Windows worker fleets failing to start

[horsmand]

Trying to deploy a worker fleet (or new instance into an existing worker fleet) that uses any Windows AMI is failing due to a bug in a script that we execute on the host as part of its initialization process (in the UserData). The nature of the bug means that it affects all versions of RFDK in production, and all worker fleets that try to deploy using an AMI with a Windows operating system.

The script that is failing attempts to install and configure the CloudWatch agent onto the instance. Because this script fails, the script that is supposed to configure the Deadline worker to connect to the render queue and start it never gets executed and the health check fails, causing the CDK deployment to fail and roll back.

Since the failure prevents CloudWatch from setting up properly, no logs are uploaded to CloudWatch and viewable from the AWS Console, and the host gets terminated.

Log statement seen that signals we're falling back to the latest version of the CloudWatch agent rather than the version we try to pin to: https://github.com/aws/aws-rfdk/blob/mainline/packages/aws-rfdk/lib/core/scripts/powershell/configureCloudWatchAgent.ps1#L26
Error message that was observed: https://github.com/aws/aws-rfdk/blob/mainline/packages/aws-rfdk/lib/core/scripts/powershell/configureCloudWatchAgent.ps1#L52

Environment

• CDK CLI Version: all
• CDK Framework Version: all
• RFDK Version: all
• Deadline Version: all

---

This is 🐛 Bug Report