[packapp] Support conda install

[asavaritayal]

If environment.yaml is present, conda install the dependencies instead of pip install-ing them.

[asavaritayal]

@elprans correct me if I'm wrong but the Conda repository only offers pre-built wheels, so there is no requirement for --build-native-deps path on this one right?

[daguito81]

Any news on this issue?

[yubozhao]

Would love to see any update/news on this feature.

[anthonychu]

@anirudhgarg @vrdmr Do you have the latest guidance on this?

[georghildebrand]

I would also be very much interested in this!

[martikevin]

Hey @anirudhgarg @vrdmr we are also waiting for conda support because some dependencies like geopandas cannot be installed over pip install command.

Please an update!?

[yputter]

Also waiting for this, as turbodbc fails with pip (cannot find wheel, so tries to build, but then fails on dependence pybind11, https://github.com/blue-yonder/turbodbc/issues/59)

[martinjoshua]

Would also love to see an update on this feature request.

[Servot]

Also waiting for this, as scikit-garden is failing with pip

[yubozhao]

We end up using custom Dockerfile. You can find what we did here: https://github.com/bentoml/BentoML/blob/master/docker/azure-functions/Dockerfile

Basically, build on top this and add miniconda

[owenlamont]

I'm also holding out for this - would love Azure functions to have the same support for conda environments as Azure ML.

[stefanushinardi]

Hi @owenlamont, we are planning to publish an image with conda dependencies that you could use in custom container on a premium plan. Would that work for you? or you're planning to use consumption plan?

[owenlamont]

Hi @stefanushinardi - I'm ignorant of what plan my workplace has. I'll have to enquire with the with our data engineering team.

I was aware conda environments could be used by specifying a custom container Docker image but I had hoped I could let Azure Functions abstract Docker away and it would work as written on the description of this ticket - that I could just replace the requirements.txt with an environment.yml and everything else would auto-magically work.

[stefanushinardi]

I see, got it. I dont have a timeline for that yet, but appreciate the nudge. Will discuss internally about the support. cc: @vrdmr @Hazhzeng @AnatoliB

[kyle-pena-nlp]

My general experience is that the conda ecosystem (base, conda-forge, etc.) seems to work a bit better in general. I think more of the community's testing happens on these packages as opposed to pip packages. My feeling is that adding support for an environment.yaml file is pretty critical. I ran into similar limitations with revoscalepy and finding it only supported packages from the pip ecosystem. Perhaps there is a general misconception about the prominence of pip at MS?

[abouroubi]

Is there any news on this ? I use conda for my environment management and I frankly don't see my self going back to venv, it's a weird choice to force user to use an environment manager.

[yuhaodai]

years passed but nothing.

[MarkTensenSgt]

would love this implemented as well!

[abouroubi]

@anthonychu any news on this ? It's been 3 years ...

[athinton]

Yes - I'm pretty new to python, but it seems that conda is the superior environment and package manager over pip and venv. It is disappointing that we have to resort to using docker images to utilise the best tools for the job.

+1 from me to include this feature in Azure Functions.

[AaronM44]

Also looking for this feature

[espenhk]

Yep, also looking for the same feature. Love conda and rarely ever have issues with it, so it's a significant snag that you need workarounds to be able to use azure-functions with conda...

[Kolt45]

4 years in and no change. Have to assume this is profit issue for the business otherwise it is a missed opportunity for adding new projects to Azure Functions.