Right now, grid search variables are resolved before the random samples are generated.

If we could toggle the order of resolution, we could support this: https://discuss.ray.io/t/is-there-a-way-to-run-the-same-hyperparameter-configuration-multiple-times/1412/12

... to make it easier to read Vespa documentation on an e-reader / offline

Vespa documentation is generated using Jekyll from .md and .html files, look into options for generating the artifact as part of site generation (there might be plugins we can use here)

For SC Operator it may be a good idea to generate CRD manifests from inside a docker container.
This should provide reproducible generation step and avoid "produces different output on my machine" issues.

Linter should also fail if generation of manifests produce diff with the commited version.

https://github.com/PaddlePaddle/Serving/blob/0849586d38e56574d56b03aacb405a42d969568a/python/paddle_serving_app/reader/image_reader.py#L304

应该先判断是否为None，然后再判断它的属性