The cloud-native orchestrator for the whole development lifecycle, with integrated lineage and observability.

dagster

I'm trying to set up a deploy with an EC2 launch type, which with a network mode of `awsvpc` requires enabling <https://docs.aws.amazon.com/AmazonECS/latest/developerguide/task-networking-awsvpc.html|task ENI>, which would require <https://blog.tedivm.com/guides/2020/07/aws-ecs-with-ubuntu-and-gpu-support/|modifying> our Ubuntu AMI in a way I probably can't, as well as <https://docs.aws.amazon.com/AmazonECS/latest/developerguide/container-instance-eni.html|ENI trunking>, which is also an issue. I see comparatively few EC2 launch type configs in here, mostly Fargate. Are we certain `awsvpc` is needed? Does anyone know why?

Probably just going to add a Fargate task definition instead. This is getting too involved.

&gt; Are we certain `awsvpc` is needed?
Originally we only supported Fargate. `awsvpc` is the only network mode that AWS supports for Fargate.

I bet as our ECS support evolved to also support EC2, we never relaxed that constraint. It looks like you’ve maybe found a workaround for now by launching in Fargate instead, but if you open an issue in GitHub, we can look into relaxing the constraint.

I think we’d more or less need to thread this through `DagsterEcsTaskDefinitionConfig` instead of hardcoding it. And then give you a way to configure that on the launcher:

<https://github.com/dagster-io/dagster/blob/74eac38af3cb7c9f65306f6c151c4dad07a2d427/python_modules/libraries/dagster-aws/dagster_aws/ecs/tasks.py#L71>

I guess specifying
```use_current_ecs_task_config: false
task_definition: &lt;arn-of-a-manually-created-task-definition-without-awsvpc&gt;
run_task_kwargs:
  cluster: &lt;cluster-name&gt;```
in EcsRunLauncher config could be a workaround for this issue?