AWS re:Invent 2019: [REPEAT 1] Best practices for AWS Lambda and Java (SVS403-R1)

In this session, we follow a customer’s journey as they optimize an AWS Lambda function written in Java to meet their cold start time requirements. We start from a simple yet slow PoC and walk through all of the changes, tricks, and trade-offs we made to reduce the cold start time by over 70%. Finally, we explore new technologies such as Quarkus and GraalVM that can make Java even faster in Lambda.

[totsubo]

Worth it's weight in gold! I'm suffering from this issue and this talks helps me understand why.

[nazeelak]

Using Java on Lambda, and how GraalVM and its AOT compilation feature help, and advantage of Micronaut framework explained in depth, Thanks for sharing.

[sarangkumar]

I don't understand one thing, in the talk i see that an initialisation time(init_duration) gets higher by moving code outside of the handler and then handler code execution time(duration) gets lower which will not have impact on overall time duration, means if we do the reverse by making handler time higher than initialisation time still the entire duration would be same. So if we do either of these, the cold start would be the same.

[UCTNXPiSlAhgUyGlBwg]

I have tried aws 2.0 library. the conclusion is 1.0 slightly faster than 2.0

[jorgevillarreal]

Serverless: Pay only for what you use and don't worry about maintaining servers where your code will be running.

But.... not all of your system components should be deployed in a serverless fashion.... I would say if you need to go through all of this to deploy that particular component, the benefits of serverless don't seem that attractive anymore.

Make sure to understand your workflow and use a variety of technologies.