feat: build the Docker Driver for arm64

[tucksaun]

What this PR does / why we need it:
Add ARM64 build and release of the Docker driver in Drone pipeline

Which issue(s) this PR fixes:
Fixes #5682

Special notes for your reviewer:
I would have loved to have a unified x64 and arm64 build but apparently Docker drivers does not support multi arch images. So instead I went with tweaking the build steps to allow cross building the image for ARM64 and added the instructions to do so in drone.yml.
It seems there's a drift in the images published on Docker Hub versus the one documented for use. I updated it here but this might be wrong.

Checklist

• Reviewed the CONTRIBUTING.md guide (required)
• Documentation added
• Tests updated (no tests for this AFAIK)
• CHANGELOG.md updated
• Changes that require user attention or interaction to upgrade are documented in docs/sources/upgrading/_index.md

[CLAassistant]

All committers have signed the CLA.

[jeschkies]

@tucksaun could you resolve the conflicts?

[jeschkies]

You have to change the jsonnet file uinstead. I'll generate the Drone YML then.

[tucksaun]

done

[tucksaun]

@jeschkies a friendly ping to let you know this PR has been rebased but requires your intervention for the .drone/drone.yml generation 🙂

[tucksaun]

@jeschkies conflicts resolved

[guvmao]

Is there any issue preventing the advancement of this pull request?

[tucksaun]

Is there any issue preventing the advancement of this pull request?

there were new conflicts I just resolved. excluding those, I believe someone (@jeschkies?) has to regenerate the drone.yml now that the jsonnet file is updated.
and that's it?

[jeschkies]

@tucksaun I'm sorry. I've switched teams and have missed GitHub notifications.

I would have loved to have a unified x64 and arm64 build but apparently Docker drivers does not support multi arch images.

I was about to mentioned the multi arch image. How is the arm image tagged differently?

[tucksaun]

IIRC docker plugins does not support multi arch images (at least it did not went I worked on this)

…

On Thu 22 Aug 2024 at 21:21, Karsten Jeschkies ***@***.***> wrote: @tucksaun <https://github.com/tucksaun> I'm sorry. I've switched teams and have missed GitHub notifications. I would have loved to have a unified x64 and arm64 build but apparently Docker drivers does not support multi arch images. I was about to mentioned the multi arch image. How is the arm image tagged differently? — Reply to this email directly, view it on GitHub <#9247 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAGUNZUMY3BU6T4BP5DTQ3DZSZB4LAVCNFSM6AAAAABLF7PFYSVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDGMBVGU3DENRZGI> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[francescor]

Hi, I think your PR would be very helpful

is it a lot of work to solve conflicts to this PR can be accepted?

[tucksaun]

@francescor not much for the Dockerfile so this is done.

however this is a bit different for the pipeline because Drone has been recently removed (see #14273) and it seems like the docker plugin is neither build not released anymore (I can't find any reference to it anyway).

@jeschkies, this means I'm not able to update the pipeline for testing or releasing anymore 🤷

[francescor]

I see this one https://hub.docker.com/r/miacis/loki-docker-driver

is related to this context?

[tucksaun]

I see this one https://hub.docker.com/r/miacis/loki-docker-driver

is related to this context?

I have no idea

[jeschkies]

We'd still have to pushblish the new image

[jeschkies]

I'm fine with landing this but as I understand we would not be able to publish the ARM Docker image as it would override

[francescor]

Hi, we really badly need the ARM images, it will allow us to move all our setup into ARM with considerable savings :)
@tucksaun do you have the ARM image available? we may take it from you until this PM is solved
thanks

[tucksaun]

I'm fine with landing this but as I understand we would not be able to publish the ARM Docker image as it would override

They should be built with different tags. The ARM version uses an explicit tag so this should not impact x64 users

[tucksaun]

We'd still have to pushblish the new image

AFAICT the full releasing of the plugin needs to be restore

[tucksaun]

Hi, we really badly need the ARM images, it will allow us to move all our setup into ARM with considerable savings :)

@tucksaun do you have the ARM image available? we may take it from you until this PM is solved

thanks

I didn't build it for a while (and bad timing as I'm away for conferences) but I can try to have a look in the coming days

[tucksaun]

@francescor I just pushed the plugin at tucksaun/loki-docker-driver:main-arm64

please note I didn't have the opportunity to test it so this comes with no warranty 😁

[francescor]

lovely, thank you!

please note I didn't have the opportunity to test it so this comes with no warranty 😁

yes, of course :)

[francescor]

@francescor I just pushed the plugin at tucksaun/loki-docker-driver:main-arm64

forgive me, @tucksaun how do I pull your image?

docker pull ghcr.io/tucksaun/loki-docker-driver:main-arm64

[tucksaun]

`docker plugin install tucksaun/loki-docker-driver:main-arm64 --alias loki --grant-all-permissions` should make it work

