Short answer: yes—if you switch away from ECS to a local Process or Docker work pool on your EC2 host, you can get near‑zero start latency. ECS (even on EC2) typically adds seconds due to polling, task definition plumbing, and container scheduling. Your options - Fastest: Process work pool (subprocess on the same EC2 box) - Typical start time: ~0.1–0.3s - Tradeoff: no container isolation - Steps:

prefect work-pool create "local-process" --type process
    prefect worker start -p local-process

Then point your deployment at

local-process

. - Very fast: Docker work pool (local Docker on the same EC2 box) - Typical start time: ~0.5–2s with the image cached - Tradeoff: needs Docker daemon on that EC2 host - Steps:

uv run --with 'prefect[docker]' prefect work-pool create "local-docker" --type docker
    uv run --with 'prefect[docker]' prefect worker start -p local-docker

In your deployment, use a pinned image tag and pre‑pull it on the host for immediate starts. If you want to keep ECS (EC2 launch type) but cut latency as much as possible - Reduce worker polling interval (default is 10s):

export PREFECT_WORKER_QUERY_SECONDS=2

- Ensure EC2 capacity is warm (ASG min capacity > 0, no scale‑up delay). - Reuse task definitions to avoid registrations: - In the ECS work pool variables, enable

match_latest_revision_in_family: true

or supply a fixed

task_definition_arn

. - Keep images cached on instances: - Use pinned tags (not

latest

) and pre‑pull on your EC2 instances; avoid “always pull” behavior. What I recommend for your setup - Since you run Prefect Server and workers on a single EC2 instance and all envs share that host, use: - Process pool for the absolute fastest dev/test/startup. - Docker pool for fast starts plus container isolation. - Keep your ECS EC2 pool for production later if you need horizontal scale across instances. Quick deployment example (process)

from prefect import flow

@flow
def my_flow():
    ...

if __name__ == "__main__":
    my_flow.deploy(
        name="ultra-fast",
        work_pool_name="local-process",
    )

Question: Is running flows directly on that same EC2 host acceptable from an isolation/compliance standpoint, or do you need them to run inside ECS tasks? That will determine whether we move you to Process/Docker or just aggressively tune your ECS pool.