Sometimes workers deployed by the Azure batch worker manager deploy a stale image (i.e. an older image before recent changes were made to the worker).
However, when I docker pulled the image with the same name and tag, I got the updated image. So I'm not sure what's happening. Likely, it has to do with the way Azure is pulling the image.
The image name is set by this line: worker_image: str = 'codalab/worker:' + os.environ.get('CODALAB_VERSION', 'latest')
and then it's used in the TaskContainerSettings here:
Sometimes workers deployed by the Azure batch worker manager deploy a stale image (i.e. an older image before recent changes were made to the worker).
However, when I docker pulled the image with the same name and tag, I got the updated image. So I'm not sure what's happening. Likely, it has to do with the way Azure is pulling the image.
The image name is set by this line:
worker_image: str = 'codalab/worker:' + os.environ.get('CODALAB_VERSION', 'latest')
and then it's used in the TaskContainerSettings here: