Increase performance with snapstart/crac and client priming

[hamburml]

Hello,

I try to let my lambdas answer faster after a coldstart. I use snapstart for that. When using s3Client or dynamodbClient the lambda needs a bit time (maybe 1 or 2 seconds) before it can use the client.

I was able to let the whole lambda answer in round 200 ms with crac.

@ApplicationScoped
@Startup
public class Somename implements Resource {

    @Inject
    S3Client s3Client;

    @PostConstruct
    void init() {
        Core.getGlobalContext().register(this); // Register for CRaC lifecycle
    }

    void doSth() {

        // code which is used when lambda runs

    }

    @Override
    public void beforeCheckpoint(Context<? extends Resource> context) throws Exception {
        try {
            s3Client.listBuckets();
        } catch (Exception e) {
            log.info("listBuckets: " + e.getMessage());
        }
        log.info("before checkpoint");
    }

    @Override
    public void afterRestore(Context<? extends Resource> context) throws Exception {
        log.info("after restore");
    }
}

I do not care if an exception is thrown because listBuckets is not working when for example policies are not correctly set. I simply want the s3Client to be ready for action when the lambda gets restored via memory snapshot. The quarkus docu mentions client priming https://quarkus.io/guides/aws-lambda-snapstart and has an example for the DynamoDBClient which is build in a static-block. But if I do it like that I do not use the quarkus-specific ClientProducers in the runtime dependency.

Any recommendations how this can be done more elegant or is my approach sufficient enough?