ADR 0009: Use one production workload state¶
Status: Accepted Decision date: 2026-07-22 Owners: Workload owner and architecture owner
Context¶
The current workload is operated by a small team and only one persistent production environment is required. State must still be remote, locked, recoverable, and isolated from bootstrap trust.
Purpose and decision drivers¶
Avoid environment and cost sprawl while preserving safe infrastructure concurrency and recovery.
Decision¶
Maintain one production workload state. State/OIDC bootstrap has a separate administrative lifecycle and cannot be destroyed or privilege-expanded by routine workload apply. Pull-request previews and local/emulator tests provide nonpersistent validation.
Consequences and accepted risks¶
- There is no persistent staging environment with production fidelity.
- State backend and OIDC bootstrap require documented recovery and break-glass procedures.
- Production applies are serialized and environment-protected.
- A future staging state is a deliberate new decision.
Alternatives considered¶
- Dev/stage/prod states now: rejected due to cost and operational need.
- Local state: rejected due to concurrency, recovery, and secret risk.
- One state including its own mutable trust: rejected because routine apply could endanger recovery.
Validation and revisit triggers¶
Test locking, version recovery, access separation, and reviewed-plan application. Revisit when change volume, contributors, or release risk justifies persistent staging.