Unable to run pipeline-runner image under docker on amd64 system

[bpow]

Not sure that this issue should be filed here or with the seqr-loading-pipelines project, but when working through the LOCAL_INSTALL.md document I run into issues with docker-compose trying to start the pipeline-runner container.

When running docker-compose up -- pipeline-runner, I get:

elasticsearch is up-to-date
Starting seqr_pipeline-runner_1 ... done
Attaching to seqr_pipeline-runner_1
pipeline-runner_1  | exec /entrypoint.sh: exec format error
seqr_pipeline-runner_1 exited with code 1

Running docker run -it gcr.io/seqr-project/pipeline-runner:gcloud-prod gives a little more info:

WARNING: The requested image's platform (linux/arm64) does not match the detected host platform (linux/amd64) and no specific platform was requested
exec /entrypoint.sh: exec format error

Running docker run --platform linux/amd64 -it gcr.io/seqr-project/pipeline-runner:gcloud-prod doesn't help, the error then includes the following:

docker: Error response from daemon: image with reference gcr.io/seqr-project/pipeline-runner:gcloud-prod was found but does not match the specified platform: wanted linux/amd64, actual: linux/arm64.

However, the :latest tag, on the other hand (instead of :gcloud-prod), does have the linux/amd64 platform.

I don't know if this difference is intended (e.g., is it intended to run only on arm64 on gcloud?), or accidental (did a developer build on an M1 mac and push/tag). Either way, it would be nice for the LOCAL_INSTALL.md file to refer to an available docker image:tag which can be run in the same platform as the other images.

[hanars]

As discussed here, amd64 is not a supported platform. I'm not sure why the latest tag works for that, but I would not rely on that continuing to work. In the linked discussion are instructions for build your own image for the platform

[bpow]

OK, so I'm not the only one to have run into this :).

I'm fine with building my own image, but I think it would be helpful, if the arm64 architecture is the only one supported for this step, that this be mentioned in the LOCAL_INSTALL.md document under the requirements. It may be that this is not the case when using "Option #1" from that document, but following through "Option #2" in that document, there is nothing that indicates that we need to use a different platform for the pipeline-runner container than for the seqr instance.

Also, the platform of the default pipeline-runner instance looks to have changed just in August of this year-- prior to that, the images being made had amd64 arch:

for i in 20221019_171650 20221007_143623 20220829_165356 20220526_130649 20220526_125536 20220428_114713 20220413_114830 20220318_162331 20220216_152748 20220126_144423 20220126_144230 20220125_165333
do
echo -n "$i "
docker inspect gcr.io/seqr-project/pipeline-runner:$i | grep Arch
done

Gives:

20221019_171650         "Architecture": "arm64",
20221007_143623         "Architecture": "arm64",
20220829_165356         "Architecture": "arm64",
20220526_130649         "Architecture": "amd64",
20220526_125536         "Architecture": "amd64",
20220428_114713         "Architecture": "amd64",
20220413_114830         "Architecture": "amd64",
20220318_162331         "Architecture": "amd64",
20220216_152748         "Architecture": "amd64",
20220126_144423         "Architecture": "amd64",
20220126_144230         "Architecture": "amd64",
20220125_165333         "Architecture": "amd64",

So, I think it would still be helpful at the least to address this more directly in the LOCAL_INSTALL.md document, or even better to ask the pipeline-runner team to produce a separately-tagged image for the same arch that you use for gcr.io/seqr-project/seqr:gcloud-prod so that it is possible to run the docker-compose.yml from a single machine without use of a vm for the separate arch.

[hanars]

You are correct, I did not actually realize that we changed platform support a couple months ago. That was actually a bug, and I have now released an updated image build for amd64. Sorry about that, and thanks for your helpful insight!