Support Docker images for Apple Silicon/M1 arm64 platform

[muthenberg]

Hi,

Using the latest docker images on my MacBook Pro M1 with Apple Silicon I get the following warning:
WARNING: The requested image's platform (linux/amd64) does not match the detected host platform (linux/arm64/v8) and no specific platform was requested

It seems the image works fine, but I guess it uses Rosetta 2 for amd64 emulation. I would prefer a native image. Are there any plans to support this?

I would be willing to create those images on my machine and provide them to you, if you have no such MacBook. However I do not know, whether this is desired and whether I would be able to build the images. My C++ experience is pretty dated and based on using Visual Studio 6.0.

[hoerup]

I assume this will depend both on getting GH actions working for building docker image in general ( see #6137 ) AND github adding support for M1 ( actions/runner-images#2187 )

Oh yeah, also needs to have debian support in general for the M1 and having debian generating baseimage for this architecture.

[hoerup]

@muthenberg I may have misunderstood something
Is M1 a "normal" arm64/arm V8 ??

In that case I guess the GH actions for building cross platform containers could be utilized ?

You could look into creating a PR expanding the current GH actions to support crossplatform

[muthenberg]

@hoerup That is a good question. I know that M1 has an ARM processor, but I am not very knowledgable about different kinds of those processors. Wikipedia tells me that the instruction set is arm V8.4-A.

If you create such a PR I would be happy to try to run it on my machine and tell you the results.

[SiarheiFedartsou]

We already build M1 containers on CI, will be available in 5.27.

[Kradenko]

Hi, would this mean that 5.27 would then also work on Raspberry Pi? If so, when will 5.27 be available?

[SiarheiFedartsou]

@Kradenko hm, I think so, but it is to be tested. Could you build docker image from master and try? I can provide instructions on how it can be done if needed.

[Kradenko]

@SiarheiFedartsou Please! I have been looking for a howto all day.

[SiarheiFedartsou]

@SiarheiFedartsou Please! I have been looking for a howto all day.

You can just build Docker image like this:

docker build -t osrm-local --platform arm64 -f docker/Dockerfile .

Then you can just try to run OSRM as usual, but changing osrm/osrm-backend to osrm-local to use image you just built rather then version from Docker Hub. Let me know if it is not clear.

[Kradenko]

@SiarheiFedartsou Please! I have been looking for a howto all day.

You can just build Docker image like this:

docker build -t osrm-local --platform arm64 -f docker/Dockerfile .

Then you can just try to run OSRM as usual, but changing osrm/osrm-backend to osrm-local to use image you just built rather then version from Docker Hub. Let me know if it is not clear.

Will give it a go later tonight and let you know.

[Kradenko]

Apologies for the incredibly late response. This method has served me very well for using it on the Raspberry Pi.

I only wish we had more memory to run the OSRM on the Pis. Will ask if there is a way to use swap space at some stage. For testing this is fine now.

just to add to that I have placed pre-compiled docker images on my repository here: https://hub.docker.com/r/kradenko/osrm-backend