diff --git a/.github/workflows/cicd-main.yml b/.github/workflows/cicd-main.yml index 84dfc425cb3..b4b5a9518ff 100644 --- a/.github/workflows/cicd-main.yml +++ b/.github/workflows/cicd-main.yml @@ -285,7 +285,7 @@ jobs: echo "image_tag=$IMAGE_TAG" | tee -a "$GITHUB_OUTPUT" ephemeral-runner-routing: - uses: NVIDIA-NeMo/FW-CI-templates/.github/workflows/_ephemeral_runner_routing.yml@d0878a142525ff79366ca96fbc0b306cd8d64670 # v1.9.0 + uses: NVIDIA-NeMo/FW-CI-templates/.github/workflows/_ephemeral_runner_routing.yml@4fe957d4ea5816af1d67fae130bbab4b434c10b3 # v1.9.2 with: lane: ${{ vars.NEMO_CI_EPHEMERAL_LANE }} legacy_label: ${{ vars.NON_NVIDIA_H100_RUNNER }} @@ -297,7 +297,7 @@ jobs: uses: NVIDIA-NeMo/FW-CI-templates/.github/workflows/_cicd_preflight.yml@d0878a142525ff79366ca96fbc0b306cd8d64670 # v1.9.0 with: default_runner_prefix: ${{ vars.DEFAULT_H100_RUNNER }} - non_nvidia_runner_prefix: ${{ needs.ephemeral-runner-routing.outputs.runner_label }} + non_nvidia_runner_prefix: ${{ needs.ephemeral-runner-routing.outputs.runner_label_base }} default_test_data_path: ${{ vars.DEFAULT_H100_TEST_DATA_PATH }} non_nvidia_test_data_path: ${{ vars.NON_NVIDIA_H100_TEST_DATA_PATH }} default_registry: ${{ vars.DEFAULT_H100_CONTAINER_REGISTRY }} @@ -555,7 +555,7 @@ jobs: dockerfile: docker/Dockerfile platform: linux/amd64 registry: ${{ needs.org-member-pre-flight.outputs.registry }} - runner: ${{ needs.org-member-pre-flight.outputs.runner_prefix }} + runner: ${{ startsWith(needs.org-member-pre-flight.outputs.runner_prefix, 'ephe-v2-') && format('{0}-a{1}', needs.org-member-pre-flight.outputs.runner_prefix, github.run_attempt) || needs.org-member-pre-flight.outputs.runner_prefix }} target: release build-contexts: | nemo-rl=. @@ -626,7 +626,7 @@ jobs: vars.UV_BUILD_CACHE == 'enabled' && needs.build-container.result == 'success' }} - runs-on: ${{ needs.org-member-pre-flight.outputs.runner_prefix }} + runs-on: ${{ startsWith(needs.org-member-pre-flight.outputs.runner_prefix, 'ephe-v2-') && format('{0}-a{1}', needs.org-member-pre-flight.outputs.runner_prefix, github.run_attempt) || needs.org-member-pre-flight.outputs.runner_prefix }} env: REGISTRY: ${{ needs.org-member-pre-flight.outputs.registry }} IMAGE_NAME: ${{ vars.CI_CONTAINER_NAME }} @@ -689,7 +689,7 @@ jobs: matrix: include: - script: Docs_Tests - runner: ${{ needs.org-member-pre-flight.outputs.runner_prefix }} + runner: ${{ startsWith(needs.org-member-pre-flight.outputs.runner_prefix, 'ephe-v2-') && format('{0}-a{1}', needs.org-member-pre-flight.outputs.runner_prefix, github.run_attempt) || needs.org-member-pre-flight.outputs.runner_prefix }} needs: [pre-flight, build-container, org-member-pre-flight, cicd-wait-in-queue] if: >- ${{ @@ -751,7 +751,7 @@ jobs: }} - runs-on: ${{ needs.org-member-pre-flight.outputs.runner_prefix }} + runs-on: ${{ startsWith(needs.org-member-pre-flight.outputs.runner_prefix, 'ephe-v2-') && format('{0}-a{1}', needs.org-member-pre-flight.outputs.runner_prefix, github.run_attempt) || needs.org-member-pre-flight.outputs.runner_prefix }} name: ${{ matrix.script }} steps: - name: Checkout @@ -761,7 +761,7 @@ jobs: env: HF_TOKEN: ${{ secrets.HF_TOKEN }} with: - runner: ${{ needs.org-member-pre-flight.outputs.runner_prefix }} + runner: ${{ startsWith(needs.org-member-pre-flight.outputs.runner_prefix, 'ephe-v2-') && format('{0}-a{1}', needs.org-member-pre-flight.outputs.runner_prefix, github.run_attempt) || needs.org-member-pre-flight.outputs.runner_prefix }} script: ${{ matrix.script }} registry: ${{ needs.org-member-pre-flight.outputs.registry }} test_data_path: ${{ needs.org-member-pre-flight.outputs.test_data_path }} @@ -870,7 +870,7 @@ jobs: matrix: script: ${{ fromJSON(needs.pre-flight.outputs.h100_functional_test_scripts) }} needs: [pre-flight, prepare-megatron-lock, build-container, cicd-unit-tests, functional-test-script-check, org-member-pre-flight, cicd-wait-in-queue] - runs-on: ${{ needs.org-member-pre-flight.outputs.runner_prefix }} + runs-on: ${{ startsWith(needs.org-member-pre-flight.outputs.runner_prefix, 'ephe-v2-') && format('{0}-a{1}', needs.org-member-pre-flight.outputs.runner_prefix, github.run_attempt) || needs.org-member-pre-flight.outputs.runner_prefix }} if: >- ${{ always() && @@ -904,7 +904,7 @@ jobs: env: HF_TOKEN: ${{ secrets.HF_TOKEN }} with: - runner: ${{ needs.org-member-pre-flight.outputs.runner_prefix }} + runner: ${{ startsWith(needs.org-member-pre-flight.outputs.runner_prefix, 'ephe-v2-') && format('{0}-a{1}', needs.org-member-pre-flight.outputs.runner_prefix, github.run_attempt) || needs.org-member-pre-flight.outputs.runner_prefix }} registry: ${{ needs.org-member-pre-flight.outputs.registry }} image: ${{ vars.CI_CONTAINER_NAME }} test_data_path: ${{ needs.org-member-pre-flight.outputs.test_data_path }} @@ -988,7 +988,7 @@ jobs: needs.functional-test-script-check.result == 'success' && !cancelled() }} - runs-on: ${{ needs.org-member-pre-flight.outputs.runner_prefix }} + runs-on: ${{ startsWith(needs.org-member-pre-flight.outputs.runner_prefix, 'ephe-v2-') && format('{0}-a{1}', needs.org-member-pre-flight.outputs.runner_prefix, github.run_attempt) || needs.org-member-pre-flight.outputs.runner_prefix }} name: fast_${{ matrix.script }} steps: - name: Checkout @@ -998,7 +998,7 @@ jobs: env: HF_TOKEN: ${{ secrets.HF_TOKEN }} with: - runner: ${{ needs.org-member-pre-flight.outputs.runner_prefix }} + runner: ${{ startsWith(needs.org-member-pre-flight.outputs.runner_prefix, 'ephe-v2-') && format('{0}-a{1}', needs.org-member-pre-flight.outputs.runner_prefix, github.run_attempt) || needs.org-member-pre-flight.outputs.runner_prefix }} script: ${{ matrix.script }} image-tag: ${{ needs.pre-flight.outputs.image_tag }} registry: ${{ needs.org-member-pre-flight.outputs.registry }}