Lambda

Lambda

Most popular serverless compute service in AWS (maybe also in the world)

• About executing code in reaction to a certain event
  • can have multiple events
• can configure various aspects about this execution
• a regional service, related: EC2 VS Lambda
• Memory allocation
  • You can allocate memory from 128 MB to 10,240MB (10 GB), in 1 MB increments
  • A you increase the memory allocation, AWS proportionally increases CPU power / network bandwidth / disk I/O
• Billing
  • $\text{Allocated memory (GB)}\times \text{Execution time (s)}$
    • Gigabyte-seconds (GBS)→ core metric AWS uses to calculate compute cost of lambda
    • If you allocate 1024 MB (1 GB) of memory & function runs for 500 milliseconds (0.5 seconds) →> billed for 0.5 GB-seconds
  • more memory = more CPU power → more cost
  • AWS lambda power tuning tool
    • devs run data-driven tests against their functions using this to find this exact sweet spot between cost and execution speed
• Limitations
  • 15 minutes → if the request is 15 min 1 sec, then it will timeout
  • 250MB limit → the total size of files (custom code + dependencies + libraries + lambda layer etc) cannot exceed 250MB
• Removing unused code
  • Lambda has hard limits on deployment sizes, so u sometimes remove code from libraries manually

Diagram

• diagram

Code

• functions
  • the things that contain your code
  • when you create a function (when making a lambda instance) you can create one or choose a pre-provided one
• you can upload the code (zip file, docker file) or define (write) the code in the Lambda management console
• choose supported programming language (in theory lambda supports all languages) & advanced setup (ex. environment variables)
  • ex. Node.js, ☕Java, Rust, C++, Golang (recommended)

Event

• Choose one of the many supported event sources (ex. file was uploaded to S3)
  • there is a broad variety
• filtering - certain variations of a given event (ex. choosing only .jpg for event when file is uploaded to S3)

Configuration

• timeout, like how long the code might run at most
  • you pay for the time your code is being executed
• choose architecture - x86_64 or arm64
• attach EFS (Elastic File System) as needed
• Attach an execution role for permissions
  • the code executed can have certain permissions
• You can connect your lambda function to a VPC
  • you can give ur code access to a VPC in case it needs more communication
• configure the memory that should be available
  • will impact the pricing

Example

• You could create a new Lambda and create event to S3 bucket
• Give your function the appropriate permissions to access the bucket
• Then you could have code that extracts data from the bucket when event is triggered

Monitoring

• in a function u create, u also have monitoring capabilities enabled out of the box → uses CloudWatch behind the scenes
• In configuration options, you can even enable extra monitoring, like XRay
• related: Monitoring workloads