[CLOUD-3188] Move GC logs to PREPEND_JAVA_OPTS to bypass the specific EAP checks

[yersan]

The solution is a bit hacky, but I haven't found a better approach, so maybe this one is acceptable.

EAP in standalone.sh will avoid adding any GC log file configuration if there is any JAVA_OPTS setting related to GC logs. This check clashes with the jvm_specific_diagnostics for JDK 11, where the -Xlog:gc::utctime decorator is enabled.

To prevent EAP skips the GC log file configuration due to the jvm_specific_diagnostics settings, we could remove the settings from the JAVA_OPTS and add them to PREPEND_JAVA_OPTS. That change will bypass the EAP specific checks.

jira issue: https://issues.redhat.com/browse/CLOUD-3188

[jfdenise]

+1, looks fine.
With your fix in, we have both gc logs in console and in gc.log file. That is allowed (we can have multiple -Xlog property set in command line).
I understand the motivation of the check in standalone.sh but this check doesn't seem right: https://github.com/wildfly/wildfly-core/blob/master/core-feature-pack/src/main/resources/content/bin/standalone.sh#L271
It does check for OpenJDK8 and OpenJDK 9+ Xlog:gc although we could have for example -Xlog:age,gc. Furthermore the added VM properties are also for OpenJ9.
https://github.com/wildfly/wildfly-core/blob/master/core-feature-pack/src/main/resources/content/bin/standalone.sh#L283

This seems a bug to me.

[jfdenise]

Actually, the grep part could fail in case JDK_DIAGNOSTICS change the syntax. Shouldn't we move the JDK_DIAGNOSTICS in all cases to PREPEND_JAVA_OPTS?

[yersan]

I did in that way to apply the changes only when it is strictly necessary based on the current standalone.sh, and hence to minimize any possible side effects when the JDK_DIAGNOSTICS does not contain a -Xlog:gc setting.

However, I agree @jfdenise. Indeed, we should end up without duplicates in the JAVA_OPTS so any bits adding the same settings after processing this, does not make sense, so order here is irrelevant. Well, the -Xlog settings are a bit more tricky in this aspect since the same setting is overwritten by the following -Xlog argument configuring the same aspect. e.g if JDK_DIAGNOSTICS defines a log file by using -Xlog:gc:file=/one/path and the standalone.sh adds a different path using -Xlog:gc:file=/other/path, the latest configuration wins. It could make sense in this context, although enabling JAVA_DIAGNOSTICS is a user choice, so I could expect to see how those settings overrides the final JAVA_OPTS.

[yersan]

I've just removed the grep check and appending the jvm specific diagnostics if it was activated and there is one