You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
- deleted logs/ directory and added it to .gitignore
- removed RELEASE_PYTHON_VERSION and standardized on PYTHON_VERSION makefile variable
- helper functions to parse makefile target and extract important metadata as makefile variables
- add retries to podman push in build_image makefile function
- dynamically build workbench directory / dockerfile filename based on target
- standardized makefile image targets as <accelerator>-<feature>-<scope>-<os>-<python version>
- single deploy-% target for all images
- single undeploy-% target for all images
- singe test-% target for all images
- new e2e-% target that runs $* + deploy-$* + test-$* + undeploy-$*
- updated/simplified make_test.py in light of Makefile changes
- pass kustomize output to kubectl via stdin to avoid accidental checkin of personal settings
- refactored notebooks/ repo file hierarchy to consistently leverage subfolders for accelerator-specific resources
- renamed runtimes folder to runtime to match target name
- jupyter/cuda + jupyter/rocm
- runtime/cuda + runtime/rocm
- updated kustomize resources for consistency
- image name used an manifest name prefix
- -workbench used as manifest name suffix
- using labels transformer as commonLabels deprecated
- containerPort named workbench-port
- removed spec.containers.command from codeserver/rstudio to let server start
- images.newTag aligned with makefile target
- added emptyDir volume mount to all workloads
- added startupProbe to our accelerator images
- using term "workbench" as opposed to "notebook" consistently throughout manifests
- updated various Dockerfile to match new folder hierarchy where necessary
- refactored test_jupyter_with_papermill to support testing needs of all workbenches + runtimes
- scripts/makefile_utils directory created
- numerous usability enhancements to the logic
- reduce hardcoding of "magic" strings by parsing kustomize output to identify workload names and ports
- scan for open port and use that when verifying container starts via kubectl port-forward
- confirms container starts for all workbenches (not just jupyter)
- confirms required libraries installed in container (now applied to jupyter notebooks as well)
- moved all validate-xxx target logic into script for better consolidated maintenance
- relies on makefile to pass metadata parsed from target name to avoid duplicating logic
- TODO:
- fix any problems in GHA due to above changes
- add NAMING.md file to explain the "rules" around our makefile target names and all the places in our development flow that is impacted
- fix openshift/release due to above changes
- cleanup now-defunct/legacy makefile targets
Related-to: https://issues.redhat.com/browse/RHOAIENG-23291
0 commit comments