AWS Lambda Warm Starts Can Introduce Hard-to-Catch Bugs in Production

AWS Lambda reuses execution environments across invocations in what is known as a warm start, which improves speed and reduces cost by avoiding repeated initialisation. While placing clients like DynamoDB outside the handler is a recommended practice for reusing connections, it also means mutable variables and cached values from a previous invocation persist into the next. Code that assumes a clean slate on every run will behave correctly during manual testing — which typically triggers cold starts — but may produce subtle bugs under real production traffic. The root issue is that warm starts skip initialisation entirely and jump straight to the handler, carrying over any leftover state. The general guidance is to keep shared clients and connections outside the handler, while ensuring any request-specific or time-sensitive values are initialised fresh inside it.