feat(#8841): Support multiplatform images

[1yuv]

Description

Build multi-platform images
#8841

Code review checklist

• Readable: Concise, well named, follows the style guide, documented if necessary.
• Documented: Configuration and user documentation on cht-docs
• Tested: Unit and/or e2e where appropriate
• Internationalised: All user facing text
• Backwards compatible: Works with existing data and configuration or includes a migration. Any breaking changes documented in the release notes.

Compose URLs

If Build CI hasn't passed, these may 404:

• Core
• CouchDB Single
• CouchDB Cluster

License

The software is provided under AGPL-3.0. Contributions to this project are accepted under the same license.

[1yuv]

Working demo of external contributor: this build and internal contributor

[mrjones-plip]

yay!!! This has been so long in the coming, it's just a real joy to finally see it in the form of PR - nice work!

tl;dr - I think we may need to change const BUILD_PLATFORMS = ['linux/amd64', 'linux/arm64']; to have /v8 at the end like this: const BUILD_PLATFORMS = ['linux/amd64', 'linux/arm64/v8'];

This SO post talks about it, but that talks about linux/arm64/v8 normalizes to linux/arm64 - but what about the other way around?

read below for my logic

---

I'm doing some more research, but in my earlier tests I noted:

I do note that it's not arm64 AND amd64, just amd64

This is because I tested the cht-sentinel:4.6.0-8841-multiplatform-images image. Running a complete test, there's some more disparities, specifically that both sentinel and API are only amd64. They're both based on alpine:3.19 upstream. I'm running tests with, sentinel, for example:

docker run --rm mplatform/mquery public.ecr.aws/medic/cht-sentinel:4.6.0-8841-multiplatform-images

Running an exhaustive test we see that none of the upstream images support arm64 and they all support arm64/v8. All CHT images are the 4.6.0-8841-multiplatform-images tag:

Medic Images	linux/amd64	linux/arm64	linux/arm64/v8
cht-sentinel	Y	N	N
cht-couchdb	Y	Y	N
cht-nginx	Y	Y	N
cht-haproxy-healthcheck	Y	Y	N
cht-haproxy	Y	Y	N
cht-api	Y	N	N
Upstream Images
alpine:3.19 (sentinel)	Y	N	Y
couchdb:3.3.3 (couchdb)	Y	N	Y
nginx:1.25.1-alpine (nginx)	Y	N	Y
python:3.10.13-alpine (haproxy-healthcheck)	Y	N	Y
haproxy:2.6 (haproxy)	Y	N	Y
alpine:3.19 (api)	Y	N	Y

[mrjones-plip]

thanks @1yuv ! You know how long it takes to build new images after a commit/push based on your earlier efforts?

[1yuv]

Hi @mrjones-plip , thanks for extensive test and details on the image and platforms they support. After updating build platfroms to linux/arm64/v8, I can see all medic images now support both amd64 and arm64.

Medic Images	linux/amd64	linux/arm64
cht-sentinel	Y	Y
cht-couchdb	Y	Y
cht-nginx	Y	Y
cht-haproxy-healthcheck	Y	Y
cht-haproxy	Y	Y
cht-api	Y	Y

[mrjones-plip]

Awesome work @1yuv ! I was hoping we'd bet linux/arm64/v8 out of this and not just linux/arm64 🤔 Still, I'll approve this as is - I believe it's good to go. My local tests on Linux show no regressions, which is great.

That said I think we should get at least one other engineer to sign off on this before merging to master.

@dianabarsan or @nydr - any thoughts or concerns here? I kinda wanna get another set of eyes on this given that the new images can have far reaching implications if something goes off the rails. Maybe you can let us know before next Tue (Mar 12th) so we can keep this PR moving?

NB - I'll be out starting Mar 11th, back in the office Mar 18th.

[mrjones-plip]

tentatively approving pending sign off from @dianabarsan or @nydr

[1yuv]

Regarding time taken for building and publishing image, given we're building 2 platform images for 6 services, app compilation time has gone up by nearly 1 minute 30 seconds, specially publish for testing where we're building multi-platform images. This is in comparison with recent single platform build action .

However, the overall build and test time is unaffected or reduced in comparison with some previous builds since we saved time in Publishing branch build by using regctl which detects pre-existing layers and prevent unnecessary i/o and has resulted in faster publishing of images compared to using docker pull, tag, and push.

[dianabarsan]

Cool. Thanks for getting this done, @1yuv !