Skip to content

README.md

Latest commit

 

History

History
70 lines (56 loc) · 3.1 KB

README.md

File metadata and controls

70 lines (56 loc) · 3.1 KB
Raw

Lambda Managed Instances example

This SAM application deploys Voker on AWS Lambda Managed Instances and verifies that one execution environment safely processes several invocations at once. The live probe checks concurrent request/response matching, invocation-scoped request and X-Ray metadata, isolated application errors, and cooperative work stoppage shortly before the Lambda deadline.

The function and capacity provider use Arm64. Lambda selects a supported Graviton instance type for the capacity provider. Although C9g instances are available through EC2 in some Regions, Lambda Managed Instances must separately support an instance type before it can be placed on an allow-list.

The stack is intentionally self-contained. It creates a VPC, three public subnets, an internet gateway, a Lambda capacity provider, and a published function alias. Public subnets are the simplest and lowest-cost connectivity choice for a short-lived development probe; use private subnets with NAT or VPC endpoints for production workloads.

Prerequisites

  • Go matching the version in go.mod
  • AWS CLI v2, AWS SAM CLI, and jq
  • An AWS Region where Lambda Managed Instances is available
  • Deployment permissions for CloudFormation, Lambda, EC2, IAM, S3, and lambda:PassCapacityProvider
  • iam:CreateServiceLinkedRole if the account has never created a Lambda Managed Instances capacity provider

Deploying the first published version can take several minutes. Lambda launches three managed EC2 instances by default for Availability Zone resilience, and those instances remain billable until the capacity provider is deleted.

Deploy and test

make validate
make deploy
make probe
make delete

The default stack name is voker-managed-instances. Region and credentials use normal AWS CLI resolution unless AWS_REGION and AWS_PROFILE are passed on the command. For example:

make deploy AWS_REGION=your-region AWS_PROFILE=your-profile STACK_NAME=my-voker-probe
make probe AWS_REGION=your-region AWS_PROFILE=your-profile STACK_NAME=my-voker-probe
make delete AWS_REGION=your-region AWS_PROFILE=your-profile STACK_NAME=my-voker-probe

Always run make delete after testing. The template configures a maximum of 30 capacity-provider vCPUs and eight concurrent invocations per execution environment. Override them with MAX_VCPU_COUNT and MAX_CONCURRENCY on the make deploy command when needed.

What the handler demonstrates

The handler uses atomic counters for shared process state and returns the event ID, Lambda request ID, trace ID, execution-environment ID, configured maximum concurrency, and observed concurrency. A real Managed Instances handler must apply the same concurrency discipline to all mutable globals, caches, clients, and shared /tmp paths.

Each invocation receives an independent context.Context. Managed Instances do not forcibly stop Go code when an invocation times out, so the example installs a 500 ms deadline guard and returns a typed error before Lambda's deadline. Real applications should choose a buffer large enough for their next unit of work and stop all side effects when the context is done.