CI: moxygen adapter build fails due to disk space on GitHub Actions runner

[englishm-cloudflare]

Problem

The moxygen upstream Docker image (ghcr.io/facebookexperimental/moqrelay:latest-amd64) is ~5.6GB. After building moq-rs from source, the GitHub Actions ubuntu-latest runner runs out of disk space when pulling and extracting the moxygen image:

ERROR: failed to register layer: write /tmp/fbcode_builder_getdeps-ZappZbuildZfbcode_builder-root/build/wangle/lib/libwangle_ssl.a: no space left on device

The adapter build step now has continue-on-error: true (#19) so this doesn't block the rest of the workflow, but it means moxygen Docker-based interop tests are effectively skipped. Remote endpoint tests against moxygen still run fine.

Possible solutions

1. Free disk space before builds — Remove pre-installed software on the runner (Android SDK, .NET, Haskell, etc.) to reclaim ~20-30GB. Common pattern in Actions workflows:

   - name: Free disk space
     run: |
       sudo rm -rf /usr/share/dotnet
       sudo rm -rf /usr/local/lib/android
       sudo rm -rf /opt/ghcrunner
       docker system prune -af

2. Use a larger runner — ubuntu-latest-xl or a self-hosted runner with more disk

3. Cache the moxygen image layers — Use docker save/docker load with Actions cache to avoid re-pulling 5.6GB each run

4. Ask moxygen maintainers about a smaller image — The 5.6GB image includes the full fbcode build tree; a multi-stage build producing a slimmer runtime image would help everyone

Context

• First observed in run 21808896604
• Workaround merged in fix(ci): continue on adapter build failure #19
• Related: fix(ci): fix Docker test failures — jq syntax and cert permissions #18 (fixed jq syntax that previously masked this by silently skipping the build)

[englishm-cloudflare]

Another workaround @kixelated just suggested would be to make the adapter pull out the necessary binaries and supporting libraries and then throw away earlier layers. Important to make sure things like libc versions align when using that approach.

[sandarsh]

Let me look into this. Both the client image and the relay image probably suffer from this.

[sandarsh]

@englishm-cloudflare This should be resolved now, as I have switched to using multi-stage builds. Let me know!