CodeDeploy is doing in-place deployments across EC2 instances in an Auto Scaling group using the CodeDeployDefault.OneAtATime configuration. After a deployment finishes, two of five instances still run the previous revision while three run the new one. What is the most likely cause?

The most likely cause is that EC2 Auto Scaling launched new instances during the deployment. When new instances are launched in an Auto Scaling group, they typically use the AMI and user data defined in the launch template or configuration. If the deployment was still in progress and the Auto Scaling group scaled out, these new instances would have received the older application revision present in the AMI, not the new revision being deployed by CodeDeploy. CodeDeploy's OneAtATime configuration updates instances sequentially, so new instances launched mid-deployment wouldn't be part of the current deployment target group. Incorrect options: If instances failed to download the package or the agent wasn't installed, CodeDeploy would report deployment failures for those instances, not a partial deployment where they continue running the old revision without error. A failing AfterInstall hook would cause the deployment to fail on those instances, potentially rolling back, but CodeDeploy would report this failure. The scenario describes instances running the old revision without an explicit failure reported by CodeDeploy for those specific instances.