https://dagster.io/ logo
#ask-ai
Title
# ask-ai
s

Shane Keller

03/12/2024, 3:46 PM
I'm running a dagster job on a hybrid deployment with an ECS agent and max concurrency setting of 10, but the job is only running 2 workers at once.
s

Scout

03/12/2024, 3:46 PM
To address the issue of your Dagster job running with fewer workers than the specified max concurrency setting in a hybrid deployment with an ECS agent, we should first ensure that the configuration is set up correctly. In a hybrid deployment, the Dagster agent is responsible for executing your code and streaming metadata about code execution to Dagster Cloud's Agent API. For ECS specifically, the agent executes Dagster jobs as AWS ECS tasks. The configuration for concurrency and worker count can be influenced by several factors, including the ECS task definition, the Dagster run launcher configuration, and the run coordinator settings. Here are a few steps you can take to troubleshoot the issue: 1. Check the ECS Task Definition: Ensure that the ECS task definition for your job has the appropriate CPU and memory allocation to support the desired number of workers. If the task definition does not have enough resources, ECS may not be able to start the desired number of tasks. 2. Review the Dagster Run Launcher Configuration: The
EcsRunLauncher
configuration in your
dagster.yaml
should be reviewed to ensure it is set up to launch the correct number of tasks. If you have a custom task definition, make sure it is correctly referenced in the configuration. 3. Inspect the Run Coordinator Settings: The
QueuedRunCoordinator
settings in your
dagster.yaml
can affect how many runs are pulled from the run queue. If you have set a max concurrency limit here, it could be limiting the number of concurrent runs. You can adjust the settings to increase the number of workers that are dequeued concurrently. 4. Check for Other Concurrency Limits: Ensure that there are no other concurrency limits set at the asset or job level that could be restricting the number of concurrent runs. 5. Monitor ECS Service: Check the ECS service and cluster metrics to see if there are any errors or resource constraints that could be preventing the desired number of tasks from running. 6. Review Dagster Logs: Check the logs for the Dagster daemon and the ECS tasks to see if there are any errors or warnings that could provide clues as to why the concurrency is not as expected. If after reviewing these areas the issue persists, it may be helpful to consult the documentation for more detailed guidance on configuring ECS agents and run concurrency, or to seek further assistance from the Dagster community or support channels.
8 Views