Skip to main content

Lambda Limits

Knowing the hard structural walls of the platform is what separates a certified developer from someone guessing in production. 📈🛑

AWS Lambda scales like an absolute beast, but it operates inside strict physical guardrails. If you try to force a workload past these boundaries, the control plane will drop a hard exception and halt your task. For the DVA-C02 exam, knowing these numbers by heart lets you instantly identify when to skip Lambda and deploy straight to AWS Fargate, ECS, or EC2.

Key Takeaways​

🏎️ 1. Execution Runtimes (In-Flight Boundaries)​

When your code is actively executing inside a Firecracker microVM instance, these constraints are enforced per invocation:

  • Memory Allocation Slider: Ranges from 128 MB up to 10,240 MB (10 GB).

  • The Modern Incremental Increment: While older documentation used 64 MB increments, the platform allows you to slide your memory configuration in granular 1 MB increments.

  • vCPU Scaling Law: You do not control CPU directly. At exactly 1,769 MB, your function gets the compute equivalent of one full, dedicated vCPU core.

  • Hard Timeout Gate: Maximum execution time is strictly capped at 900 seconds (15 minutes). If your job needs 16 minutes to compile a dataset, Lambda is a bad choice—migrate to Fargate.

  • Environment Variable Payload: The combined total size of all your configuration environment strings cannot exceed 4 KB. If you need to pass massive JSON configs or high-volume parameters, you must pull them at runtime from AWS Systems Manager Parameter Store or Secrets Manager instead.

  • Local Ephemeral Storage (/tmp): Need to drop a heavy binary or scratchpad dataset to disk? The local ephemeral storage defaults to 512 MB for free, but you can configure it all the way up to 10 GB.

📦 2. Deployment Packages (Payload Material Storage)​

When you are zipping code or shipping artifact files, the AWS control plane enforces these delivery limits:

  • The Direct Upload Zip Limit (Compressed): 50 MB. If your zipped artifact file is 51 MB, a direct CLI push fails. You must upload the archive into an Amazon S3 Bucket first and reference it in your deployment.
  • The Absolute Extraction Ceiling (Uncompressed): 250 MB. No matter how you deliver your asset (Direct Zip, via S3, or stacked across up to 5 Lambda Layers), the total unzipped footprint inside the container filesystem cannot cross 250 MB!
  • The Container Image Loophole: If you deploy your serverless function via Lambda Container Images (Docker artifacts pushed to Amazon ECR), the 250 MB rule is completely vaporized. You get a massive container footprint capacity of up to 10 GB!

📊 3. Absolute Regional Default Quotas​

These quotas apply at an account level per individual AWS region:

  • Default Concurrency Threshold: 1,000 concurrent executions running simultaneously across all functions in your account. If a single unreserved function spikes and eats up all 1,000 units, your other services will get throttled (HTTP 429 Rate Exceeded). This is a soft limit—you can open an AWS support ticket to scale this pool up into the tens of thousands!

🔲 4. The Quick-Reference Sizing Grid​

Parameter Metric VectorPlatform System Boundary LimitIs It Flexible/Adjustable?Exam Architectural Failure Pattern
Max Timeout Window15 Minutes (900 seconds)❌ Hard Platform CeilingScenario: Heavy video encoding taking 25 minutes. Fix: Move to ECS/Fargate.
Max Memory Allocation10,240 MB (10 GB)❌ Hard Platform CeilingScenario: Data analytics processing job requiring 16 GB RAM. Fix: Move to EC2/Glue.
Environment Variables4 KB Total❌ Hard Platform CeilingScenario: Storing a massive collection of raw app configurations. Fix: Fetch from SSM.
Uncompressed Zip Size250 MB (Includes Layers)❌ Hard Platform CeilingScenario: Heavy data science libraries bloating your deployment. Fix: Use Container Images (10 GB).
Default Concurrency Pool1,000 Concurrent UnitsAdjustable via SupportScenario: High-volume API spikes freezing internal reporting jobs. Fix: Apply Reserved Concurrency.

Exam Tips​

  • The 30-Minute Job Trap: If an exam question claims a corporate developer wants to migrate a cron job that processes data for exactly 30 minutes onto a standard Lambda function to optimize costs—look for the answer that rejects the design. Lambda will kill the execution after 15 minutes flat. Suggest AWS Fargate tasks triggered via an Amazon EventBridge Schedule rule instead.
  • The Massive Machine Learning Package: If a scenario mentions a developer trying to push an advanced Python image classification script that uses heavy pre-trained models, causing the local zip code package to expand to 400 MB—look straight for Lambda Container Images via Amazon ECR. Containers unlock that 10 GB footprint limit and cruise past the 250 MB zip barrier smoothly!