In current pipeline first schedule strategy, we assign all of resources to each actor/worker instead of ray GCS management.
It's good way to place all of actors once in single job cluster, but for multitenacy there are many job master actors so they can not manage shared resource of others. It would better to fit more elastic strategy and let gcs do more things in multitenacy mode.
In current pipeline first schedule strategy, we assign all of resources to each actor/worker instead of ray GCS management. It's good way to place all of actors once in single job cluster, but for multitenacy there are many job master actors so they can not manage shared resource of others. It would better to fit more elastic strategy and let gcs do more things in multitenacy mode.