0.27.0
What's Changed
- [feature] Added new S3Location and S3Bucket types that render S3 links in the Dashboard, and
documented them - [feature] Added a new Image type that render images in the Dashboard, and documented it
- [example] Added TorchLightning + Resnet + Ray example
- [example] Added Ray AIR example
- [example] Added various enhancements to the example Testing Pipeline
- [improvement] BREAKING CHANGE:
image_layers
field insematic_pipeline
bazel macro
now ONLY gets passed to image layering, and not also to the Sematic binary target.
if you are usingimage_layers
to express dependencies of a pipeline, you will now
need to duplicate them in thedeps
field. - [improvement] Added support for Python 3.10
- [improvement] Added support for Python 3.10-style type hints
- [improvement] The Dashboard login flow will now redirect to the requested page instead of to
the homepage - [improvement] Displaying the user who started a Run in the Dashboard Run list page and in the
Run history drop-down - [improvement] Various documentation updates and fixes
- [improvement] Various log message improvements and enhancements
- [improvement] Sped up Docker image generation through various improvements
- [improvement] Automatically update npm dependencies when building the wheel
- [improvement] Added deep-linking selected panel stickiness in the Dashboard
- [improvement] Filtered only pipelines launched by the current user in the Dashboard homepage
last run widget - [improvement] Documented Ray integration architecture
- [improvement] Improved wheel building portability by defaulting to bash
- [improvement] Improved reaction time of the CloudResolver after a very long-running cloud Run
has ended - [improvement] Improved handling of incorrect cleanup of or errors in subprocess spawned by
user code - [improvement] Various internal refactorings, cleanups, and build improvements
- [improvement] Switched from MB and GB to GiB and MiB in the Ray configurations, for
consistency and standardization reasons - [improvement] Added checks and documentation for the new libmagic dependency
- [improvement] Resolution failures caused by Resolver errors now get marked as failed instead
of canceled - [bugfix] Pinned MNIST example pipeline dependencies after a new dependency release broke the
execution - [bugfix] Fixed a bug that prevented the browser back/forward buttons from working correctly
in the Dashboard - [bugfix] Fixed a bug that sometimes prevented the Dashboard Run tree panel from updating
- [bugfix] Fixed sorting of the Runs in the Dashboard Run tree panel by creation time
- [bugfix] Fixed a race condition where Kubernetes job updates would be incorrectly handled,
and added safeguards for incorrect Run state transitions - [bugfix] Fixed a bug where killing a subprocess spawned by user code would prematurely
terminate the Run, or leave the Resolver in an inconsistent state - [bugfix] Fixed a race condition that would prevent a fresh deployment on Kubernetes due to
missing resources - [bugfix] Fixed a bug in the Dashboard that prevented the log panel error messages from
refreshing when seeking to a different Run - [bugfix] Fixed a bug where failure to notify the Dashboard or Resolver with Run updates via
socketio messages would cause the Resolution to fail
Full Changelog: v0.26.0...v0.27.0
Compatibility
Sematic Servers on this version can support pip packages back to v0.24.1
.