That message usually means Prefect successfully tried to hand the flow run to ECS, but ECS could not start the task. In other words, this is typically an infrastructure submission/startup issue, not a failure in your flow code. A few common causes for random/intermittent ECS failures: Most likely - Transient ECS/Fargate capacity issues - Image pull failures (

CannotPullContainerError

) - Docker Hub/ECR rate limiting - Networking issues with subnets / ENIs / security groups - IAM permission issues for the execution role - CloudWatch logging permission issues if log configuration is enabled A few concrete things to check: 1. Look at the worker logs The ECS worker usually logs the real AWS exception that gets summarized into

Flow run could not be submitted to infrastructure

. 2. Look at the ECS task stop reason In the AWS ECS console, inspect the failed task and check: -

stoppedReason

-

stopCode

- container

reason

If you see things like: -

TaskFailedToStart

-

CannotPullContainerError

-

ResourceInitializationError

-

RESOURCE:MEMORY

/

RESOURCE:CPU

those usually identify the root cause. 3. Check whether the image pull is flaky This is very common for nightly bursts. - If you are pulling from Docker Hub, rate limits can cause intermittent failures. - If using private ECR/GHCR, verify auth and permissions. Relevant examples: - Docker Hub rate limit causing this exact Prefect/ECS message - CannotPullContainerError example - Private registry credentials discussion 4. Check CloudWatch log permissions If CloudWatch logging is enabled for the ECS task, missing

logs:*

permissions on the task execution role can prevent task startup. Example: - CloudWatch permission issue causing submission failure 5. Check capacity / networking around the nightly run time - Are many tasks starting at once? - Are subnets running out of IPs / ENIs? - Are you using

FARGATE_SPOT

and seeing occasional placement failures? Useful docs: - Prefect ECS worker docs - Manual ECS deployment / troubleshooting - Work pools and infrastructure concepts A practical short checklist: - Check the exact ECS stopped reason for one failed run - Check whether all failures happen at the same time window nightly - Confirm what image registry you use (Docker Hub, ECR, GHCR, etc.) - Confirm whether CloudWatch logging is enabled - See if the worker logs show retries or AWS API errors If you want, send me: - the full error text from one failed run - the ECS task stoppedReason - whether you’re using Fargate or EC2 launch type - which image registry you use and I can help narrow it down quickly.