Tobias Goerke
Tobias Goerke
Hello @botasb, we eventually moved on and started using KFServing instead. While KFServing allows for TFServing to be used, it can also execute arbitrary docker images and thus serve CatBoost...
We are hosting Kubeflow on premises. It took us several weeks to make this work. So: definitely no afternoon task but worth the effort. For testing purposes I'd recommend using...
Does this correspond to your ideas, @chensun and @zijianjoy?
As discussed with @zijianjoy (see [google docs comments](https://docs.google.com/document/d/1WdPjTgoDj5AdPNYVzqYwVIWKsWjPPphYD3Ck4DFxSLA)), disabling the cache for new runs / jobs is now implemented in the backend. The UI hasn't changed and still features a...
> @TobiasGoerke can you rebase to master and fix the merge conflict? > > Please try my new cache server at mtr.devops.telekom.de/ai/cache-server:latest with the environment variables MAXIMUM_CACHE_STALENESS=P0DT0H3M30S DEFAULT_CACHE_STALENESS=P0DT0H1M31S I've tested...
Any way users can delete artifacts and invalidate caches in the UI is fine for me. Your suggestion to not offer these features separately might even be more user-friendly. As...
> > E.g. if you delete a run it tries to delete the artifacts and if you delete an experiment it deletes all runs and therefore all corresponding artifacts. >...
> As desired i moved it into its own distribution because gcs and aws will need different settings. So far this only works for minio. I hope that is fine...
Works nicely now. Pipelines do run, artifacts are stored in the correct path, caching works, no more errors in the metacontroller. lgtm!