update workflow

Author: kiran-thumma

.github/workflows/enroot-tests.yml

@@ -47,7 +47,7 @@ on:
         type: boolean
         default: false
       testbed_file:
-        description: 'Path to testbed file (e.g. tests/enroot/testbeds/mi325.yaml) - defaults to secrets.TESTBED_FILE'
+        description: 'Path to testbed file (overrides secret-based testbed). If not provided, uses SINGLE_NODE_TESTBED_FILE or MULTI_NODE_TESTBED_FILE secrets (which should contain YAML content).'
         required: false
         type: string
         default: ''
@@ -106,6 +106,44 @@ jobs:
           python3 -m pip install --upgrade pip
           pip install -r tests/enroot/requirements.txt
       
+      - name: Create testbed file from secret
+        if: |
+          ${{
+            github.event_name == 'push' ||
+            (github.event_name == 'workflow_dispatch' && (
+              (matrix.test_name == 'test_single_node_pytorch' && inputs.run_single_node_test == true) ||
+              (matrix.test_name == 'test_multi_node_distributed_pytorch' && inputs.run_multi_node_test == true) ||
+              (matrix.test_name == 'test_multi_node_rccl' && inputs.run_rccl_test == true)
+            ))
+          }}
+        working-directory: tests/enroot
+        env:
+          SINGLE_NODE_TESTBED: ${{ secrets.SINGLE_NODE_TESTBED_FILE }}
+          MULTI_NODE_TESTBED: ${{ secrets.MULTI_NODE_TESTBED_FILE }}
+        run: |
+          # Create testbed files from secrets (secrets contain YAML content)
+          mkdir -p testbed
+          
+          # Write single-node testbed if secret exists
+          if [ -n "$SINGLE_NODE_TESTBED" ]; then
+            printf '%s\n' "$SINGLE_NODE_TESTBED" > testbed/single_node_tb.yml
+            echo "Created testbed/single_node_tb.yml from secret"
+          else
+            echo "[WARNING] SINGLE_NODE_TESTBED_FILE secret is not set"
+          fi
+          
+          # Write multi-node testbed if secret exists  
+          if [ -n "$MULTI_NODE_TESTBED" ]; then
+            printf '%s\n' "$MULTI_NODE_TESTBED" > testbed/multi_node_tb.yml
+            echo "Created testbed/multi_node_tb.yml from secret"
+          else
+            echo "[WARNING] MULTI_NODE_TESTBED_FILE secret is not set"
+          fi
+          
+          # List created testbed files for debugging
+          echo "Testbed files created:"
+          ls -la testbed/ || echo "No testbed directory"
+      
       - name: Run enroot tests
         if: |
           ${{
@@ -123,11 +161,11 @@ jobs:
           
           # Determine testbed file and docker image based on test type and event
           if [ "${{ github.event_name }}" = "push" ]; then
-            # For push events: use test-type-specific secrets and default images from batch scripts
+            # For push events: use test-type-specific testbed files and default images from batch scripts
             if [ "$TEST_NAME" = "test_single_node_pytorch" ]; then
-              TESTBED_FILE="${{ secrets.SINGLE_NODE_TESTBED_FILE }}"
+              TESTBED_FILE="testbed/single_node_tb.yml"
             else
-              TESTBED_FILE="${{ secrets.MULTI_NODE_TESTBED_FILE }}"
+              TESTBED_FILE="testbed/multi_node_tb.yml"
             fi
             DOCKER_IMAGE=""
             NO_INSTALL="false"
@@ -138,9 +176,9 @@ jobs:
               TESTBED_FILE="${{ inputs.testbed_file }}"
             else
               if [ "$TEST_NAME" = "test_single_node_pytorch" ]; then
-                TESTBED_FILE="${{ secrets.SINGLE_NODE_TESTBED_FILE }}"
+                TESTBED_FILE="testbed/single_node_tb.yml"
               else
-                TESTBED_FILE="${{ secrets.MULTI_NODE_TESTBED_FILE }}"
+                TESTBED_FILE="testbed/multi_node_tb.yml"
               fi
             fi
             NO_INSTALL="${{ inputs.no_install }}"
@@ -156,6 +194,17 @@ jobs:
             fi
           fi
           
+          # Validate testbed file exists
+          if [ ! -f "$TESTBED_FILE" ]; then
+            echo "[ERROR] Testbed file not found: $TESTBED_FILE"
+            echo "Please ensure the appropriate secret is set:"
+            echo "  - SINGLE_NODE_TESTBED_FILE for single-node tests"
+            echo "  - MULTI_NODE_TESTBED_FILE for multi-node tests"
+            echo "Or provide a custom testbed_file input via workflow_dispatch."
+            exit 1
+          fi
+          echo "Using testbed file: $TESTBED_FILE"
+          
           # Run RCCL test differently (pytest directly)
           if [ "$TEST_NAME" = "test_multi_node_rccl" ]; then
             # For RCCL test: extract version tag from docker image if provided
@@ -168,10 +217,13 @@ jobs:
               echo "Using RCCL Docker image version: $DOCKER_IMAGE_VERSION"
             fi
             
+            # Convert testbed file to absolute path before changing directory
+            TESTBED_FILE_ABS="$(pwd)/$TESTBED_FILE"
+            
             # Set PYTHONPATH and cd to testsuites directory for pytest
             export PYTHONPATH=$(pwd):$PYTHONPATH
             cd testsuites
-            python3 -m pytest test_enroot.py --testbed "$TESTBED_FILE" -k test_multi_node_rccl --no-install --no-uninstall
+            python3 -m pytest test_enroot.py --testbed "$TESTBED_FILE_ABS" -k test_multi_node_rccl --no-install --no-uninstall
           else
             # For other tests: use run_test.py
             python3 run_test.py "$TEST_NAME" "$DOCKER_IMAGE" "$NO_INSTALL" "$NO_UNINSTALL" "$TESTBED_FILE"