Testing changes on Amazon ECS typically means building and pushing an image, registering a task-definition revision, waiting for deployment, and repeating the process when something fails. That delay undermines coding agents’ ability to make rapid changes and test them. MetalBear’s new mirrord for ECS is designed to let developers, CI jobs, and coding agents run modified service code locally while using the context of a live staging task: its environment variables, network access, and IAM task role. The local code can reach resources such as private databases, queues, and internal APIs without creating a separate test environment.
Setup requires adding a small container to each service’s task definition and deploying that configuration once. When a task starts, the container loads mirrord into the app container without changing the application image. The task and the developer’s machine each make outbound HTTPS connections to a sessions-manager, hosted by MetalBear or run in the customer’s Kubernetes cluster; this avoids opening inbound ports. Requests sent to the staging service with a matching session header are routed to the local code, while other requests continue to the deployed service. Each person or agent can therefore test a separate version against the shared staging environment without disrupting other users. After setup, changes can be tested by rerunning the service rather than rebuilding and rolling it out. A Terraform-based sample demonstrates the workflow with an ECS service and private RDS database.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.