Engagement went up, and it was because the app got slower
Time-in-app rose because loading took longer. A metric that improves under a regression is measuring the wrong thing.
practices
42 posts
The lock held across a network call, and the deadlock that followed
A mutex held for microseconds is fine. Held across an unbounded wait, it serialises your service.
The rollout with no off switch, discovered at 2am
A flag that gates the new path but not the data migration is not a kill switch.
The config change that was really a deploy, and had none of the safeguards
Config bypasses code review, CI, canaries and rollback. It changes behaviour just as much.
The N+1 query that looked fast in development
Ten rows locally, ten thousand in production. The query count scales with data.
The cache that returned someone else's data
A cache key missing the tenant is a data leak with a 100% hit rate.
Every timeout you set independently adds up to one that is wrong
Three services with 30-second timeouts give the user 90 seconds.
An estimate is a distribution, and you reported the mode
Two weeks means the most likely case. What anyone planning around you needs is the spread.
The code review that should have been a conversation
Forty comments on a pull request is a design disagreement wearing a review costume.
The most valuable pull request that quarter deleted 4,000 lines
Dead code is not free. It is read, maintained, migrated, and it shapes decisions.
Idempotency is a product decision before it is a technical one
Deciding what counts as the same request is a business question. The key is the easy part.
Logs are for the questions you have not thought of yet
A metric answers a question you already had. The difference is cardinality.