Container: LTS build overwrites abacus-intel:latest without PEXSI, breaking develop's PEXSI build test

[cheerly-pku]

Summary

develop's "Build extra components with Intel toolchain" job enables PEXSI (-DENABLE_PEXSI=ON) and fails with Could not find PEXSIConfig.cmake, because the shared ghcr.io/deepmodeling/abacus-intel:latest image currently does not contain PEXSI. That image was last pushed by an LTS build, whose Dockerfile.intel does not install PEXSI.

Direct evidence (image inspected)

The config of ghcr.io/deepmodeling/abacus-intel:latest (fetched from ghcr) has:

• label org.opencontainers.image.revision = 62cac09ae7d09ad55bd5c607267f70df74638662
• label org.opencontainers.image.created = 2026-08-23T12:27:04Z

and its build history contains:

RUN ... git clone https://github.com/deepmodeling/abacus-develop.git -b LTS --depth 1 \
    && cd abacus-develop \
    && cmake -B build -DENABLE_DEEPKS=ON -DENABLE_LIBXC=ON -DENABLE_LIBRI=ON -DENABLE_RAPIDJSON=ON ...

i.e. the image is built from the LTS branch and has no PEXSI install step (-DENABLE_PEXSI absent; no GKlib/METIS/ParMETIS/SuperLU_DIST/PEXSI build commands).

Corroborating evidence

1. develop's Dockerfile.intel installs PEXSI. It has a ###### PEXSI PART section (lines 39-141) and builds with -DENABLE_PEXSI=ON (Dockerfile.intel:158). LTS's Dockerfile.intel does not (its build line is -DENABLE_DEEPKS=ON -DENABLE_LIBXC=ON -DENABLE_LIBRI=ON -DENABLE_RAPIDJSON=ON, origin/LTS:Dockerfile.intel:61).

2. Both branches push the same tag abacus-intel:latest. develop's devcontainer.yml:59 and LTS's devcontainer.yml:33 both use type=raw,value=latest; LTS's workflow triggers on branches: [develop, LTS].

3. Timing matches exactly:

   • LTS "build_container_and_push (intel)" job completed (pushed) at 2026-08-23T12:58:05Z.
   • develop's PEXSI job passed right before (32639410989, 12:26, 19m47s) and failed right after (32641514949, 13:09, 2m44s).

Root cause

abacus-intel:latest is shared by the develop and LTS container builds, but only develop's Dockerfile.intel installs PEXSI. An LTS push overwrote the tag with a PEXSI-less image (the develop rebuild failed with no space left on device, so it could not restore PEXSI), breaking develop's PEXSI-enabled CI job.

Suggested fixes

• Give LTS its own image tag (e.g. abacus-intel-lts:latest) instead of overwriting abacus-intel:latest.
• Or install PEXSI in LTS's Dockerfile.intel as well.
• Separately, fix the runner disk exhaustion (no space left on device) in the develop intel build.

[Growl1234]

I believe, this has already been fixed by #7781. It's also duplicate of #7780.

[cheerly-pku]

Thanks @Growl1234 for pointing this out. #7781 indeed fixes it by publishing LTS images as abacus-*-lts, and #7780 already tracks the same problem. Closing this as a duplicate.