[jonathanfoster]

I have a production service running on AWS Lambda and I haven't run into any major challenges. The Lambda service is responsible for authenticating to a downstream third party service and proxying requests along with an access token. I would consider this a simple use case.

CloudWatch have provided me all the visibility necessary to troubleshoot issues. I think the important thing here is to have a good logging strategy (logs are only as good as what you put in them). In my case, I made sure info messages were logging for the start and end of use cases (e.g., "reseting password", "password reset successfully"), warn messages for non-fatal errors (e.g. "username not found"), and error messages for fatal errors (e.g. "unable to connect to database").

The only frustrating limitation I've run into is when the Lambda function times out before receiving a response from the downstream service. At one point, the downstream service was having major performance issues and responses times were crazy high. This meant I couldn't get a response code and had to run the downstream calls locally to troubleshoot.

Performance is not great (most requests are in the 400-500 ms range), but it's more than adequate for my use case. A large portion of the response time is likely due to the downstream service, but there are cold starts that spike response time way out of normal range.

Overall, I'm really happy with AWS Lambda and it's definitely top of my list when taking on a new project. I'm really interested in experimenting with AWS Mobile Hub in the future. It doesn't get much better than one-stop serverless shopping.