Prediction API get model version

[jamborta]

I have a single model deployment which I am planning to update occasionally. Just wondering if there is a way for the prediction API to return the model version that it is currently serving, as it can be ambiguous during a rolling update. Currently it only returns an empty header:

'meta': {'puid': 'lt91ru9u5f8j8530m9dl40rrqo', 'tags': {}, 'routing': {}}}

[ukclivecox]

We certainly want to add a set of default fields taken from the resource definition such as image name and version. This work we need feedback on what people want by default.

The wrappers at present focus on allowing the user to just return the predictions but we need to make it easy to pass back arbitrary meta data.

Perhaps we can use this issue as a basis for what we would like to see.

[jamborta]

Image name and version would be sufficient for me.

[ukclivecox]

If you could test with 0.2.4-SNAPSHOT images? @jamborta
See the Pull request #246 for details.

[jamborta]

Great! Thank you.

[jamborta]

@cliveseldon trying to test this, I'm using the image seldonio/seldon-core-s2i-python3:0.3-SNAPSHOT but getting the old format. It is set to REST.

[ukclivecox]

You need to use images from 0.2.4-SNAPSHOT when installing seldon-core. Can you check you are using these. And you have redownloaded if so as they are SNAPSHOT images and recreated periodically.

[jamborta]

@cliveseldon This was working last week, but today I reinstalled seldon-core-0.2.4-SNAPSHOT, but it is returning the old format.

[ukclivecox]

OK, that's strange. Running the latest helm examples should give you an example output like below:

Can you tell us more about how you are testing so we can try to dig deeper?

[jamborta]

@cliveseldon sorry, just realised that it picked up some old jar file from cache when installed. Works ok now.