[autoscaler] 'KeyName' assertion error when migrating to 1.2.0

When migrating from the legacy autoscaler config format to the new format in 1.2.0, I’m getting a KeyName assertion error in `ray/autoscaler/_private/aws/config.py:_configure_key_pair.

The problem seems to be that the new format requires the elimination of the head_node and worker_nodes keys from the cluster config, but in the _configure_key_pair function, it still does assertion checks for KeyName under config[“head_node”] and config[“worker_nodes”]. And for some reason, KeyName is not used from within available_node_types:ray.head.default:node_config, where I’d defined it.

For additional context, this issue doesn’t block my migration process because I just have to add the KeyName key to head_node and worker_nodes in the config and everything works perfectly fine, so I’m leaving this topic here just as a reference point in case it’s useful.

Thanks @Vishnu – it is indeed still the case that KeyName must still be specified under head_node and worker_nodes fields – this is a bit of technical debt that we haven’t resolved yet.

1 Like