Fixing AWS with AWS

DevOpsDays NYC 2018 Ignite talk

https://www.devopsdays.org/events/2018-new-york-city/program/mike-fiedler/

Transcript

1. Fixing AWS with AWS
2. Photo by Ben White on Unsplash
3. Wrangler of the Unusual Roller Derby Referee Twitter: @mikefiedler GitHub: @miketheman Mike Fiedler © Alex Erde
4. Photo by Will van Wingerden on Unsplash
5. Photo by Ashim D’Silva on Unsplash
6. Photo by Guillaume Bolduc on Unsplash
7. Photo by Peter Nguyen on Unsplash
9. Henderson 1921 FLXIBLE Sidecar, Wikimedia
10. Photo by Todd Diemer on Unsplash
11. global service DaemonSet daemon-scheduler
12. One Task Per Host - ECS? • Run on Boot via user-data ◦ Unmanaged task, doesn’t restart ◦ Update requires host replacement • Placement Strategies/Constraints ◦ distinctInstance / memberOf ◦ Count mismatch ◦ Doesn’t follow Autoscaling
13. Photo by Denys Nevozhai on Unsplash
14. CloudWatch Event filter { "source": [ "aws.ecs" ], "detail-type": [ "ECS Container Instance State Change" ] }
15. AWS Lambda Handler (trimmed) def lambda_handler(event, context): service = os.getenv('ECS_SERVICE_ARN') cluster = event["detail"]["clusterArn"] adjust_service_desired_count( ecs_client(), cluster, service )
16. def adjust_service_desired_count(ecs_client, cluster, service): running_service = ecs_client.describe_services(cluster=cluster, services=[service]) desired_task_count = running_service["services"][0]["desiredCount"] clusters = ecs_client.describe_clusters(clusters=[cluster]) registered_instances = clusters["clusters"][0]["registeredContainerInstancesCount"] if desired_task_count != registered_instances: response = ecs_client.update_service( cluster=cluster, service=service, desiredCount=registered_instances, ) return response
18. 1.2 mil invokes 4 services 20 clusters ~$17/mo #serverless Lambda Pricing Calculator
19. Photo by Hunter Haley on Unsplash
20. Twitter: @mikefiedler GitHub: @miketheman Mike Fiedler You too can extend AWS!