Deployment is ansible-playbook run by hand with an interactive vault password. Choose how far automation goes during Phase 0.
Recommendation: keep Ansible as the source of truth and add human-triggered deploys from Gitea (workflow_dispatch) for selected apps. No automatic production deploys and no GitOps controller yet.
Work (Claude): comment with options, including how secrets would reach a deploy job. Decision (Eric): choose.
Acceptance: decision recorded and D4 and D5 updated.
Deployment is `ansible-playbook` run by hand with an interactive vault password. Choose how far automation goes during Phase 0.
**Recommendation**: keep Ansible as the source of truth and add human-triggered deploys from Gitea (`workflow_dispatch`) for selected apps. No automatic production deploys and no GitOps controller yet.
**Work (Claude)**: comment with options, including how secrets would reach a deploy job.
**Decision (Eric)**: choose.
**Acceptance**: decision recorded and D4 and D5 updated.
claude
added this to the CI/CD 4: Continuous delivery milestone 2026-11-09 17:07:10 +00:00
Decision recorded (Eric, 2026-11-10): the recommended model.
For Phase 0, eom.dev delivers like this:
Ansible stays the source of truth. Roles and software-infrastructure define every deployment, and anything a workflow deploys goes through the same playbooks and roles.
Deploys are human-triggered. Selected apps get a workflow_dispatch deploy workflow that takes an explicit input (for example an image SHA for www). Nothing deploys to production on push or merge.
No GitOps controller yet. Argo CD or Flux stays an evaluation item in L2 (#79).
Hand-run ansible-playbook remains valid for everything without a workflow, and as the fallback when a workflow is unavailable.
How secrets reach a deploy job. These are the options I considered. D4 (#72) implements one, and Eric creates and places all credentials there.
Recommended: credentials live with the deploy runner, not in Gitea. A separate deploy runner (D4) runs in its own namespace. Its vault password and its narrowly scoped Kubernetes service account are provided to that runner by Eric, outside Gitea, so the shared CI runner never has them and no workflow on it can request them. Only approved workflows can target the deploy runner's label.
Gitea Actions secrets scoped to one repo, used only by jobs on the deploy runner. Simpler to set up, but the secret is stored in Gitea and is available to any workflow in that repo, so it relies more on review of workflow changes.
An external secrets manager (for example Vault or SOPS with a KMS). It's the most flexible, but it's a new service to run, which is too much for Phase 0.
Under every option: no agent reads or handles credentials, deploy logs must not print them, and the deploy identity can change only its target namespaces.
Follow-ups updated: D4 (#72) and D5 (#73) have comments with what this decision means for them. Closing this as decided.
— eom-cicd-roadmap
**Decision recorded (Eric, 2026-11-10): the recommended model.**
For Phase 0, eom.dev delivers like this:
- **Ansible stays the source of truth.** Roles and `software-infrastructure` define every deployment, and anything a workflow deploys goes through the same playbooks and roles.
- **Deploys are human-triggered.** Selected apps get a `workflow_dispatch` deploy workflow that takes an explicit input (for example an image SHA for www). Nothing deploys to production on push or merge.
- **No GitOps controller yet.** Argo CD or Flux stays an evaluation item in L2 (#79).
- **Hand-run `ansible-playbook` remains valid** for everything without a workflow, and as the fallback when a workflow is unavailable.
**How secrets reach a deploy job.** These are the options I considered. D4 (#72) implements one, and Eric creates and places all credentials there.
1. **Recommended: credentials live with the deploy runner, not in Gitea.** A separate deploy runner (D4) runs in its own namespace. Its vault password and its narrowly scoped Kubernetes service account are provided to that runner by Eric, outside Gitea, so the shared CI runner never has them and no workflow on it can request them. Only approved workflows can target the deploy runner's label.
2. Gitea Actions secrets scoped to one repo, used only by jobs on the deploy runner. Simpler to set up, but the secret is stored in Gitea and is available to any workflow in that repo, so it relies more on review of workflow changes.
3. An external secrets manager (for example Vault or SOPS with a KMS). It's the most flexible, but it's a new service to run, which is too much for Phase 0.
Under every option: no agent reads or handles credentials, deploy logs must not print them, and the deploy identity can change only its target namespaces.
**Follow-ups updated:** D4 (#72) and D5 (#73) have comments with what this decision means for them. Closing this as decided.
— eom-cicd-roadmap
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Deployment is
ansible-playbookrun by hand with an interactive vault password. Choose how far automation goes during Phase 0.Recommendation: keep Ansible as the source of truth and add human-triggered deploys from Gitea (
workflow_dispatch) for selected apps. No automatic production deploys and no GitOps controller yet.Work (Claude): comment with options, including how secrets would reach a deploy job.
Decision (Eric): choose.
Acceptance: decision recorded and D4 and D5 updated.
I agree with your recommendation. Please proceed with that plan.
Decision recorded (Eric, 2026-11-10): the recommended model.
For Phase 0, eom.dev delivers like this:
software-infrastructuredefine every deployment, and anything a workflow deploys goes through the same playbooks and roles.workflow_dispatchdeploy workflow that takes an explicit input (for example an image SHA for www). Nothing deploys to production on push or merge.ansible-playbookremains valid for everything without a workflow, and as the fallback when a workflow is unavailable.How secrets reach a deploy job. These are the options I considered. D4 (#72) implements one, and Eric creates and places all credentials there.
Under every option: no agent reads or handles credentials, deploy logs must not print them, and the deploy identity can change only its target namespaces.
Follow-ups updated: D4 (#72) and D5 (#73) have comments with what this decision means for them. Closing this as decided.
— eom-cicd-roadmap