SageMaker Canvas Auto Update, Auto Predict - Fails

In Amazon SageMaker Canvas, automatic dataset updates and batch predictions only run while you're logged into the Canvas application. If you're not seeing your automatic updates trigger at the scheduled time, the most likely cause is that you're not logged into the Canvas application when the scheduled update is supposed to occur.

When you log out of Canvas, all automations (including automatic dataset updates and batch predictions) are paused. They will resume only when you log back into the application.

To troubleshoot your issue:

1. Make sure you're logged into the Canvas application during the scheduled update times.

2. Check if the automatic update configuration is enabled. You can verify this by:

• Going to the ML Ops section in the left navigation pane
• Selecting the Automations tab
• Checking the Configuration tab to ensure the Auto update toggle is turned on for your dataset

3. If the configuration shows as enabled but jobs aren't running, you might need to check if there are any schema issues with the new files in your S3 location. For automatic updates, Canvas doesn't perform schema validation, and if the new files don't match the existing schema, you may encounter errors.

4. The job history for automatic updates isn't updated in real-time. There's typically a delay of about 5-10 minutes after the scheduled run time before the job history is updated in the interface.

Since you mentioned that manual updates still trigger the linked predictions correctly, the connection between your dataset and prediction is working. The issue appears to be specifically with the automatic triggering of updates.
Sources
Configure automatic updates for a dataset - Amazon SageMaker
Edit your automatic dataset update configuration - Amazon SageMaker
if sagemaker dataset gets a new version automatically, does the model automatically rebuild? | AWS re:Post
sagemaker canvas dataset auto update, when is job history updated? | AWS re:Post