CodingLevel 6
Dependencies and deploys
Easynot started
Mediumnot started
Hardnot started
Deploys, rollouts & flagsHow to push broken code to production without anyone noticing.3 minDependency & vendor failuresWhy a 3rd-party analytics API going down can take your entire checkout page with it.3 minBad data deployWhen code rollback doesn't save you because the database is already corrupted.3 minCanary deploy and the bad versionSend the new version a thin slice of traffic first, watch it like a hawk, and let the watchdog pull it back the moment it misbehaves.3 minCold storage retrieval timeoutsWhen asynchronous realities break synchronous expectations.3 minConfiguration errorsWhen perfectly good code is destroyed by a bad JSON file.3 minDependency CVEsYour app is mostly other people's code. When a vulnerability is published against one of those libraries, you're exposed, even if you've never heard of the library that pulled it in.3 minEnvironment misconfigurationWhen perfectly good code destroys the production database.3 minFeature store timeoutsWhen fetching data for a real-time ML prediction takes too long and crashes the user request.3 minML canary: the NaN trapWhy a single missing value can silently lobotomize an entire neural network.3 minModel serving cold startThe first request to a fresh replica pays for everything that hasn't loaded yet.3 minModel serving dependency fallbackWhen the model your service calls goes dark, you don't go down with it, you degrade to something simpler that still answers.3 minModel serving registry outageAlready-warm replicas keep serving from memory when the registry dies, it's the new pods that can't find out which model to load, so you can't add capacity exactly when you need it.4 minTLS certificate expiryA clock crosses the certificate's valid-until line, and one by one the handshakes start failing.5 minTokenizer version mismatchA model only ever sees numbers, feed it numbers from the wrong dictionary and it answers fluently to a question you never asked.3 minTrain-serve skewWhen the code you wrote in your Jupyter Notebook doesn't match the production backend.3 min