Default Eclipse JVM should not have -XmX parameter

[grandinj]

I have noticed that recent installers, when using the default supplied VM to run Eclipse, has an -XmX parameter.

This leads to an extremely frustrating experience with large projects. I suddenly found my debugger doing all sorts of very weird things, which led to a loooooooong goose chase before I found the cause.

Either
(a) the platform needs some indication that it is running into memory issues
(b) just leave that parameter out, in which case the VM will grow to use available memory on the machine.

(I vote for option b)

Thanks.

[iloveeclipse]

I believe option (b) makes sense. On "usual" working environments today we have way more RAM as before and JVM would "automatically" reserve more than default 2GB (SDK sets -Xms256m -Xmx2048m in the sdk.product).

The only problem I see "smaller" configurations like Jenkins platform builds that usually suffer from too strict resources limits and where JVM would set smaller heap.
But there we could set VM heap limits in Jenkins files.

[merks]

FYI, this issue was discussed here as well:

eclipse-packaging/packages#37

Most folks are using one of the EPP packages, so what the Platform does has much narrower impact than what EPP does...

[jonahgraham]

As I mentioned in eclipse-packaging/packages#37 (comment) the EPP normally aims to follow where the Platform/SDK leads. This has been the case because we used to inherit the Platform settings, that was changed recently(ish) which gives us the option to drop the xmx from EPP.

In this case perhaps EPP should lead and platform/SDK can follow along later?

[iloveeclipse]

In this case perhaps EPP should lead and platform/SDK can follow along later?

Or we can just try to do this in SDK and see if something breaks on Jenkins/official build & fix Jenkins files etc in case we still need Xmx option. I will push a PR in a moment.

[iloveeclipse]

Or we can just try to do this in SDK and see if something breaks on Jenkins/official build & fix Jenkins files etc in case we still need Xmx option.

Any objections? => #1471

[iloveeclipse]

So far there were no regressions in the SDK tests, we will see if some Jenkins jobs would fail.

The default SDK eclipse.ini looks like this now (Linux build):

-startup
plugins/org.eclipse.equinox.launcher_1.6.600.v20231012-1237.jar
--launcher.library
plugins/org.eclipse.equinox.launcher.gtk.linux.x86_64_1.2.800.v20231003-1442
--launcher.defaultAction
openFile
--launcher.appendVmargs
-vmargs
-Dosgi.requiredJavaVersion=17
-Dosgi.dataAreaRequiresExplicitInit=true
-Dorg.eclipse.swt.graphics.Resource.reportNonDisposed=true
--add-modules=ALL-SYSTEM
-Dorg.slf4j.simpleLogger.defaultLogLevel=off
-Djava.security.manager=allow

[iloveeclipse]

I see a first OOM in the Jenkins test:
https://ci.eclipse.org/platform/job/eclipse.platform/job/PR-731/5/console

@laeubi : I can't understand what maved output says and in which task it had OOM. It also looks like there are multiple tasks running in parallel (see org.eclipse.core.tests.resources test output mixed with all other entries).

There is also one maven error [ERROR] MavenReportException: Error while generating Javadoc.

Can you help understand following:

1. Which Eclipse/tycho task has reported [ERROR] java.lang.OutOfMemoryError: Java heap space? Is it org.eclipse.core.tests.resources test? Something else?
2. How to configure that one to use more heap (as it looks like Jenkins is very low on resources)? If that is about test execution, I assume we have to add <argLine>-Xmx2G to the pom.xml file in question. But to which one (see first question)?

[laeubi]

It also looks like there are multiple tasks running in parallel

Platform build is executed in parallel

Which Eclipse/tycho task has reported [ERROR] java.lang.OutOfMemoryError: Java heap space? Is it org.eclipse.core.tests.resources test? Something else?

You can see the limits here: https://ci.eclipse.org/platform/job/eclipse.platform/job/PR-731/5/consoleText

  resources:
      limits:
        cpu: "2000m"
        memory: "4096Mi"
      requests:
        cpu: "1000m"
        memory: "4096Mi"

but I don't really understand how eclipse.ini relates to Jenkins build execution... so if we know how/if this makes a difference there might be a way to further configure this, e.g. for maven one would use .mvn/jvm.config but tests are using different setting depending on they are forked or not...

[iloveeclipse]

but I don't really understand how eclipse.ini relates to Jenkins build execution...

If the tycho starts SDK product during tests, it uses eclipse.ini we've changed in #1471.

I honestly don't know what tycho does to construct the test application and if that inherits eclipse.ini coming from the SDK product. If tycho based test instance inherits that file, it also doesn't use -XmX2g defaults anymore and probably uses only 1GB heap (1/4 from machine RAM on Jenkins).

But still open question: any clue which task has reported OOM?

[laeubi]

But still open question: any clue which task has reported OOM?

infocenter-product 1.0.0-SNAPSHOT .................. FAILURE [ 7.372 s]

is the module failing the build, but this might only be a side-effect of the JVM itsel running out of memmory, thats why I said, I can't see that a PROCESS (that can read eclipse.ini) is failing, the JVM that running the BUILD seems going OOM...

[iloveeclipse]

What I mean is something like eclipse-platform/eclipse.platform#767.
But assuming all tests has to be adopted, is there a more elegant solution (as adding this option in every pom)?

[laeubi]

If you want to share things in common, it is best placed here in pluginManagement section of the build:
https://github.com/eclipse-platform/eclipse.platform.releng.aggregator/blob/master/eclipse-platform-parent/pom.xml

[iloveeclipse]

infocenter-product 1.0.0-SNAPSHOT .................. FAILURE

I don't see which / where is this FAILURE reported in the log (above the table)? The last relevant entry I see is an INFO
10:16:48.454 [INFO] --- tycho-p2-repository:4.0.4-SNAPSHOT:assemble-repository (default-assemble-repository) @ infocenter-product

[iloveeclipse]

If you want to share things in common, it is best placed here in pluginManagement section of the build: https://github.com/eclipse-platform/eclipse.platform.releng.aggregator/blob/master/eclipse-platform-parent/pom.xml

So this line has to be adopted to have -Xmx option:

eclipse.platform.releng.aggregator/eclipse-platform-parent/pom.xml

Line 529 in 10afb28

<argLine>${surefire.testArgLine} ${surefire.platformSystemProperties} ${surefire.systemProperties} ${surefire.moduleProperties}</argLine>

[iloveeclipse]

Change is reverted, due possibility that is causes OOM in Jenkins builds.
If someone sees OOM's an platform jenkins builds after next IBuild, please comment here.

[iloveeclipse]

OK, looks like we see OOM's elsewhere without this patch, see eclipse-platform/eclipse.platform#764.

So I plan to re-submit #1471 as soon as we have solution for more RAM on Jenkins, see https://gitlab.eclipse.org/eclipsefdn/helpdesk/-/issues/3885 .

See #1483

[iloveeclipse]

eclipse-platform/eclipse.platform#784 provided now 8 GB RAM for Platform builds on Jenkins, let see if that helped or not.

[jonahgraham]

Has there been any negative effects of this? EPP will make this same change too unless someone raises a concern inhttps://github.com/eclipse-packaging/packages/issues/37

[iloveeclipse]

Has there been any negative effects of this?

Only for poor VM (Jenkins) instances with <= 4GB RAM.

For PC's/VM's with >=8 GB everything should be fine.