46:00 - ECS workers pull tasks from SQS at their own pace, so even if SQS is full of messages, their CPU level should not be affected. Shouldn't ECS scaling be triggered by an SQS alarm?
Turns out I misspoke in the talk 🤷♂. The ECS services use Application Auto Scaling with a custom metric. The custom metric is generated using a Lambda function that reads the backlog in the SQS queues
05:57 -- IMDB re-architects to serverless microservices
16:32 -- Amazon Global Ops Robotics protects workloads with cell-based architecture
22:19 -- Amazon Relay app (middle mile) goes multi-Region to keep trucks moving
34:55 -- Amazon Classification and Policy Platform (CPP) uses shuffle sharding to limit blast radius
46:14 -- Amazon Search runs chaos engineering to be ready for Prime Day, any day
46:00 - ECS workers pull tasks from SQS at their own pace, so even if SQS is full of messages, their CPU level should not be affected. Shouldn't ECS scaling be triggered by an SQS alarm?
Turns out I misspoke in the talk 🤷♂. The ECS services use Application Auto Scaling with a custom metric. The custom metric is generated using a Lambda function that reads the backlog in the SQS queues
@@setheliot1 interesting, thank you!