…

On Fri 11 Oct 2024 at 11:25, Fra R ***@***.***> wrote: @francescor <https://github.com/francescor> I just pushed the plugin at tucksaun/loki-docker-driver:main-arm64 forgive me, @tucksaun <https://github.com/tucksaun> how do I pull your image? docker pull ghcr.io/tucksaun/loki-docker-driver:main-arm64 — Reply to this email directly, view it on GitHub <#9247 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAGUNZUNBF6RVH4RY5KQPLDZ26KR7AVCNFSM6AAAAABLF7PFYSVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDIMBXGAYDONRXGI> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[francescor]

thanks @tucksaun

I can install it

# drain node from swarm manager
$ docker node update --availability drain my_arm_node
# then, in that node
$ docker --version 
Docker version 26.1.4, build 5650f9b
$ docker plugin ls
ID        NAME      DESCRIPTION   ENABLED
$ docker ps 
CONTAINER ID   IMAGE     COMMAND   CREATED   STATUS    PORTS     NAMES

# install plugin
$ docker plugin install tucksaun/loki-docker-driver:main-arm64 --alias loki --grant-all-permissions
main-arm64: Pulling from tucksaun/loki-docker-driver
Digest: sha256:fb6b5790f298972624e01ab1bbee6e7a2eb1c4fdd8609426c65fd589d826750f
c4a9207ddf3d: Complete 
Installed plugin tucksaun/loki-docker-driver:main-arm64

but then apparently the "old" plugin keeps showing up :(

$ docker plugin ls 
ID             NAME          DESCRIPTION           ENABLED
2d6ac6b33590   loki:latest   Loki Logging Driver   true

and indeed, I still see that x86_64 emulator:

$ ps aux | grep x86_64-binfmt
root     3288290  3.8  1.2 1939332 202244 ?      Ssl  06:30   0:02 /usr/libexec/qemu-binfmt/x86_64-binfmt-P /bin/docker-driver /bin/docker-driver
root     3288337  0.0  0.0   6020  1920 pts/4    S+   06:31   0:00 grep --color=auto x86_64-binfmt

here the inspect (which correctly shows your plugin)

$ systemctl restart docker
$ docker plugin inspect loki
[
    {
        "Config": {
            "Args": {
                "Description": "",
                "Name": "",
                "Settable": null,
                "Value": null
            },
            "Description": "Loki Logging Driver",
            "DockerVersion": "27.2.0",
            "Documentation": "https://github.com/grafana/loki",
            "Entrypoint": [
                "/bin/docker-driver"
            ],
            "Env": [
                {
                    "Description": "Set log level to output for plugin logs",
                    "Name": "LOG_LEVEL",
                    "Settable": [
                        "value"
                    ],
                    "Value": "info"
                },
                {
                    "Description": "Activate pprof debugging endpoint for the given port.",
                    "Name": "PPROF_PORT",
                    "Settable": [
                        "value"
                    ],
                    "Value": ""
                }
            ],
            "Interface": {
                "Socket": "loki.sock",
                "Types": [
                    "docker.logdriver/1.0"
                ]
            },
            "IpcHost": false,
            "Linux": {
                "AllowAllDevices": false,
                "Capabilities": null,
                "Devices": null
            },
            "Mounts": null,
            "Network": {
                "Type": "host"
            },
            "PidHost": false,
            "PropagatedMount": "",
            "User": {},
            "WorkDir": "",
            "rootfs": {
                "diff_ids": [
                    "sha256:c4a9207ddf3d51da7effcf838f81641ef9c37d4861908f319bcc9e86bb767aa4"
                ],
                "type": "layers"
            }
        },
        "Enabled": true,
        "Id": "2d6ac6b33590d8cd13657347c1e0bdbfcc183ebc5937f2234ea3b50c10091164",
        "Name": "loki:latest",
        "PluginReference": "docker.io/tucksaun/loki-docker-driver:main-arm64",
        "Settings": {
            "Args": [],
            "Devices": [],
            "Env": [
                "LOG_LEVEL=info",
                "PPROF_PORT="
            ],
            "Mounts": []
        }
    }
]

[francescor]

The host is an AWS t4g.xlarge

# uname -a
Linux boat-worker-arm-1d15 6.2.0-1018-aws #18~22.04.1-Ubuntu SMP Wed Jan 10 22:31:58 UTC 2024 aarch64 aarch64 aarch64 GNU/Linux

with installed qemu-user-static & docker-buildx-plugin

If I remove it with apt purge qemu-user-static and then remove and reinstall the plugin, I cannot enable it

[tucksaun]

@francescor I managed to reproduce the issue, I will try to have a look today or tomorrow

[tucksaun]

@francescor should be fixed

[francescor]

Super!

I confirm your loki image does not need x86_64-binfmt (so it can be removed: in Ubuntu 22.04 apt purge qemu-user-static)

thank you so much @tucksaun

I'll provide more feeds in case I hit issues, but I can see logs in our grafana, so it should be OK

Now we can definitively replace swarm node with ARM instances, with great savings!

Thanks!