DevOps · Cluster
Configuration Management
Desired state and environment parity.
- 01 Convergent State Application and Why a Run That Changes Nothing Still Costs Money How a desired-state tool decides whether a resource already matches, why that decision is a diff and never a comparison of files, and what idempotence actually guarantees.
- 02 Idempotency Keys and Command Resources That Cannot Be Described Declaratively Why some operations cannot be expressed as desired state, how guard and sentinel patterns approximate idempotence, and why the approximation leaks under concurrency.
- 03 The Dependency Graph Behind an Apply: Ordering Edges, Cycles, and Partial Failure How a configuration tool derives execution order from resource references, why the graph is not the file, and what a failed node leaves behind for the next run.
- 04 Environment Parity as a Byte-Level Claim: What Divergence Between Stages Actually Costs Why staging and production diverge even when both are built from the same configuration, which differences change behaviour, and how to make parity a measurable property.
- 05 How a Configuration Value Resolves: Precedence, Merge Modes, and Missing Keys The lookup rules that turn many layers of data into one value, why first-found and deep-merge hierarchies disagree, and why undefined and empty are different states.
- 06 Template Rendering Under Variable Input: Determinism, Escaping, and Secret Leakage Why a configuration template can produce a different file from identical inputs, how quoting inside structured formats fails, and where rendered secrets end up on disk.
- 07 Externalised Configuration and Immutable Artifacts: Where the Boundary Sits Why the same artifact should run in every environment, which configuration must move outside it, how the binding happens at startup, and what fails when the boundary is drawn wrong.
- 08 Drift Detection Without Reading Every Host: Sampling, Fingerprints, and Reconciliation How to find hosts whose state no longer matches their declaration, what a fingerprint over a state subtree can and cannot prove, and how detection frequency trades against read cost.
- 09 Break-Glass Changes: Overrides That Are Auditable and Expire How to let an operator change a host during an incident without permanently losing the declaration, why an override needs an owner and a deadline, and how the loop treats it.
- 10 Plan and Apply Across a Fleet: Partial Convergence and the Limits of Ordering Why a fleet-wide apply cannot be ordered like a single host, what the plan file captures and where it is stale, and how batches expose half-migrated intermediate states.