Zero-downtime k8s deployments behind ALB controller¶
This post dives into the initial setup for k8s releases as well as finally solving a long-present issue that has been present when rolling out changes.
The setup¶
Our setup involved the AWS ALB ingress controller which managed and provision respective loadbalancer resources based on events in k8s, and exposing our services to the world by creating said ingresses. The ingress controller would then update target routes as pods get cycled in and out. Sounds like a plan right?
The problem¶
Even though we expected release rollouts to be zero-downtime, when performing releases we noticed that if we hit the services at the right time, there was a few second window where we would see 502 Bad Gateway!
