2.1.1.3. Lambda Deployment and Versioning
2.1.1.3. Lambda Deployment and Versioning
First Principle: Lambda deployments enable agile updates, while versioning and aliases ensure controlled rollouts and safe testing of new application code in production.
Lambda's versioning model lets you deploy updates safely — but only if you understand the relationship between $LATEST, versions, and aliases.
When you update a function, you're editing $LATEST — a mutable working copy. This is fine for development, but production should never point to $LATEST because any update immediately affects all callers. Instead, publish a version to create an immutable snapshot of your code and configuration. Version 1, version 2, version 3 — each is frozen and gets its own ARN.
Aliases are the key abstraction. An alias like PROD points to a specific version (say, version 5). Your API Gateway integration points to the alias, not the version. When you deploy version 6, you update the alias — and if something breaks, you point the alias back to version 5 instantly, without redeploying any code.
Packaging dependencies: Lambda never installs dependencies for you — a .zip package must already contain them. For Python, run pip install -r requirements.txt -t <dir> and zip that directory together with your code; for Java, build a single "uber" JAR with the Maven Shade (or Gradle Shadow) plugin. Layers are the alternative when several functions share the same dependencies.
For Node.js, the .zip holds the handler file (app.js) plus the installed node_modules/, and leaves out tests/ and other dev-only files to keep the package small.
The real power comes with weighted aliases for traffic shifting. Instead of sending 100% of traffic to version 6, you send 10% (canary) and monitor errors. If CloudWatch alarms trigger, CodeDeploy automatically rolls back the alias. This is the Lambda deployment pattern the exam tests most heavily — understand that traffic shifting requires aliases, not direct version references.
Scenario: You've developed a new version of your Lambda function and want to deploy it to production. You need to test it with a small percentage of user traffic first before a full rollout, and have the ability to quickly revert if issues arise.
⚠️ Exam Trap: A Lambda version is immutable — you can't change its code or configuration. An alias points to a version and CAN be updated. Traffic shifting between versions requires aliases, not direct version references.
