Automate Model Examples and remove last references to Scikit-Learn v0.23.2 #146
Conversation
…em. Requires running apps WITHOUT gunicorn to avoid zombies
…ing exampels, fix broken link, remove beta flag warning
…t base image - not permanent
…d Help Message and skip H2O. Required bumping xgboost
| # WARNING: Killing the subprocess may not kill any workers spawned by the process (e.g. gunicorn!) | ||
| p.kill() | ||
| p.wait() | ||
| capture_output(stdout.name, stderr.name) | ||
| raise |
There was a problem hiding this comment.
The shutdown functionality for the prediction services is kinda wonky right now, especially when testing. It seemed like the best way to handle it was by using Python's subprocess module & just killing the parent process (and for the sake of our examples, the apps run with the dev server so we never risk leaving zombie workers).
- The builtin
/shutdownendpoint only works forgunicornservers (whenproduction=True) - this is a known bug in the Model SDK package. - Not all tests will call the
/shutdownendpoint - like when running python scans. Nor do we expect all the servers to be run withproduction=True - Python's
kill()here will kill the process, but if you are using gunicorn, then the spawned workers are left as zombies
| scikit_0.23: | ||
| deployment: scikit_0.23_deployment.yml | ||
| python: | ||
| deployment: python_deployment.yml |
There was a problem hiding this comment.
I'm replacing the scikit_0.23 reference with python, and adding the current set of base images. We will need to update this when the release time comes (or figure out some way to automate this).
Follow up action would be to update the copy on our Certifai Dev to have python
…to support function() syntax
… env before doing install in pipeline
…del requirements. Undo set up of conda for docker related commands
|
Successful PR Pipeline instance: https://gocd.dci-dev.dev-eks.insights.ai/go/tab/build/detail/cortex-certifai-examples-pr/72/examples/1/Notebooks |
1 participant
Issue: https://github.com/CognitiveScale/certifai/issues/4812
Overview of changes:
run_test.shscript, for running containerized model examples locally while skipping h2o modelsscikit0.23reference - replace withpythonxgboostfrom1.2.0to1.7.2- this was required to be able to installxgbooston the M1 Mac (v1.2.0 was released in 2020!!)