Dmitri Volkov

Configuration is source code and should be reviewed like it, which is a change of habit rather than a change of tooling. That makes the file readable as an override list but means a value set twice is a bug rather than a style question. The related problem is a value set in the file and also set in the environment, where precedence is a property of the loader and not documented anywhere you will look. Types are the other half. Most loaders infer a type from the string, so a version number of a certain shape, a value that looks like a date and a value that looks like arithmetic are all cases where the inferred type is not what was meant. Quoting fixes it and is worth doing by default. Structured configuration has the same parsing hazards as any structured document, and the YAML ones apply in full. Drift is the thing worth measuring. Compare what is configured against what is running, per environment, and the difference is a list of changes nobody deployed. Doing it on a schedule catches the manual fix somebody made at 2am and never wrote down. Secrets belong in a secret store, not in a file in the repository, and I cover what happens when they are already there. A configuration file nobody has looked at since it was written is the most common source of a drift report, and the diff is what turns it from a mystery into a line. Generating configuration from one source and editing it by hand afterwards is how two environments stop agreeing, and the hand edit is always the one nobody wrote down.