Ship infrastructure before services, with environment approvals

A queue migration, a new cluster setting, a changed secret: the services that depend on them must not deploy first. When infrastructure and services live in different repositories, “first” is easy to get wrong, especially when someone has to approve the infrastructure change before it runs.

In a flow, the infrastructure box comes first and the worker that depends on it is connected on success. The worker only starts once the infrastructure deploy has finished and succeeded. If it fails, the worker is skipped, and nothing half-migrated goes out.

Kodmos is a GitHub App: runs are started by the app, and your environment protection rules still apply. When a job targets a protected environment with required reviewers, the run waits on GitHub, and Kodmos shows the deployment as Needs approval with the rest of the chain waiting behind it. Approving on GitHub lets it carry on.

name: Apply
on:
  workflow_dispatch:

jobs:
  apply:
    runs-on: ubuntu-latest
    # Required reviewers are set on this environment, in the repository's settings.
    environment: Production
    steps:
      - uses: actions/checkout@v4
      - run: ./scripts/apply.sh

Plan the change as an event with a deploy time, say 07:30, when traffic is low. The Platform team owns infra and worker, so they are the ones asked for branches and reminded before the deadline. Once the list locks, whoever is on at 07:30 only has to press Deploy; the environment’s reviewers approve the run on GitHub.

Everyone involved gets an email when the deploy starts and another when it has finished, with each project’s result. DevOps also hears about failures, and every step stays in the event’s activity for the post-mortem nobody wants to write.