ray-train: wire in reusable multi-node EFA/NCCL test

- pipeline: add run-efa-test input + efa-test job (uses _reusable.efa-tests.yml,
  2x EFA GPU instances, all_reduce_perf across nodes). Add ci-config job to read
  platform; gate EFA to the ec2 config only (interconnect is identical across
  eks/ec2 — run the expensive 2x p4d test once). release-gate now needs efa-test.
- pr-gpu: add test/efa/** path + efa-test-change filter; run EFA on build or EFA
  test changes (matches pytorch convention)
- autorelease-gpu: run-efa-test: true so releases are gated on multi-node EFA

Author: Eren-Jeager123

.github/workflows/ray-train.autorelease-gpu.yml

@@ -34,5 +34,6 @@ jobs:
     with:
       config-file: ${{ matrix.config_file }}
       tag-suffix: ${{ github.run_id }}
+      run-efa-test: true
       release: true
     secrets: inherit

.github/workflows/ray-train.pipeline.yml

@@ -31,6 +31,11 @@ on:
         required: false
         type: boolean
         default: true
+      run-efa-test:
+        description: "Run multi-node EFA/NCCL test (2x EFA GPU instances)"
+        required: false
+        type: boolean
+        default: false
       release:
         description: "Run release job after tests pass"
         required: false
@@ -42,6 +47,16 @@ on:
         value: ${{ jobs.build.outputs.image-uri }}
 
 jobs:
+  ci-config:
+    runs-on: ubuntu-latest
+    outputs:
+      platform: ${{ steps.parse.outputs.platform }}
+    steps:
+      - uses: actions/checkout@v6
+      - id: parse
+        run: |
+          echo "platform=$(yq '.metadata.platform // ""' '${{ inputs.config-file }}')" >> $GITHUB_OUTPUT
+
   build:
     if: ${{ inputs.build }}
     concurrency:
@@ -97,6 +112,20 @@ jobs:
       config-file: ${{ inputs.config-file }}
       image-uri: ${{ needs.build.outputs.image-uri || '' }}
 
+  # Multi-node EFA/NCCL test (2x EFA GPU instances). The interconnect layer is
+  # identical across the eks/ec2 stages, so run it once — on the ec2 config only —
+  # to avoid doubling the expensive 2x p4d cost.
+  efa-test:
+    if: ${{ always() && !failure() && !cancelled() && inputs.run-efa-test && needs.ci-config.outputs.platform == '' }}
+    needs: [build, ci-config]
+    concurrency:
+      group: ${{ github.workflow }}-efa-test-${{ inputs.config-file }}-${{ github.ref }}
+      cancel-in-progress: false
+    uses: ./.github/workflows/_reusable.efa-tests.yml
+    with:
+      config-file: ${{ inputs.config-file }}
+      image-uri: ${{ needs.build.outputs.image-uri || '' }}
+
   release-gate:
     if: >-
       github.ref == 'refs/heads/main' &&
@@ -105,7 +134,7 @@ jobs:
       inputs.release &&
       !failure() && !cancelled() &&
       needs.build.result == 'success'
-    needs: [build, sanity-test, security-test, telemetry-test]
+    needs: [build, sanity-test, security-test, telemetry-test, efa-test]
     runs-on: ubuntu-latest
     outputs:
       should-release: ${{ steps.check.outputs.should-release }}

.github/workflows/ray-train.pr-gpu.yml

@@ -17,6 +17,7 @@ on:
       - "test/sanity/**"
       - "test/security/data/ecr_scan_allowlist/ray/**"
       - "test/telemetry/**"
+      - "test/efa/**"
       - "!docs/**"
 
 permissions:
@@ -40,6 +41,7 @@ jobs:
       build-change: ${{ steps.changes.outputs.build-change }}
       sanity-test-change: ${{ steps.changes.outputs.sanity-test-change }}
       telemetry-test-change: ${{ steps.changes.outputs.telemetry-test-change }}
+      efa-test-change: ${{ steps.changes.outputs.efa-test-change }}
     steps:
       - uses: actions/checkout@v6
       - uses: dorny/paths-filter@v4
@@ -58,6 +60,8 @@ jobs:
             telemetry-test-change:
               - "scripts/docker/telemetry/**"
               - "test/telemetry/**"
+            efa-test-change:
+              - "test/efa/**"
 
   discover:
     needs: [gatekeeper]
@@ -85,5 +89,6 @@ jobs:
       run-sanity-test: ${{ needs.check-changes.outputs.build-change == 'true' || needs.check-changes.outputs.sanity-test-change == 'true' }}
       run-security-test: ${{ needs.check-changes.outputs.build-change == 'true' }}
       run-telemetry-test: ${{ needs.check-changes.outputs.build-change == 'true' || needs.check-changes.outputs.telemetry-test-change == 'true' }}
+      run-efa-test: ${{ needs.check-changes.outputs.build-change == 'true' || needs.check-changes.outputs.efa-test-change == 'true' }}
       release: false
     secrets: inherit