Enable control of logging to be more granual

SeldonIO / MLServer

An inference server for your machine learning models, including support for multiple frameworks, multi-model serving and more

https://mlserver.readthedocs.io/en/latest/

Apache License 2.0

729 stars 183 forks source link

Enable control of logging to be more granual #968

Open dtpryce opened 1 year ago

dtpryce commented 1 year ago

Currently setting debug=true stops ALL logging which is fine for some use cases but others may want to still see some logging (ours). The behaviour we would love to see is that we can set per endpoint or group of endpoints a log level or verbosity. e.g. inference_logging = INFO | DEBUG | WARN | CRITICAL | OFF, metric_logging = same, probe_logging = same. And that this is then reflected up into Seldon Core.

adriangonz commented 1 year ago

Hey @dtpryce ,

Thanks for opening this one!

That's an interesting suggestion. Initially, I was just thinking of keeping the probes logs separate, but I agree that a second flag for metrics could also make sense.

dtpryce commented 1 year ago

Hey @adriangonz yeah was making sure to be thorough I guess and can definitely see some use case in being able to turn on different endpoint logging selectively. For us it would be probe = OFF, metric = INFO, inference = INFO I think and ensuring this goes up to Seldon so that we have the right logging format at that level has become important too!

I will follow along, sadly not sure I can contribute too much on this one.