When running a flow from prefect cloud using a Coiled Dask executor, I'm noticing long waits (on the order of minutes) before the first task starts running. Is that expected? Are there any best practices (or even not-so-best practices) that can be used to shorten that loop? I'm hoping to make it part of a dev iteration cycle, which would make those two minutes pretty common, and thus expensive
I'm also asking the Coiled people if there are any options on their side
06/14/2022, 2:31 PM
Yes expected just because that’s the Dask cluster spinup side (if you see the same delay for
). They would be the right ones to ask.
On the Prefect side, I think you can decrease the ECS spin up time by backing it with an EC2 instance that can cache the image?