Recommend docker/podman as easiest install route

[jmetz]

Dependencies of pyimagej (imports as imagej) from the imagej app, means that install of this module is significantly more complicated than advertised on the readme.

Recommend that the suggested install route should be docker/podman, or possibly micromamba.

@oeway @Nanguage - what would you think? We could go with mamba/conda/micromamba as recommended in the pyimagej install docs, though docker (cf https://github.com/mamba-org/micromamba-docker ) could be easier?

[oeway]

The imagej-app is here for testing, the idea is to run it inside a conda env, see my note here:

bioengine/apps/imagej/main.py

Lines 16 to 17 in 660cf0a

	# TODO: download the conda env and unpack it
	# Then launch the code inside the conda env

The apps in this folder are basically launcher apps, not the actual heavy lifting process. The apps will responsible to pull docker image, download prebuilt conda env etc. then run the actual services inside those environments, possibly send it to a compute node, ssh to another server, launch in in K8s. This will be implemented via the hypha-launcher.

[jmetz]

Then the issue is simply that cloning the repo, installing the requirements file dependencies, and running python bioimageio.engine start_server simply doesn't work as expected IMO.

The apps fail to load because they're not properly installed.

Perhaps we could have a simple minimal route for just running this repo as is?

For that, a dockerfile/containerfile should work.

[jmetz]

Alternatively if the conda env file is sufficient then I guess that's fine too - in which case it's probably just a case of implementing the to-do you mentioned, and adding some basic documentation to that effect.

[jmetz]

Presumably though, the prebuilt conda env is quite platform specific (as it must bundle openjdk and maven?), so I would still tend to recommend the docker approach over this...

[oeway]

Exactly, the conda env will then run inside a common container(e.g. the triton container), so we don't need to pull so many docker images locally, that will ensure it works cross platform.