Skip to content

Latest commit

 

History

History

kubernetes

README.md

Java vertical scaling test

Nodes

This addon assumes that there are 6 nodes in the cluster. Nodes are labeled in a special way so that Kubernetes schedules 1 Java pod per node. Currently, the addon does not check if there are enough nodes, so if nodes count is under 6, one of the pods won't be scheduled.

Also, Jelastic nodes displayNames are modified to make statistics analysis simpler.

Deployments

This addon creates 6 K8s deployments in java namespace. Deployments are defined in .yaml files and can be edited if necessary.

The addon processes retrieved yamls and replaces placeholders with desired values (at this moment only memory limits and JAVA_OPTS). You can make more things configurable is necessary.

curl ${baseUrl}/deployments/adoptopenjdk.yaml | sed "s/%JAVA_OPTS%/${globals.common} ${globals.adoptopenjdk}/g" | sed "s/%MEM_LIMIT%/${settings.adoptopenjdkLimit}/g" | kubectl apply -f -

Memory Limits

The default memory limit is 3Gi. You can change it for each pod in ${settings.jdkFlavor}:

- type: string
  name: zulujdkLimit
  caption: Zulu Pod Mem Limit
  default: 3Gi
  inputType: hidden

Remove inputType: hidden if you want to be able to define memory requests for containers on UI.

JAVA_OPTS

There are common JAVA_OPTS defined in {globals.common} and JAVA_OPTS defined for each individual pod - $globals.jdkFlavor

globals:
  common: -Xmx3g -Xms32m -XX:+UseCompressedOops
  zing: -XX:-AutoTuneResourceDefaultsBasedOnXmx
  openjdk: -XX:+UnlockExperimentalVMOptions -XX:+UseZGC -XX:ZUncommitDelay=1 -XX:ZCollectionInterval=30
  openj9: -XX:+IdleTuningCompactOnIdle -XX:+IdleTuningGcOnIdle -XX:IdleTuningMinIdleWaitTime=1 -Xjit:waitTimeToEnterDeepIdleMode=1000
  adoptopenjdk: -XX:+UnlockExperimentalVMOptions -XX:+UseShenandoahGC -XX:ShenandoahGCHeuristics=compact
  liberica: -XX:+UseG1GC -XX:G1PeriodicGCInterval=10k
  zulujdk: -XX:+UnlockExperimentalVMOptions -XX:+UseEpsilonGC

Redeploy

When this addon is repeatedly applied to a Kubernetes cluster, all existing deployments in java namespace are forcefully deleted, thus, there is no need to manually clean up resources before a repeated deployment.

Monitoring

The addon assumes that Grafana is installed in a cluster and is available at ${env.url}/grafana. Several dashboards are imported to track Java pods memory utilization.

Once installed, the addon will print URL and names of dashboards to check out.