I've been in the game for a decade now using AWS Lambda, and I recently had a discussion with my Engineering Manager who claimed that if a Lambda function isn't used for a certain period (around 30 minutes), AWS might remove its image, causing cold starts that can take up to 15 minutes to respond. I argued that cold starts max out at a few seconds under normal conditions unless there's some crazy workload. He insists he's had extensive experience with this and that it's accurate. What's the real deal here?
3 Answers
You can simply verify this yourself. When people get into these disputes, it’s often best to focus on hard data rather than opinions. If cold starts are lengthy on your end, diagnose this properly. If not, don’t waste time stressing over it.
I've dealt with cold start issues in the past, but it was more like 30 to 90 seconds, and we just pounded the API to keep things warm. AWS has made huge strides in reducing these delays, and I’ve never seen it take that long unless something else was wrong. Your manager needs the latest info.
Make sure to check Lambda SnapStart if this is a concern for him! But honestly, from my experience, cold start delays shouldn’t reach anywhere close to 15 minutes unless there are outside factors at play. If anything, the cold starts have significantly improved over time.
LOL, that’s exactly what I think! I’m pushing for Lambda too, since K8s has been over-provisioned. My boss thinks we’d incur extra fees just to prevent AWS from retracting our Lambda images during off-peak hours. It’s all just misinformation, but I don’t want to step on any toes in my first week.