Releases: project-codeflare/codeflare-sdk
v0.4.3 Head Node Usability
Fixes issue where small jobs would fail on OpenShift when Ray scheduled them on head node.
v0.4.2 TorchX Scheduler Updates and Fixes
What's Changed
- Change default namespace logic by @anishasthana in #85
- add rdzv_backend parameter to DDPJobDefinition by @MichaelClifford in #87
- make workspace user accessible by @MichaelClifford in #89
New Contributors
- @anishasthana made their first contribution in #85
Full Changelog: v0.4.1...v0.4.2
Caching Bug Fix
Fixes an issue where old TorchX code was being cached and breaking Job submission until refresh.
v0.4.0 Job Management Update
The SDK now has the ability to:
- Submit jobs to a Ray cluster (via TorchX)
- Submit jobs directly to MCAD (via TorchX)
- Check submitted job status
- Check submitted job logs
- Cancel a submitted job
Also included are other minor bug fixes and improvements (mainly clarity on starting vs. running Ray clusters).
v0.3.1 Allow for user defined ports in Auth
What's Changed
- Remove hard-coded port 6443 in auth by @MichaelClifford in #74
Full Changelog: v0.3.0...v0.3.1
v0.3.0 Info and Auth Updates
Updates:
- New all-encompassing cluster.status() method
- New cluster.details() method
- New cluster.wait_ready() method
- New CodeFlareStatus.STARTING status
- Functions get_current_namespace(), list_all_clusters(), and list_all_queued() now all function as intended
- A ton of bug fixes and error-handling for both regular use and edge-cases (not logged in, login expired, no resources created yet,
- resource failure, cluster setup in-progress, etc.)
- A LOT more testing (up to 95%)
- Improved messaging on both expected and unexpected results
Authentication: abstracted from cluster creation/deletion, now can independently authenticate and logout
v0.2.2 Requirement and Dependency Update
Corrects requirements and package dependencies by switching from ray 2.1.0
to ray[default]
2.1.0, and corrects allowed python version range to be in line with rich
compatibility.
v0.2.1 Auth Skip TLS Option
Adds a new option to the TokenAuthentication
object to skip TLS verification.
v0.2.0 Ray Job Options
Adds the ability to:
- View a list of submitted Ray jobs
- View status of a given Ray job
- View logs of a given Ray job
all directly through the SDK
v0.1.7 Authentication and Descriptions
Added new OCP cluster authentication options as well as docstrings and type hints to all modules. Also reformatted and added format-checking rules for developers.