#1395 - fix delay between the first active user(s) and the first request(s) in Karate-Gatling

[theathlete3141]

Relevant Issues : #1395
Type of change: Bug fix for existing feature

Whilst running CatsSimulation in karate-gatling and using logging statements, I found that the line

ScanResult scanResult = new ClassGraph().acceptPaths("/").scan()

appears to be very slow (at least when using a high user rate e.g. ramp 100 users during 10 seconds).

This is called from:

Suite.java - Resource resource = ResourceUtils.getResource(workingDir, name)
Suite.java - karateBase = read("classpath:karate-base.js")
Runner.java - Suite suite = new Suite(builder)
KarateAction.scala - Runner.callAsync(name, tags.asJava, arg, perfHook)

To try to speed this up, we should do it once:

private static final ScanResult scanResult = new ClassGraph().acceptPaths("/").scan();

However, now PerfHookTests.java testPerfHook6 fails, but only when run collectively as the collective test suite (it does not fail when ran on it's own). It seems that sometimes scanResult.getResourcesWithPath(removePrefix(path)) returns null and leads to a NullPointerException rather than the expected "Feature not found exception".

Digging a bit further, the first three times this method is called it will always returns a non-null value (in this case an empty ResourceList), but after this it can (and does) return null. Therefore we need to add the handling:

if (rl == null) {
	rl = ResourceList.emptyList();
}

Note that rather than

private static final ScanResult scanResult = new ClassGraph().acceptPaths("/").scan();

We could instead do this lazily if preferred:

private static ScanResult scanResult;
private synchronized static ScanResult getScanResult() {
	if (scanResult == null) {
		scanResult = new ClassGraph().acceptPaths("/").scan();
	}
	return scanResult;
}

Or:

private static ScanResult scanResult;
private static final Object scanResultLock = new Object();
private static ScanResult getScanResult() {
	if (scanResult == null) {
		synchronized (scanResultLock) {
			if (scanResult == null) {
				scanResult = new ClassGraph().acceptPaths("/").scan();
			}
		}
	}
	return scanResult;
}

[ptrthomas]

@theathlete3141 this is great stuff, I had a hunch this was causing some impact, e.g. the CPU utilization during the build and had tried a static init - but your solution sounds great. thanks for the investigation