Skip to main content
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] [List Home]
Re: [cbi-dev] OOM in Standalone TCK build job and question about where do the default NLP_PROTOCOL_OPTS settings get configured?


On Tue, Jun 2, 2020 at 9:17 AM Mikael Barbero <mikael.barbero@xxxxxxxxxxxxxxxxxxxxxx> wrote:
Hi Scott,

The JNLP_PROTOCOL_OPTS is defined here https://github.com/eclipse-cbi/jiro-agents/blob/master/default.libsonnet#L37. It is then used in the docker image and in the podtemplates configurations to define environment variables.  It's a default value for all provided agents and for the default jnlp containers on the cluster. We recognize that this can be too low of a value for some projects (see bug https://bugs.eclipse.org/bugs/show_bug.cgi?id=560283) but it works in 95% of the cases so we want to keep the default low. 

The way you override it in https://github.com/eclipse-ee4j/jakartaee-tck/blob/master/Jenkinsfile#L130 is the correct way to do it for a pipeline job.

Unfortunately, job [2] is a freestyle job, so you cannot override it by yourself. Two options: 
  1. you either move to a pipeline job (probably with just a jnlp container in the pod template) and you do the same kind of trick as you did for [2], 
  2. or we increase the value for all your freestyle jobs on your instance. It means that each build will consume more resources and may block other jobs from running until they complete (when you reach the higher end of your resources quota).
Which solution do you prefer? IMO, you should go with #1, it will give you greater control and freedom in tweaking resources for each jobs individually.

Mikaël, 

Thank you for explaining, #1 sounds like the best solution, as our (memory) needs will change over time.

Cheers,
Scott

 

Cheers,

Mikaël Barbero 
Team Lead - Release Engineering | Eclipse Foundation
🐦 @mikbarbero
Eclipse Foundation: The Platform for Open Innovation and Collaboration

Le 2 juin 2020 à 04:12, Scott Marlow <smarlow@xxxxxxxxxx> a écrit :

Hi,

I would like to understand what the container/VM memory size is that goes with the "OutOfMemoryError: Java heap space" failures we get collecting job results from standalonetck-nightly-build-run-master [1].  

I do see from standalonetck-nightly-build-run-master/47/injectedEnvVars [2] that JNLP_PROTOCOL_OPTS contains -Xmx256m.  Can we identify where the JNLP_PROTOCOL_OPTS setting is coming from for [2] and also understand how much memory is available for a possibly higher -Xmx256m?  

Just for comparison, the [1] job launches [3] which requests (Java max of) -Xmx2048m to be used for JNLP, of a possible OS/container/VM max memory limit of 2176Mi (as controlled via [4]).

Scott

_______________________________________________
cbi-dev mailing list
cbi-dev@xxxxxxxxxxx
To unsubscribe from this list, visit https://www.eclipse.org/mailman/listinfo/cbi-dev

_______________________________________________
cbi-dev mailing list
cbi-dev@xxxxxxxxxxx
To unsubscribe from this list, visit https://www.eclipse.org/mailman/listinfo/cbi-dev

Back to the top