hms-dbmi-cellenics / issues

This repository is used to report and track issues
1 stars 0 forks source link

Autoscale worker pod replicas from zero #77

Closed alexvpickering closed 6 months ago

alexvpickering commented 8 months ago

Background

ECS Kubernetes is typically ~50% of the cost of a low-usage Cellenics deployment (deployment without minimum of 1 user at all times). In staging environments, we avoid this cost by running pipeline jobs in batch (pay by usage) and keeping worker pod replicas=0. This creates an annoyance where we have to manually scale worker pods to replicas=1 in order to manually test a staged environment. It additionally means that developers can only test a staged PR in a deployment that they have access to (e.g. only HMS developers can check HMS staging environment and Biomage must re-stage). If this scaling is automated, both of the above frictions will be eliminated. It would also create the opportunity to eliminate idle worker/pipeline pods and the associated costs for low-usage production environments.

Proposal

PRs