On this page4 sections
In a hypothetical cleanup, a team moves customer overrides out of a crowded configuration file. The file looks better on Friday. On Monday, a responder trying to explain an export has to open another service to find out which settings the worker used. The export behaves exactly as it did before. Finding out why now takes another lookup.
The KISS principle asks us to keep things simple. For configuration, start by counting the behaviors the team has to live with. A clean default helps the people who set up a service. Someone still needs to understand, test and recover every behavior the service promises to support.
The settings survived the move
Take two independent switches, each with an on and an off position. Leave the first one off and the second still has two positions. Turn the first on and there are two more. Ten independent switches give you up to 1,024 combinations by the same arithmetic. Some combinations may be blocked by the service’s rules, so count the ones it can actually reach and distinguish those it promises to support. The exercise tells you more than a line count ever could.
Read diagram description
The diagram shows a two-by-two grid. The columns are flag A off and on, and the rows are flag B off and on. All four cells are possible if the switches are independent. Apply the service’s rules before deciding which states it can reach and must support.
A customer override is a setting that takes precedence over the default for that customer. The original export still gets the override after the cleanup. If the worker can show its effective configuration, the values it used after applying defaults, customer overrides and deployment settings, the responder can trace its behavior back to the source. That makes an existing state easier to understand. Removing an unused override does different work: the team has one less state to understand in the first place.
Google’s configuration design guide discusses keeping ordinary settings simple while allowing deliberate overrides. That tradeoff makes sense here. Most customers can use the default, and the exceptions that remain should be easy to see when someone has to explain an export.
Find the people and data that still depend on the override
An expiry date gets someone to look at the override. The calendar cannot tell them whether the last customer moved. Look for remaining users, the team that can migrate them and the evidence that will make it safe to remove the setting. Two storage paths may look like clutter until you discover that one is how a customer gets old exports back. Keep a necessary recovery path visible and test the supported combinations it creates.
The transition has a configuration of its own. Suppose a job updates the override, then stops before it updates the dependent export schedule. Report those steps separately so the responder can find where the work stopped. A single failure label could send them back to repeat the update that already took effect. Even after every live customer has moved, check the old records, queued messages and scheduled jobs that can still ask a worker to use the previous behavior.
The old setting can meet new data
Follow an export through the proposed change. The worker reads the new setting and writes in a new format. Later, someone restores the old setting. Now the worker has the old configuration and a file made by the new behavior. The review needs to answer whether the worker can read that file and how an operator will recognize the combination if it cannot.
Use a canary rollout to exercise the changed behavior on a limited population. Keep track of which settings, customer cases and recovery paths it actually covered. A consequential combination left untested belongs in the risk registry when it needs a decision beyond the rollout.
Let the file get shorter when the work does
Move a small group first and watch both normal exports and recovery. Keep the former path while it can still serve as a valid fallback. Remove it when the agreed retirement conditions are met, including what happens to the data and jobs that still refer to it.
Until then, the most helpful simplification might be a longer screen. Give the responder the settings the worker used, the source of each value and a clear path back if the change goes wrong. When the last dependency on the override is gone, the shorter file will finally describe a simpler service.
Source context
Source links appear within this article. Read them alongside the author’s analysis and evaluate the guidance against your environment.
Report an error or outdated detail