The JVM-Runtime is notorious for it's long, multi-seconds coldstarts. Workarounds in the form of lambda warm up magic are rather ugly.

With the addition of native lambda runtimes, we can use graalvm to build a native image which should be significantly faster.

=>

The native lambda generated by this repo with a 256 MB lambda needs roughly 150 ms for a coldstart.

Also used tech: gradle + terraform

• basic gradle setup
• terraform ec2 (for build, graalvm only works on *nix)
• graalvm task works
• create http client for lambda runtime interface (http + json parsing)
• implement lambda runtime interface
• terraform lambda using graalvm image
• replace manual ec2 build with codebuild from git repo and automatic lambda rollout

Currently, the lambda function gets created by terraform. On a clean slate (no state), the creation would fail because the expected zip won't be present inside the s3 bucket (no codebuild run yet). Therefore a fake zip will be used during the lambda creation. The codebuild job will call lambda:UpdateFunctionCode.

Alternatively, the codebuild job could also handle the creation of the function, if it is not yet created. Both approaches aren't without flaw.

• fighting-cold-startup-issues-for-your-kotlin-lambda-with-graalvm
• AWS Lambda Runtime Interface
• A simple native HTTP server with GraalVM
• HttpServer.kt
• palantir/gradle-graal

• compute_type = "BUILD_GENERAL1_SMALL" keeps crashing, possible because it needs more RAM -> BUILD_GENERAL1_MEDIUM
• Maybe you forgot to apply terraform?!