I was wondering whether someone here shared the same experience. I have a Python Lambda which runs some “simple” requests for recommending images based on a vector database (qdrant). The package itself already has a size of 65MB (e.g. qdrant client and firebase). Although I turned on provisioned concurrency I still have cold starts of roughly 5 seconds. I even created a dummy lambda which has no imports and responds a 200 immediately and has the same packages installed - still 5 sec cold start. Using loggings I triple checked that provisioned concurrency is utilized. I am thinking about a complete rewrite in TypeScript of the Lambda to handle these cold starts. Do you have the same experiences regarding Python Lambdas? Is there a neat trick I am not aware of? Continue reading...
