eclipse / jifa

🔬 Online Heap Dump, GC Log, Thread Dump & JFR File Analyzer.
https://eclipse.github.io/jifa/
Eclipse Public License 2.0
543 stars 96 forks source link

`---jifa.elastic-worker-idle-threshold=30`, does this parameter control the worker's survival time? #260

Closed mark8s closed 8 months ago

mark8s commented 8 months ago

I want to make the worker pod survive longer, but it seems that when I use this args configuration, it has no use. Did I use it wrong?

image

The startup parameters of my complete jifa-master are:

      args:
            - --jifa.role=master
            - --jifa.scheduling-strategy=elastic
            - --jifa.storage-pvc-name=jifa-pvc
            - --jifa.storage-path=/jifa-storage
            - --jifa.service-account-name=jifa-service-account
            - --jifa.elastic-worker-image=eclipsejifa/jifa:latest
            - --jifa.elastic-worker-idle-threshold=30
D-D-H commented 8 months ago

does this parameter control the worker's survival time

Yes.

But in the current implementation, master won't pass this config to elastic workers.

You can try to modify the logic in org.eclipse.jifa.server.service.impl.K8SWorkerScheduler#scheduleAsync.

https://github.com/eclipse/jifa/blob/61ec0e4049721ecaee04e9124e1f52850ef23cac/server/src/main/java/org/eclipse/jifa/server/service/impl/K8SWorkerScheduler.java#L114

D-D-H commented 8 months ago

Another way is to build a new worker image with config updated.

D-D-H commented 8 months ago

I'll file a PR to fix the problem that master cannot control all configs of elastic workers.

mark8s commented 8 months ago

@D-D-H ok,tks~