skypilot
skypilot copied to clipboard
[Sky Onprem] Features and Fixes for Sky Onprem
Features
- [ ]
sky admin status
- New CLI command that lists all user jobs for the administrator - [ ] Add local cluster spillover to multiple clouds
- [ ] When registering a local cluster, make it file-based (no need to run
sky launch -c [LOCAL_CLUSTER] ''
) - [ ] Make Sky installation from admin perspective only for the root account (or even better, remove Sky installation completely for admin)
- [ ] Sky YAML script to automatically install Ray/Sky and create local cluster on a public cloud
Fixes
- [x] Add schema check for cluster config yaml #1044
- [ ] Detect more GPU resource types for detecting accelerators on the local cluster
- [x] Refactor local cluster methods out of
cli.py
#1043 - [ ] Make local cluster job submission logic only switch users once (in
sky_app_[ID].py
) - [x] Fix potential python conflict between Ray cluster's python (root user's python) and user's python #1030
- [x] When
sky queue [CLUSTER]
is specified, do not outputLocal cluster is not initialized...
#1038
Other Features mentioned in #801 (lower priority):
- [ ]
sky admin useradd
- CLI command to add and remove users on the local cluster - [ ]
sky admin clean
- Recover from a bad state if onprem cluster breaks