They gave me a MacBook Pro at work. I built a docker image locally. I pushed it to Amazon cloud to run. I got “ Compatibility problem! The Mac has Apple M3 chip based on the ARM technology, while the computer in the Amazon cloud has a chip with x64 instruction set. Oops… Docker images are processor-architecture and OS specific. An image built for ARM cannot run on x64 and vice versa. Fortunately, Docker provides tools for building cross-platform and multi-platform images. Unfortunately, this is still an area under development, so the situation changes quickly and the information about it is often outdated or confusing.<\/p>\n “OS” is an operating system, like Windows or Linux.<\/p>\n “Architecture” is a family of compatible processors. Intel and AMD compatible processors are labeled “Platform” is a combination of OS and architecture, e.g. “ Each “regular” image has an OS and an architecture associated with it. These can be viewed using the following command<\/p>\n The output would be something like this: At the time of writing (April 2024) local Docker image storage only supports single-platform images. I.e. each local\u00a0 image will have a specific architecture and platform. However, it is possible to build an image for another platform, not the same as the local computer.<\/p>\n Images built in this way can be pushed to DockerHub.<\/p>\n The trouble is that when pulling an image, one has to know which image to pick for the current platform, which may be inconvenient and error-prone.<\/p>\n It would be nice to have a single image tag that produces an image for the right platform when pulled.<\/p>\n E.g. I want to be able to do<\/p>\n and pull the latest Apache image that works on my machine, without worrying which platform it is.<\/p>\n Docker offers two solutions to this problem: manifests and multi-platform images.<\/p>\n Manifests video: https:\/\/www.youtube.com\/watch?v=AQeGdMuJWIM&ab_channel=AntonPutra<\/a><\/p>\n Manifest is a special entity that acts as a collection of images. Note that at the time of writing manifests cannot be created locally, they always live on a remote server. Thus, manifests can only contain images If I built two platform specific images This way, when someone does Multi-platform images video: https:\/\/www.youtube.com\/watch?v=ZxsAXNCiSZw<\/a><\/p>\n It is possible to build a single image that contains multiple platform-specific sub-images using Then we make this builder default:<\/p>\n Then we build our multi-platform image and push it to the remote server in the same command. Keep in mind, a multi-platform image cannot be stored in the local image storage:<\/p>\n This would create a single entity For an example, inspect httpd image from Apache<\/a>.<\/p>\n Images create using Docker multi-platform support is a new area that is still developing. You need to worry about platforms if your dev platform is different from your hosting platform. Local storage currently does not support multi-platform images (but see https:\/\/docs.docker.com\/storage\/containerd\/<\/a>). Docker offers two ways to create multi-platform images on the server: manifests and <\/p>\n <\/p>\n","protected":false},"excerpt":{"rendered":"invalid exec format<\/code>” error. What gives?<\/p>\nTerminology<\/h1>\n
amd64<\/code> or x86_64<\/code>. 64-bit ARM processors are labeled arm64<\/code>, aarch64<\/code>, or armv8<\/code>. Here amd64<\/code> and arm64<\/code>” are ‘standard’ names, and others are aliases. E.g. if you ask docker to build an aarch64<\/code> image, it obliges, but then reports image architecture as arm64<\/code>‘.<\/p>\nlinux\/amd64<\/code>“.<\/p>\nInspecting Image Platform<\/h1>\n
docker inspect --format='{{.Os}}\/{{.Architecture}}' image_tag<\/i><\/code><\/p>\n'linux\/arm64'<\/code><\/p>\nLocal Images are Single Platform<\/h1>\n
docker build . --platform linux\/amd64 -t ikriv\/testimage:1.0-intel<\/code><\/p>\ndocker build . --platform linux\/arm64 -t ikriv\/testimage:1.0-mac<\/code><\/p>\nMulti-Platform Solutions<\/h1>\n
docker pull httpd:latest<\/code><\/p>\nManifests<\/h1>\n
ikriv\/testimage:1.0-intel<\/code> and ikriv\/testimage:1.0-mac<\/code>, I can create a multi-platform manifest as follows:<\/p>\ndocker push ikriv\/testimage:1.0-intel<\/code><\/p>\ndocker push ikriv\/testimage:1.0-mac
\n<\/code><\/p>\ndocker manifest create ikriv\/testimage:1.0 ikriv\/testimage:1.0-intel ikriv\/testimage:1.0-mac<\/code><\/p>\ndocker manifest annotate ikriv\/testimage:1.0 ikriv\/testimage:1.0-intel --os linux --arch amd64<\/code><\/p>\ndocker manifest annotate ikriv\/testimage:1.0 ikriv\/testimage:1.0-mac --os linux --arch arm64<\/code><\/p>\ndocker manifest push ikriv\/testimage:1.0<\/code><\/p>\ndocker pull ikriv\/testimage:1.0<\/code>, they will get either “intel” or “mac” version, depending on their platform. All three entities, testimage:1.0<\/code>, testimage:1.0-intael<\/code>, and testimage:1.0-mac<\/code> will be visible on Dockerhub.<\/p>\nMulti-Platform Images<\/h1>\n
docker buildx<\/code> command. To do so, one needs to create a named docker builder:<\/p>\ndocker buildx create --name mybuilder<\/code><\/p>\ndocker buildx use mybuilder<\/code><\/p>\ndocker buildx build --platform linux\/amd64,linux\/arm64 -t ikriv\/testbuildx --push .<\/code><\/p>\nikriv\/testbuildx:latest<\/code> on the DockerHub server. It contain images for two platforms. docker pull<\/code> will download the image for the current platform.<\/p>\nWhat’s the difference?<\/h1>\n
docker build --platform<\/code> and using docker buildx<\/code> produce slightly different results. In my test, buildx produced smaller images, but took longer time. The nature of this difference is subject of further research. However, functionally these images behave identically when pulled.<\/p>\nConclusion<\/h1>\n
docker buildx<\/code> command. They produce slightly different results, but seem to be functionally equivalent. Most popular images on DockerHub support multiple platforms, when you pull, you get a version that is compatible with the local machine.<\/p>\n