make docker build multistage

[nilsnolde]

Fixes #1232

Multistaging the ORS image brings down the uncompressed size from 1.06 GB to 365 GB.

EDIT: actually the most important improvement I forgot😅. Before, the Java code was compiled for every container. Now it’s compiled in the image, and a container starts building a graph instantly, instead of first compiling the project for 10 mins.

I also tried for along time to keep the logic with USER etc, but eventually it seems rather like it was more redundant and didn't really do what I'd have expected it to do, even before the change. So I removed most of that which makes it a bit easier to maintain I guess.

Basically, before this change, the paths which are mounted as volumes in the docker-compose, were never really owned by whatever the UID was. They were always owned by root. If I actually provided a user ID in docker-compose.yml that matched my host one (1000), it wasn't able to create any files in those directories. AFAIU it's happening because after the image is created, those paths are mounted on the host machine, which makes them root. I tried fixing that as well, but couldn't in any reasonable time. TBH, I do that exact thing on our Valhalla image and it works. I tried replicating exactly that for 2 hours with this image and I couldn't get it to work. In the end the effect is that nothing changes compared to before, other than having 1/3 of the image size and a MUCH quicker container startup time.

More comments in the changes.

[rabidllama]

Seems OK to me, the main thing though that I am not sure about is the user stuff as from what I recall, that was added to fix problems people were having with the container not being able to write to the folders

[nilsnolde]

The way I read the code before these changes was that you'd provide a host's UID & GID per env vars which are then used for the container-internal ors user, and default to 0, ie root:

openrouteservice/docker/docker-compose.yml

Line 14 in 338979f

user: "${ORS_UID:-0}:${ORS_GID:-0}"

The idea behind that is usually that you'd like to avoid having to use sudo on the host for whatever the container creates and is mapped back to the host (plus the docker internal user is not root which is also nice when exposing as host volumes). However, I don't think that was ever properly working and my attempt to make it work wasn't very successful.

Anyways, you can take whatever you think is nice from this PR and trash the rest. I'm only using this image to test some of our routing py/js libs and I'm happy with our custom image, likely for years to come.

[MichaelsJP]

@nilsnolde Thanks for your contribution. More than welcome. I left you some comments. Just leave me a little comment if you want to provide the changes, else we'd happily just take your commits and do the rest ourselves. :)

[MichaelsJP]

Cherrypicked your commits into a new branch and made a new PR: #1272

Will be merged soon.