Lambda in 2025: Year-Long Executions and SnapStart Everywhere

AWS Lambda in 2025: Year-Long Executions and Failed-Event Destinations

Lambda keeps evolving. The latest updates allow functions to coordinate workflows for up to one year and automatically route failed events to S3. These changes expand Lambda’s use cases significantly.

What’s New in Lambda (2024-2025)

Long-Running Workflows Without Step Functions

New coordination capabilities let Lambda handle extended workflows:

import boto3
import time

def lambda_handler(event, context):
    # Start a long-running process
    process_id = start_data_pipeline(event)

    # Wait for external event (up to 1 year)
    # Lambda hibernates - no charges while waiting
    result = wait_for_completion(
        process_id,
        timeout_seconds=86400 * 365  # 1 year
    )

    return process_result(result)

S3 as Failed-Event Destination

# Configure S3 destination for failed Kafka events
aws lambda create-event-source-mapping \
    --function-name my-processor \
    --event-source-arn arn:aws:kafka:... \
    --destination-config '{
        "OnFailure": {
            "Destination": "arn:aws:s3:::my-failed-events-bucket"
        }
    }'

# Failed events saved as:
# s3://bucket/aws/lambda/events/YYYY/MM/DD/function-name/...

SnapStart for Python

Cold starts reduced from seconds to milliseconds:

# Enable SnapStart in Terraform
resource "aws_lambda_function" "api" {
  function_name = "my-api"
  runtime       = "python3.12"
  handler       = "app.handler"

  snap_start {
    apply_on = "PublishedVersions"
  }
}

# Cold start: ~100ms (vs 1-3 seconds without)