EcsRunLauncher regression - "Could not determine image for run"

treilly1 commented 5 days ago

What's the issue?

detailed outline in dagster slack https://dagster.slack.com/archives/C01U954MEER/p1732566748483749
Attempting an upgrade to dagster 1.9.2 raises error here despite my dagster.yaml configuring the run launcher as outlined in current documentation

What did you expect to happen?

No breaking ECSRunLauncher changes when upgrading. Since underlying image is specified on my ECS task def (and controlled via CICD process), I'd expect dagster to use image defined on task def.

How to reproduce?

For ECS launchers configured like documentation outlines, attempt kicking off a run.

run_launcher:
module: "dagster_aws.ecs"
class: "EcsRunLauncher"
config:
task_definition: "arn:aws:ecs:us-east-1:1234567890:task-definition/my-task-definition:1"
container_name: "my_container_name"

Dagster version

1.9.2

Deployment type

Other Docker-based deployment

Deployment details

All dagster resources are deployed to an ECS cluster

Additional information

No response

Message from the maintainers

Impacted by this issue? Give it a 👍! We factor engagement into prioritization. By submitting this issue, you agree to follow Dagster's Code of Conduct.

zyd14 commented 5 days ago

I have a PR to roll back the check I added that is causing the error; but tracing through the code it's unclear to me that this will actually fix the issue. It seems like image is being pulled from the job's repository_origin, and it seems like run_task_kwargs needs image to be non-None . There's a lot of infra to set up to be able to test this, I'm hoping @treilly1 will be able to test my changes with their existing infra if possible

lokinnnnnn commented 2 days ago

I also encountered this issue, so I returned to version 1.9.0

dagster-io / dagster