MlMappingsUpgradeIT.testMappingsUpgrade fails with testclusters

[alpar-t]

I'm working in porting bwc tests to run with testclusters and can't get this one to pass:
https://gradle-enterprise.elastic.co/s/oqrlvw6evjqa4/tests/q3vpzt4orcdaq-vgix7kzieu2xs

I'm going to mute this since it's the only one that fails with this conversion and kindly ask someone from the team to take a look.

[elasticmachine]

Pinging @elastic/ml-core

[alpar-t]

The PR that implements the conversion is #46265

[droberts195]

For the record, this test is muted in 7.5 through to master.

[droberts195]

The problem is this:

[2019-12-12T15:40:11,596][ERROR][o.e.x.m.p.l.CppLogMessageHandler] [v7.6.0-0] [controller/4715] [CDetachedProcessSpawner.cc@184] Child process with PID 4863 was terminated by signal 9
[2019-12-12T15:40:11,733][INFO ][o.e.x.m.j.p.a.AutodetectProcessManager] [v7.6.0-0] Successfully set job state to [failed] for job [ml-mappings-upgrade-job]

The test clusters code is killing the ML jobs before it kills the ES JVM.

[droberts195]

The code that causes the problem is

elasticsearch/buildSrc/src/main/java/org/elasticsearch/gradle/testclusters/ElasticsearchNode.java

Lines 840 to 841 in d442ff9

	// Stop all children first, ES could actually be a child when there's some wrapper process like on Windows.
	processHandle.children().forEach(each -> stopHandle(each, forcibly));

It needs to kill the ES JVM before the ML processes. Otherwise the ES JVM thinks the ML processes have crashed, records them as failed and then they don't start again automatically during the rolling upgrade.