A Sidekiq queue per branch, dies with the branch

Five engineers, five PR branches, all touching the background-job code. The shared Sidekiq queue in staging gets polluted: one branch's malformed jobs poison another branch's tests. Test runs flake. People blame each other.

Each branch deploys to its own container. Each container has its own daemon-managed Sidekiq process configured in programs.json checked into the repo. The branch boots; Sidekiq boots with it as a daemon program with boot: true. The branch is closed; the container is destroyed; the Sidekiq daemon dies cleanly with no orphaned jobs in someone else's Redis.

The Redis backing each Sidekiq is a per-container instance, also a daemon program, also branch-scoped. Five branches, five queues, zero cross-pollination.

Branch died, queue died, jobs gone: no DELETE FROM sidekiq_queues required.