aws-lambda

sumak + AWS Lambda

Lambda handler that writes an event to Postgres. The example is small on purpose — Lambda-specific plumbing (IAM, API Gateway, CDK/Terraform) is out of scope. What's here is the parts that matter for sumak:

1. Module-level pool. The new Pool() and sumak() call happen once per Lambda container, not once per invocation. A fresh pool per request would exhaust the RDS connection limit almost immediately.
2. max: 1. Lambda containers are single-threaded; one connection is the right default. It also acts as a natural per-function concurrency limiter.
3. Short idle timeout. RDS kills long-idle connections; Lambda freezes the process between invocations. A 10-second idle timeout keeps the pool from handing you a dead socket after a long quiet period.
4. AbortSignal from Lambda's deadline. context.getRemainingTimeInMillis() gives you the remaining budget. We convert it into an AbortSignal so an open query gets cancelled server-side rather than hitting Lambda's 15-minute kill.

Cold-start notes

sumak's initialisation is pure TypeScript with no runtime code generation, so there's no per-cold-start compile step. The compile-time benchmark in the main repo shows query-build overhead at sub-microsecond levels — that's what shows up on your cold-start trace.

If you're running on provisioned concurrency and want to be extra sure the first request doesn't trip any lazy-initialisation, add a warmup line at module scope:

db.selectFrom("events").select("id").limit(1).compile() // primes AST paths

Deploying

Use whatever your team prefers — SAM, CDK, Serverless Framework, Terraform, or plain zip + aws lambda update-function-code. This example deliberately doesn't pick one; the handler is what counts.