A routine database failover lasted 20 seconds, but our API stayed down for 41 minutes. Three layers each made up to three attempts with no delay, multiplying load on the database by 27. Here is the math, the fix and how we proved it.
Running the test suite inside our app container dropped every table in the development database. Docker Compose injected DB_DATABASE as a real environment variable, and phpunit.xml's env entries do not override existing variables. Here is the full precedence chain and three layers of defence.
A low but stubborn rate of 502s at the load balancer and socket hang up errors between services turned out to be one race: Node closed idle keep-alive connections after 5 seconds while its callers kept reusing them.
Order dates rendered fine locally but threw hydration errors for users in Asia and Europe. The server formatted dates in UTC and en-US, the browser in local time. Here is how we found it and fixed it for good.
After a harmless-looking Express refactor, every Stripe webhook failed verification and paid bookings stayed pending. A global JSON parser consumed the body before constructEvent saw it. How we found it, fixed the route and replayed the missed events.
Our fraud model scored 0.91 AUC on holdout data and close to random in production. The serving service re-implemented preprocessing in another language: the scaler was refit on each batch, one-hot categories came in a different order and text was not lowercased. Shipping the pipeline with the model fixed it.
A practitioner's reading of AWS's official summary of the October 19-20, 2025 us-east-1 disruption: a race in DynamoDB's DNS automation left an empty record, and recovery cascaded through EC2, NLB and Lambda. What happened per the report, and what we changed.
Our support agent started failing with a 400 on roughly one conversation in twelve. The model had begun returning several tool calls in one turn and our loop only answered the first. Answering every tool_call_id in order fixed it.
We switched our semantic search to a cheaper 1024-dimension embedding model and every insert and query failed against the 1536-dimension column. Worse, the half-finished fix would have mixed incompatible vectors. How we migrated with versioned columns, a backfill and dual reads.
A new model that wins offline can still lose in production, through skew, latency, or effects on users that offline data cannot show. This guide covers a staged rollout for models: offline gates, shadow traffic, canaries on business metrics, A/B tests when needed, and fast rollback.