it seems to be just a wrapper for the RavenDB server process. Which makes me wonder if we could do a better job by just implementing that code ourselves. Has this been considered at all?

@johnsimons I have been looking at that too:

• EmbeddedServer support cancellation ravendb/ravendb#20177

We could clone that code into our code base but it would just be better if they would have a client.

[johnsimons]

We could clone that code into our code base but it would just be better if they would have a client.

Do we even need to do that, like I said it is just a wrapper around Process.Start .... Process.End, TBH, not very useful

[johnsimons]

I found the usage of the Task.Delay with infinite time a bit confusing.
So I wonder if something like this would still work and be simpler, thoughts @ramonsmits ?

Suggested change

	// TODO: await EmbeddedServer.Instance.StopAsync(cancellationToken);
	var processId = await EmbeddedServer.Instance.GetServerProcessIdAsync(cancellationToken);
	var waitForCancellationTask = Task.Delay(Timeout.Infinite, cancellationToken);
	var firstTask = await Task.WhenAny(
	Task.Run(() => EmbeddedServer.Instance.Dispose(), cancellationToken),
	waitForCancellationTask
	);
	if (firstTask == waitForCancellationTask)
	{
	try
	{
	Logger.WarnFormat("Killing RavenDB server PID {0} child process because host cancelled", processId);
	var ravenChildProcess = Process.GetProcessById(processId);
	ravenChildProcess.Kill(entireProcessTree: true);
	// Kill only signals
	Logger.WarnFormat("Waiting for RavenDB server PID {0} to exit... ", processId);
	await ravenChildProcess.WaitForExitAsync(CancellationToken.None);
	}
	catch (Exception e)
	{
	Logger.ErrorFormat("Failed to kill RavenDB server PID {0}\n{1}", processId, e);
	}
	}
	serverOptions = null!;
	Logger.Debug("Stopped RavenDB server");
	var processId = await EmbeddedServer.Instance.GetServerProcessIdAsync(cancellationToken);
	try
	{
	await Task.Run(() => EmbeddedServer.Instance.Dispose(), cancellationToken);
	}
	catch (OperationCanceledException) when (shutdownCancellationToken.IsCancellationRequested)
	{
	try
	{
	Logger.WarnFormat("Killing RavenDB server PID {0} child process because host cancelled", processId);
	var ravenChildProcess = Process.GetProcessById(processId);
	ravenChildProcess.Kill(entireProcessTree: true);
	// Kill only signals
	Logger.WarnFormat("Waiting for RavenDB server PID {0} to exit... ", processId);
	await ravenChildProcess.WaitForExitAsync(CancellationToken.None);
	}
	catch (Exception e)
	{
	Logger.ErrorFormat("Failed to kill RavenDB server PID {0}\n{1}", processId, e);
	}
	}
	serverOptions = null!;
	Logger.Debug("Stopped RavenDB server");