k8sRunLauncher nodeSelector configuration is ignored when includeConfigInLaunchedRuns.enabled is true

What's the issue?

When includeConfigInLaunchedRuns.enabled is set to true in the Helm chart, the nodeSelector specified in k8sRunLauncher.runK8sConfig.podSpecConfig is overridden by the nodeSelector from user deployment configuration. This prevents proper control over which nodes run pods are scheduled on.

What did you expect to happen?

Instance-level k8sRunLauncher configuration should take precedence over inherited user deployment configuration.

How to reproduce?

1. Set includeConfigInLaunchedRuns.enabled: true in Helm values
2. Configure a specific nodeSelector in runLauncher.config.k8sRunLauncher.runK8sConfig.podSpecConfig
3. Have a different nodeSelector in user deployment configuration
4. Launch a run
5. Observe that the run pod uses the user deployment's nodeSelector instead of the configured one

Dagster version

1.13.7

Deployment type

Dagster Helm chart

Deployment details

No response

Additional information

Need to direct run pods to autoscaling node pools while keeping user deployments on default pools for resource optimization.

Message from the maintainers

Impacted by this issue? Give it a ! We factor engagement into prioritization.