Kubernetes agent configuration
This guide is applicable to Dagster+.
This reference describes the various configuration options Dagster+ currently supports for Kubernetes agents.
Viewing the Helm chart
To see the different customizations that can be applied to the Kubernetes agent, you can view the chart's default values:
helm repo add dagster-plus https://dagster-io.github.io/helm-user-cloud
helm repo update
helm show values dagster-plus/dagster-plus-agent
You can also view the chart values on ArtifactHub.
Per-deployment configuration
The workspace value of the Helm chart provides the ability to add configuration for all jobs that are spun up by the agent, across all repositories. To add secrets or mounted volumes to all Kubernetes Pods, you can specify your desired configuration under this value.
Additionally, the imagePullSecrets value allows you to specify a list of secrets that should be included when pulling the images for your containers.
Per-location configuration
When adding a code location to Dagster+ with a Kubernetes agent, you can use the container_context key on the location configuration to add additional Kubernetes-specific configuration. If you're using the Dagster+ Github action, the container_context key can also be set for each location in your dagster_cloud.yaml file, using the same format.
The following example dagster_cloud.yaml file illustrates the available fields:
# dagster_cloud.yaml
locations:
- location_name: cloud-examples
image: dagster/dagster-cloud-examples:latest
code_source:
package_name: dagster_cloud_examples
container_context:
k8s:
env_config_maps:
- my_config_map
env_secrets:
- my_secret
env_vars:
- FOO_ENV_VAR=foo_value
- BAR_ENV_VAR
image_pull_policy: Always
image_pull_secrets:
- name: my_image_pull_secret
labels:
my_label_key: my_label_value
namespace: my_k8s_namespace
service_account_name: my_service_account_name
volume_mounts:
- mount_path: /opt/dagster/test_mount_path/volume_mounted_file.yaml
name: test-volume
sub_path: volume_mounted_file.yaml
volumes:
- name: test-volume
config_map:
name: test-volume-configmap
server_k8s_config: # Raw kubernetes config for code servers launched by the agent
pod_spec_config: # Config for the code server pod spec
node_selector:
disktype: standard
pod_template_spec_metadata: # Metadata for the code server pod
annotations:
mykey: myvalue
deployment_metadata: # Metadata for the code server deployment
annotations:
mykey: myvalue
service_metadata: # Metadata for the code server service
annotations:
mykey: myvalue
container_config: # Config for the main dagster container in the code server pod
resources:
limits:
cpu: 100m
memory: 128Mi
run_k8s_config: # Raw kubernetes config for runs launched by the agent
pod_spec_config: # Config for the run's PodSpec
node_selector:
disktype: ssd
container_config: # Config for the main dagster container in the run pod
resources:
limits:
cpu: 500m
memory: 1024Mi
pod_template_spec_metadata: # Metadata for the run pod
annotations:
mykey: myvalue
job_spec_config: # Config for the Kubernetes job for the run
ttl_seconds_after_finished: 7200
job_metadata: # Metadata for the Kubernetes job for the run
annotations:
mykey: myvalue
Environment variables and secrets
Using the container_context.k8s.env_vars and container_context.k8s.env_secrets properties, you can specify environment variables and secrets for a specific code location. For example:
# dagster_cloud.yaml
location:
- location_name: cloud-examples
image: dagster/dagster-cloud-examples:latest
code_source:
package_name: dagster_cloud_examples
container_context:
k8s:
env_vars:
- database_name
- database_username=hooli_testing
env_secrets:
- database_password
| Property | Description |
|---|---|
env_vars | A list of environment variable names to inject into the job, formatted as KEY or KEY=VALUE. If only KEY is specified, the value will be pulled from the current process. |
env_secrets | A list of secret names, from which environment variables for a job are drawn using envFrom. Refer to the Kubernetes documentation for more info. |
Refer to the following guides for more info about environment variables:
Op isolation
By default, each Dagster job will run in its own Kubernetes pod, with each op running in its own subprocess within the pod.
You can also configure your Dagster job with the k8s_job_executor to run each op in its own Kubernetes pod. For example:
from dagster import job
from dagster_k8s import k8s_job_executor
@job(executor_def=k8s_job_executor)
def k8s_job():
...
Per-job and per-op configuration
To add configuration to specific Dagster jobs, ops, or assets, use the dagster-k8s/config tag. For example, to specify that a job should have certain resource limits when it runs. Refer to Customizing your Kubernetes deployment for Dagster Open Source for more info.
Running as a non-root user
Starting in 0.14.0, the provided docker.io/dagster/dagster-cloud-agent image offers a non-root user with id 1001. To run the agent with this user, you can specify the dagsterCloudAgent value in the Helm chart to be:
dagsterCloudAgent:
podSecurityContext:
runAsUser: 1001
We plan to make this user the default in a future release.
Grant AWS permissions
You can provide your Dagster pods with permissions to assume an AWS IAM role using a Service Account. For example, you might do this to configure an S3 IO Manager.
-
Update the Helm chart to associate the IAM role with a service account:
serviceAccount:
create: true
annotations:
eks.amazonaws.com/role-arn: "arn:aws:iam::1234567890:role/my_service_account_role"
This will allow your agent and the pods it creates to assume the my_service_account_role IAM role.