[BUG] [@osd/opensearch] Calling stop on an already ended Cluster producess different results on different platforms

[AMoo-Miki]

Describe the bug

OSD uses tree-kill [npm], a library that terminates a process and all of its children, for stopping processes started using Cluster. This library behaves differently on different platforms. If the pid it is asked to terminate is not running:

• on darwin and linux, it silently returns without throwing and exception
• on win32 in produces an exception

As a result, calling cluster.stop() on an already terminated cluster will throw and exception on Windows but will silently continue on other operating systems.

OpenSearch-Dashboards/packages/osd-opensearch/src/cluster.js

Lines 222 to 235 in d7004dc

	async stop() {
	if (this._stopCalled) {
	return;
	}
	this._stopCalled = true;
	if (!this._process || !this._outcome) {
	throw new Error('OpenSearch has not been started');
	}
	await treeKillAsync(this._process.pid);
	await this._outcome;
	}

This library is also used by ProcessRunnder of @osd/dev-utils and impacts its ability to successfully terminate the processes it starts.

Expected behavior
No exception thrown

Dashboards Version
main

[seanneumann]

Triage - @AMoo-Miki to apply a fix to mitigate it, but we should also follow-up with a longer term solution.

[AMoo-Miki]

The temporary workaround is in. However, the library has several issues about its inability to kill a process; a replacement needs to be found/created.