ALB Multi-Value Headers

  • By enabling multi-value headers, values share the same key in HTTP headers and query parameters will be send as an array in AWS Lambda event
1
http://example.com?key=value1&key=value2
1
"multiValueQueryStringParameters": {"key": ["value1", "value2"]}

Invocation Types

  • Synchronous invocation
    • API Gateway, ALB
  • Asynchronous invocation
    • S3, SNS, EventBridge

Event Source Mapping

  • When triggered from sources that records need to be polled
    • SQS & SQS FIFO queue
    • Kinesis Data Streams
    • DynamoDB Streams
  • Synchronous invocation

For Kinesis & DynamoDB

  • Use batch windows to accumulate records before processing
  • You can process multiple batches in parallel
    • Up to 10 batches per shard
    • One partition key per batch
    • Processions within a batch are in-order

For SQS

  • Lambda uses long polling to retrieve records
  • Number of batches is up to 10
  • For SQS FIFO, concurrency is up to the number of active message groups to keep processing in order
  • It is recommended that set the visibility time out to 6x the function timeout
  • Set DLQ on the SQS side (Event Source Mapping uses synchronous invocation)
  • Or you can use lambda destination for failures

Error Handling

  • Default
    • Lambda will keep retrying the entire batch until the items expire
    • Then the precessing for the affected shard will be paused
  • Configuration
    • Discard old events
    • Restrict retry times
    • Split the batch on error (used for resolving timeout)
  • You can set Destination for discarded events

Destinations

  • Asynchronous invocations
    • Set destinations for successful and failed events to
      • SQS, SNS, Lambda, EventBridge
    • It is recommended to use destinations in stead of DLQ of Lambda
  • Event source mapping
    • Set destinations for discarded event batch to
      • SQS, SNS
    • You can use DLQ of SQS instead

Tracing with X-Ray

  • Usage
    • Enable in Lambda configuration
    • Use X-Ray SDK in code
    • Attach AWSXRayDaemonWriteAccess policy to execution role
  • Environment variables
    • _X_AMZN_TRACE_ID: tracing header
    • AWS_XRAY_CONTEXT_MISSING: LOG_ERROR by default
    • AWS_XRAY_CONTEXT_ADDRESS: X-Ray daemon IP address and port

Lambda in VPC

  • Lambda functions have no access to Internet if you place them in a public subnet
  • You should place Lambda functions in a private subnet and attach a NAT Gateway / Instance

Performance

Put you common initial codes outside handler

1
2
3
4
# bad
def lambda_handler(event, context):
  connect_db()
  do_something()
1
2
3
4
# good
connect_db()
def lambda_handler(event, context):
  do_something()

Use /tmp directory as temporary directory

  • Max size is 512 MB
  • The directory remains when the execution context is frozen, providing transient cache that can be used for multiple invocations

Define Lambda Function in CloudFormation

Inline

1
2
3
4
5
6
7
8
9
10
Resources:
  myLambdaFunction:
    Type: AWS::Lambda::Function
    Properties:
      Runtime: python3.x
      Handler: index.handler
      Code:
        ZipFile: |
          def handler(event, context):
            return 'OK!'

S3

1
2
3
4
5
6
7
8
9
10
Resources:
  myLambdaFunction:
    Type: AWS::Lambda::Function
    Properties:
      Runtime: python3.x
      Handler: index.handler
      Code:
        S3Bucket: myBucket
        S3Key: myLambdaFunction.zip
        S3ObjectVersion: myVersionString

Layers

  • Custom Runtime
  • Externalize Dependencies