Provide more metrics about druid-exporter

[iamabhishek-dubey]

Okay we will be reconsidering it

[iamabhishek-dubey]

You can list the status of the task by using PromQL, like:-

count(druid_tasks_duration{task_status=!""}) by (task_status)

[schmichri]

Means I have to add custom rules on prometheus-adapter for a metrics which should be standard. I propose you reconsider this.

[iamabhishek-dubey]

Please use this release https://github.com/opstree/druid-exporter/releases/tag/v0.9. metrics are added

druid_running_tasks
druid_waiting_tasks
druid_completed_tasks
druid_pending_tasks

[schmichri]

Looks working. Thanks a lot!

[Aka-shi]

@schmichri Can you share on how you plan to deploy HPA with only task metrics?

[schmichri]

@Aka-shi : I plan so scale a middlemanager (statefulset) when there are too many pending tasks. The metrics are thanks to @iamabhishek-dubey are available now.

[Aka-shi]

Hey, @schmichri I understand the scale up. What I am more concerned is the scale down, if you are deploying HPA and when the controller scales up the middle manager pods based on pending tasks, the pending tasks count itself would become zero(ideally) which would in some time trigger a scale down.

IMO, to properly scale up or down, we would also need the total worker capacity of the druid cluster and after that we can use this ratio sum(druid_pending_tasks + druid_running_tasks)/druid_worker_capacity_total. This would give a fair ratio to scale up/down. Thoughts?

[schmichri]

You are right. Didn't thought this through.