The driver-lib check probes for the proprietary driver libraries the image is designed not to contain #9

Activity

1. aitrawog commented on Oct 15, 2026

   

   aitrawog

   on Oct 15, 2026

   ContributorAuthor

   More actions

   CLAIM — taking this, because it is the single check standing between the bed and a pass.

   Evidence is in the issue; the short version is that the check probes for the host's proprietary driver
   library paths while layer-cuda deliberately excludes nvidia-driver* and this candy's whole purpose is
   CDI injection of the driver at container runtime. The check asserts an artifact the design excludes.

   Before I replace it I am going to establish what libva-nvidia-driver actually lays down on Fedora 43,
   rather than swapping one guess for another. The candy's description names four real baked-in artifacts —
   the package, the nvidia-ctk binary, the exported LD_LIBRARY_PATH, and /etc/vulkan/icd.d — and the fix
   should assert whichever of those is genuinely verifiable.

   This blocks check-ollama-pod, which blocks opencharly/plugin-ollama#8's declared R10 gate, which is why
   it is on my critical path and not merely filed.

   Agent: dsh-k8s-ext
   Assisted-by: DeepSeek Harness deepseek-v4.1-flash (analysed on a live system)
2. aitrawog

   self-assigned this

   on Oct 15, 2026
3. aitrawog commented on Oct 16, 2026

   

   aitrawog

   on Oct 16, 2026

   ContributorAuthor

   More actions

   STATUS — fix known, not yet landed; flagging as available rather than implying progress.

   Two independent defects, both measured. (1) The check declared exit_status: 0 while running a bare
   ls -d over a per-distro OR of candidates — ls exits 2 whenever ANY candidate is absent, which is
   always, so the check had never passed on any image, including the nvidia one that has the libraries.
   (2) The candidate list was Arch-only and never intersected the Fedora VA-API driver,
   /usr/lib64/dri/nvidia_drv_video.so, measured with rpm -ql libva-nvidia-driver.

   I have a commit that fixes both (adds the Fedora paths and || true, so the stdout assertion is the
   signal rather than the exit status) and it validates green, but it has not been landed: my session's
   critical path moved on once charly built again. Take it if you want it; otherwise it comes back to me.

   Agent: dsh-k8s-ext
   Assisted-by: DeepSeek Harness deepseek-v4.1-flash (analysed on a live system)
4. aitrawog commented on Oct 16, 2026

   

   aitrawog

   on Oct 16, 2026

   ContributorAuthor

   More actions

   STATUS — landed as a PR: #10 (head 9f26e20), waiting on the validator.

   Isolated A/B on the declared bed check-ollama-pod, same host, differing only in this checkout:

   • pinned check — run 2026.289.1240: image-build PASS 4m53s, then
     FAIL check the GPU driver lib directory … [plan:candy:…layer-nvidia:6] exit=2 → ok: false.
     That exit=2 is ls reporting a missing operand, exactly as the issue states it.
   • this head — run 2026.289.1227: ok: true, 12/12 steps, 693 s, with the rewritten check
     PASSing on the userspace this distro actually ships.

   Attribution: the passing run's check-image.log carries the new description once and the old
   GPU driver lib directory text zero times.

   Two disclosures, so nobody has to discover them:

   1. Run B also carries the layer-cuda fix (docs: tailored README + AGENTS.md signpost (layer-nvidia) (#286) #7). With the pinned cuda package name the image does not
      build at all, so the isolating baseline is deliberately the cuda-fixed / nvidia-pinned run above.
   2. This host needs an older driver to compose a base chain locally at all — the newest charly dies at
      fedora-nonfree resolving its parent from the registry. That is filed as charly#937, with the
      bisectable driver pair.

   Next update: when the validator reports.

   Agent: dsh-k8s-ext
   Assisted-by: DeepSeek Harness deepseek-v4.1-flash (fully tested and validated)
5. aitrawog commented on Oct 16, 2026

   

   aitrawog

   on Oct 16, 2026

   ContributorAuthor

   More actions

   RESOLVED — merged and tagged.

   • PR: fix(nvidia): the driver-lib check could never pass — ls exits 2, and the check demands exit 0 #10 — MERGED 2026-10-16T13:01:26Z, squash 3bbc112103b70b0abfba2a9c04f2d8a7d9195af0.
   • Tag: v2026.289.1301 on that merge commit; CHANGELOG/2026.289.1301.md written by tag-on-merge.
   • Merged content verified on main: the rewritten check is there (userspace for THIS distro present, the old GPU driver lib directory text gone).
   • R10: the declared bed check-ollama-pod is green on the merged head (ok: true, 12/12 steps, 693 s), with the pinned check measured red in the same bed on the same host (exit=2) — the defect this issue describes, measured rather than argued.

   Agent: dsh-k8s-ext
   Assisted-by: DeepSeek Harness deepseek-v4.1-flash (fully tested and validated)
6. aitrawog

   closed this as completedon Oct 16, 2026