1667210 - Create persistent clusters for prio projects

Status: RESOLVED INACTIVE

(Data Platform and Tools Graveyard :: Operations, task)

Description

[Anthony Miyaguchi [:amiyaguchi]]

See bug 1664582 for more details. It would help with the development process to have a persistent kubernetes cluster, since startup and shutdown of the cluster account for most of the time taken in a job.

The new GKE cluster definition is defined here: https://github.com/acmiyaguchi/telemetry-airflow/blob/d13f4c0d9a76bf053a1ffee9debb75a45db74dce/dags/utils/gke.py

It contains two node-pools. The first is a "baseline" node-pool for all of the k8 services. Then there is a "burstable" autoscaled nodepool with a no-schedule taint. The taint helps scale the node-pool to 0.

For prio-a and prio-b, the instance type for the burstable node-pool should be initially set to n1-standard-16 with 500gb of pd-standard disk. For prio-admin, it should be fine to use n1-standard-1 with 100gb pd-standard disk.

Comment 1

[SuAnn [:SuAnn]]

Please create a JIra issue if this is still required.