-
Bug
-
Resolution: Cannot Reproduce
-
Minor
-
None
-
Jenkins 2.7.3 (official docker image jenkins:2.7.3) - Docker host has 16GB of memory and runs nothing else
Slave (official docker image jenkinsci/jnlp-slave:latest) - Kubernetes hosts all have 48GB of memory
Kubernetes plugin 0.8
I'm running into an issue where the slaves dynamically spun up by the Kubernetes plugin randomly cause an "unable to create new native thread" exception during the slave.jar connection phase. As far as I can tell, the exception is throws on the Jenkins master, not the slave. Both the master and slave are running in the official docker containers (jenkins:2.7.3 and jenkinsci/jnlp-slave:latest, respectively).
The first thing I did was check the process limits from the "jenkins" user inside the Jenkins master container. The limits seem to be much higher than what should be needed for a server that only has 7 jobs (where everything runs on slaves). It usually gets to the point where a build for a job is created, but doesn't actually do anything. The job hangs with either no output or the same OOM exception.
I've included the process limits and relevant logs below. Is there anything else I can provide that would help solve the problem?
—
Jenkins master process limits from inside the container: https://gist.github.com/agunnerson-ibm/dccb6c8c2edc6a498f9c377d96f43481#file-process_limits_master-txt
Slave connection log: https://gist.github.com/agunnerson-ibm/dccb6c8c2edc6a498f9c377d96f43481#file-slave-log
slave.jar log: https://gist.github.com/agunnerson-ibm/dccb6c8c2edc6a498f9c377d96f43481#file-slave-jar-log
Hanging job log: https://gist.github.com/agunnerson-ibm/dccb6c8c2edc6a498f9c377d96f43481#file-job-log