Standardize image management across control plane and proxy clusters

[blink-so]

Problem

During the Sept 30 Agentic Workshop, inconsistent storage/fetching of the non-GA Coder image (ghcr.io/coder/coder-preview) between the control plane and proxy clusters (Oregon + London) caused subdomain routing issues.

Root Cause

No standardized process for ensuring image consistency across multi-region deployment (control plane + 2 proxy clusters).

Requirements

• Document current image pull and storage mechanism for each cluster
• Identify why images diverged between control plane and proxy clusters
• Implement image version pinning/tagging strategy
• Create pre-deployment validation to verify image consistency across all clusters
• Add monitoring to detect image version drift
• Document image update procedure for multi-region deployments
• Consider migration strategy from non-GA preview images to stable releases

Success Criteria

• All clusters run identical Coder image versions
• Subdomain routing works correctly across all regions
• Automated validation catches image inconsistencies before user impact
• Zero subdomain issues during monthly workshops

Related

Sept 30 Workshop Postmortem