Unit 14.00: Configuration that differs by environment

Configuration differs by environment, and none of it belongs in a branch in the code.

Six settings, six differences

Development and production values for each, with the reason.

The code lists them.

CONFIG = {
    "app_name":         ("classifier", "classifier", "same everywhere"),
    "model_name":       ("fake-model", "real-model", "fake in dev and CI"),
    "log_level":        ("DEBUG",      "INFO",       "quieter in production"),
    "reload":           ("on",         "off",        "never on in production"),
    "workers":          ("1",          "4",          "shared state must be external"),
    "daily_budget_usd": ("0.10",       "50.00",      "a mistake in dev is cheap"),
}
print(f"{'setting':20} {'development':14} {'production':12} note")
for key, (dev, prod, note) in CONFIG.items():
    print(f"{key:20} {dev:14} {prod:12} {note}")

print("""
Every one of these comes from the environment and none from a branch in the
code. `if ENVIRONMENT == "production"` scattered through handlers is how
development behaviour reaches production unnoticed.

The fake model as the development default means a misconfigured environment
fails cheaply rather than expensively.
""")

Every one comes from the environment. if ENVIRONMENT == "production" scattered through handlers is how development behaviour reaches production - the branch is right in every place except the one somebody forgot.

The fake model as the development default means a misconfigured environment fails cheaply rather than expensively.

The mistake this prevents

The mistake is a single config file committed with production values and overridden locally. The production credentials are then in the repository, and a local override that is forgotten reaches production instead.

Takeaway

Every environment difference is a setting read from the environment, not a branch in the code. Default to the safe, cheap value.

Mark complete