From 0cd762892b50a38c66fe53440ecb7abecb4dde23 Mon Sep 17 00:00:00 2001
From: jgrusewski <jgrusewski@localhost>
Date: Wed, 25 Feb 2026 14:59:24 +0100
Subject: [PATCH] fix(ci): exclude ml crate from CPU tests, add manual GPU test
 job

52 ml tests require real CUDA (DQN, TFT, ensemble, flash_attention,
QAT, trainers). Exclude ml from workspace tests on CPU CI nodes.
Add test-gpu job (manual) for running ml tests on GPU hardware.

Also upgraded ci-build pool from GP1-XS to GP1-S (8 vCPUs, 32GB).

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---
 .gitlab-ci.yml | 16 ++++++++++++++--
 1 file changed, 14 insertions(+), 2 deletions(-)

diff --git a/.gitlab-ci.yml b/.gitlab-ci.yml
index ef1566703..c623191be 100644
--- a/.gitlab-ci.yml
+++ b/.gitlab-ci.yml
@@ -81,10 +81,22 @@ test:
   stage: test
   needs: [check]
   script:
-    # Stub libcuda.so.1 lets CUDA-linked binaries load; GPU tests skipped (run on gpu nodes)
+    # Stub libcuda.so.1 lets CUDA-linked binaries load on GPU-less nodes
     - ln -sf /usr/local/cuda/lib64/stubs/libcuda.so /usr/local/cuda/lib64/stubs/libcuda.so.1 || true
     - export LD_LIBRARY_PATH="/usr/local/cuda/lib64/stubs:${LD_LIBRARY_PATH}"
-    - cargo test --workspace --lib -- --skip model_loader::tests
+    # Run all workspace tests EXCEPT ml crate (52 CUDA-dependent tests need real GPU)
+    - cargo test --workspace --exclude ml --lib -- --skip model_loader::tests
+
+# GPU test job — runs ml crate tests on real CUDA hardware
+# TODO: Configure GPU runner with node_selector targeting gpu-inference pool
+test-gpu:
+  extends: .rust-base
+  stage: test
+  needs: [check]
+  when: manual
+  allow_failure: true
+  script:
+    - cargo test -p ml --lib
 
 # --------------------------------------------------------------------------
 # Stage 3: Build + push service images (main only, Kaniko → GitLab registry)