My bestie Claude says:
Replace complex ClusterIP + NEG configuration with simple
LoadBalancer service to improve reliability with GKE Autopilot. The
previous NEG-based setup had synchronization issues when Autopilot
moved pods between nodes/zones, causing backend service updates to
lag behind. LoadBalancer service automatically handles pod movement
and provides a dedicated TCP load balancer that's better suited for
SSH connections.
After ignoring the service for a few days, connections wouldn't work anymore since Autopilot moved the Deployment pod around to different zones, which didn't also update the backing NEGs
This approach was recommended to resolve that issue.
Signed-off-by: Jason Hall <jason@chainguard.dev>
