From ba1b77597c8ccf5cfef6c6c1f6cb95d3048d71b5 Mon Sep 17 00:00:00 2001
From: Julio Perez <37191411+jperez999@users.noreply.github.com>
Date: Sat, 14 Oct 2023 23:53:38 -0400
Subject: [PATCH] use rapids infra to run testing (#1216)

* use rapids infra to run testing

* remove branch tags from logic

* remove artifact dupe of jobs field

* add back in logic for branch identification

* add gpus flag to container call

* checking gpu with nvidia-smi

* using private container test

* adding correct address for container

* add new container pull logic to all test sets

* consolidate testing because all in one container
---
 .github/workflows/gpu.yml | 43 +++++++++++++++++++++++++--------------
 1 file changed, 28 insertions(+), 15 deletions(-)

diff --git a/.github/workflows/gpu.yml b/.github/workflows/gpu.yml
index 694ff16a9e..7a137c68fe 100644
--- a/.github/workflows/gpu.yml
+++ b/.github/workflows/gpu.yml
@@ -1,22 +1,25 @@
-name: gpu-ci
+name: GPU CI
 
 on:
   workflow_dispatch:
   push:
-    branches: [main]
+    branches:
+      - main
+      - "pull-request/[0-9]+"
     tags:
       - "v[0-9]+.[0-9]+.[0-9]+"
-  pull_request:
-    branches: [main]
-    types: [opened, synchronize, reopened]
-
-concurrency:
-  group: ${{ github.workflow }}-${{ github.ref }}
-  cancel-in-progress: true
 
 jobs:
   gpu-ci:
-    runs-on: 1GPU
+    runs-on: linux-amd64-gpu-p100-latest-1
+    container:
+      image: nvcr.io/nvstaging/merlin/merlin-ci-runner:latest
+      env:
+        NVIDIA_VISIBLE_DEVICES: ${{ env.NVIDIA_VISIBLE_DEVICES }}
+      options: --shm-size=1G
+      credentials:
+        username: $oauthtoken
+        password: ${{ secrets.NGC_TOKEN }}
 
     steps:
       - uses: actions/checkout@v3
@@ -24,6 +27,8 @@ jobs:
           fetch-depth: 0
       - name: Run tests
         run: |
+          nvidia-smi
+          pip install tox
           ref_type=${{ github.ref_type }}
           branch=main
           if [[ $ref_type == "tag"* ]]
@@ -34,17 +39,25 @@ jobs:
           if [[ "${{ github.ref }}" != 'refs/heads/main' ]]; then
               extra_pytest_markers="and changed"
           fi
-          cd ${{ github.workspace }}; PYTEST_MARKERS="unit and not (examples or integration or notebook) and (singlegpu or not multigpu) $extra_pytest_markers" MERLIN_BRANCH=$branch COMPARE_BRANCH=${{ github.base_ref }} tox -e gpu
-
-  tests-examples:
-    runs-on: 1GPU
+          PYTEST_MARKERS="unit and not (examples or integration or notebook) and (singlegpu or not multigpu) $extra_pytest_markers" MERLIN_BRANCH=$branch COMPARE_BRANCH=${{ github.base_ref }} tox -e gpu
 
+  gpu-ci-examples:
+    runs-on: linux-amd64-gpu-p100-latest-1
+    container:
+      image: nvcr.io/nvstaging/merlin/merlin-ci-runner:latest
+      env:
+        NVIDIA_VISIBLE_DEVICES: ${{ env.NVIDIA_VISIBLE_DEVICES }}
+      options: --shm-size=1G
+      credentials:
+        username: $oauthtoken
+        password: ${{ secrets.NGC_TOKEN }}
     steps:
       - uses: actions/checkout@v3
         with:
           fetch-depth: 0
       - name: Run tests
         run: |
+          pip install tox
           ref_type=${{ github.ref_type }}
           branch=main
           if [[ $ref_type == "tag"* ]]
@@ -55,4 +68,4 @@ jobs:
           if [[ "${{ github.ref }}" != 'refs/heads/main' ]]; then
               extra_pytest_markers="and changed"
           fi
-          cd ${{ github.workspace }}; PYTEST_MARKERS="(examples or notebook) $extra_pytest_markers" MERLIN_BRANCH=$branch COMPARE_BRANCH=${{ github.base_ref }} tox -e gpu
+          PYTEST_MARKERS="(examples or notebook) $extra_pytest_markers" MERLIN_BRANCH=$branch COMPARE_BRANCH=${{ github.base_ref }} tox -e gpu