@Marvin when a flow is deployed in an ECS cluster, how can I control the CPU and memory of the instance running the flow? Can I set CPU and memory at runtime, just before launching a new run?

To control the CPU and memory of an ECS instance running a Prefect flow, you will need to configure the task definition associated with the ECS service that is responsible for running the Prefect flow. These configurations would typically be set when defining the ECS service and are not usually changed at runtime. However, it seems that perhaps you might want to dynamically adjust these settings just before initiating a flow run. Unfortunately, based on the information retrieved, it's not clear whether you can set CPU and memory at runtime just before launching a new run within the standard Prefect deployment model. In general, changing ECS task definition parameters like CPU and memory settings dynamically at runtime isn't a common pattern, as these settings are part of the service configuration and are intended to remain stable across deployments. For more specific guidance or any recent feature updates that might allow runtime adjustments in Prefect 2.x, it would be wise to review the latest Prefect documentation or contact Prefect support directly. Here's a snippet of relevant information found, although it does not directly answer the question about runtime adjustments:

Images¶
At this time, managed execution requires that you run the official Prefect Docker image: prefecthq/prefect:2-latest. However, as noted above, you can install Python package dependencies at runtime. If you need to use your own image, we recommend using another type of work pool.

Read more in the official Prefect documentation It's worth noting that this information talks about managed executions and using the official Prefect Docker image, which does allow for runtime installation of Python package dependencies, but does not cover ECS specific configurations for CPU and memory resources.