Skip to content

Commit 6d2a123

Browse files
pablo-garaypablo-garay
authored
feat: mcore trigger mbridge (#2340)
Signed-off-by: Pablo Garay <[email protected]>
1 parent 66407fa commit 6d2a123

File tree

1 file changed

+183
-0
lines changed

1 file changed

+183
-0
lines changed

‎.github/workflows/trigger-mbridge-tests.yml‎

Lines changed: 183 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,183 @@
1+
# Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
2+
# SPDX-License-Identifier: Apache-2.0
3+
4+
name: Trigger MBridge Tests
5+
# Remote testing of MBridge from MCore
6+
# Triggers MBridge CI tests with current MCore commit to verify backward compatibility
7+
8+
on:
9+
# Manual trigger only
10+
workflow_dispatch:
11+
inputs:
12+
mbridge_ref:
13+
description: 'MBridge branch/ref to trigger'
14+
required: false
15+
type: string
16+
default: 'main'
17+
run_cicd_main:
18+
description: 'Run cicd-main.yml (full CI/CD)'
19+
required: false
20+
type: boolean
21+
default: true
22+
run_install_test:
23+
description: 'Run install-test.yml (quick install check)'
24+
required: false
25+
type: boolean
26+
default: true
27+
test_suite:
28+
description: 'Test suite to run (for cicd-main)'
29+
required: false
30+
type: choice
31+
options:
32+
- 'all'
33+
- 'unit-only'
34+
- 'functional-only'
35+
default: 'all'
36+
37+
jobs:
38+
# First job: Get MCore commit info (shared by all matrix jobs)
39+
get-mcore-info:
40+
runs-on: ubuntu-latest
41+
outputs:
42+
sha: ${{ steps.mcore_info.outputs.sha }}
43+
short_sha: ${{ steps.mcore_info.outputs.short_sha }}
44+
branch: ${{ steps.mcore_info.outputs.branch }}
45+
repo_url: ${{ steps.mcore_info.outputs.repo_url }}
46+
steps:
47+
- name: Checkout MCore
48+
uses: actions/checkout@v4
49+
with:
50+
fetch-depth: 0
51+
52+
- name: Get MCore commit info
53+
id: mcore_info
54+
run: |
55+
echo "sha=$(git rev-parse HEAD)" >> $GITHUB_OUTPUT
56+
echo "short_sha=$(git rev-parse --short HEAD)" >> $GITHUB_OUTPUT
57+
echo "branch=${GITHUB_REF#refs/heads/}" >> $GITHUB_OUTPUT
58+
59+
# Get repo URL from origin remote, fallback to constructing from github context
60+
REPO_URL=$(git remote get-url origin 2>/dev/null || echo "${{ github.server_url }}/${{ github.repository }}.git")
61+
echo "repo_url=${REPO_URL}" >> $GITHUB_OUTPUT
62+
63+
echo "📦 MCore commit: $(git rev-parse --short HEAD)"
64+
echo "🌿 Branch: ${GITHUB_REF#refs/heads/}"
65+
echo "📍 Repo: ${REPO_URL}"
66+
67+
# Matrix job: Trigger and monitor MBridge workflows in parallel
68+
trigger-and-monitor:
69+
needs: [get-mcore-info]
70+
runs-on: ubuntu-latest
71+
continue-on-error: true # Don't fail workflow if monitoring times out
72+
strategy:
73+
fail-fast: false # Continue other matrix jobs even if one fails
74+
matrix:
75+
include:
76+
- workflow: install-test.yml
77+
name: Install Test
78+
- workflow: cicd-main.yml
79+
name: CI/CD Main
80+
81+
name: ${{ matrix.name }}
82+
83+
steps:
84+
- name: Check if workflow should run
85+
id: should_run
86+
run: |
87+
if [[ "${{ matrix.workflow }}" == "install-test.yml" && "${{ inputs.run_install_test }}" == "true" ]]; then
88+
echo "run=true" >> $GITHUB_OUTPUT
89+
elif [[ "${{ matrix.workflow }}" == "cicd-main.yml" && "${{ inputs.run_cicd_main }}" == "true" ]]; then
90+
echo "run=true" >> $GITHUB_OUTPUT
91+
else
92+
echo "run=false" >> $GITHUB_OUTPUT
93+
echo "⏭️ Skipping ${{ matrix.workflow }} (not enabled)"
94+
fi
95+
96+
- name: Trigger ${{ matrix.workflow }}
97+
if: steps.should_run.outputs.run == 'true'
98+
id: trigger
99+
env:
100+
GH_TOKEN: ${{ secrets.PAT }}
101+
run: |
102+
echo "🚀 Triggering ${{ matrix.workflow }} | MCore: ${{ needs.get-mcore-info.outputs.short_sha }} | MBridge: ${{ inputs.mbridge_ref }}"
103+
104+
gh workflow run ${{ matrix.workflow }} \
105+
--repo NVIDIA-NeMo/Megatron-Bridge --ref ${{ inputs.mbridge_ref }} \
106+
--field mcore_commit=${{ needs.get-mcore-info.outputs.sha }} \
107+
--field mcore_branch=${{ needs.get-mcore-info.outputs.branch }} \
108+
--field mcore_repo=${{ needs.get-mcore-info.outputs.repo_url }} \
109+
--field test_suite=${{ inputs.test_suite }} \
110+
--field triggered_by=mcore-ci
111+
112+
- name: Get run ID
113+
if: steps.should_run.outputs.run == 'true'
114+
id: get_run_id
115+
env:
116+
GH_TOKEN: ${{ secrets.PAT }}
117+
run: |
118+
sleep 10 # Wait for run to appear
119+
RUN_ID=$(gh run list \
120+
--repo NVIDIA-NeMo/Megatron-Bridge \
121+
--workflow=${{ matrix.workflow }} \
122+
--limit 5 \
123+
--json databaseId,createdAt \
124+
--jq "sort_by(.createdAt) | reverse | .[0] | .databaseId")
125+
126+
echo "run_id=${RUN_ID}" >> $GITHUB_OUTPUT
127+
echo "📋 Run ID: ${RUN_ID}"
128+
129+
cat >> $GITHUB_STEP_SUMMARY << EOF
130+
## 🔄 ${{ matrix.name }} Triggered
131+
132+
**MCore:** \`${{ needs.get-mcore-info.outputs.short_sha }}\` | **MBridge:** \`${{ inputs.mbridge_ref }}\` | **Suite:** \`${{ inputs.test_suite }}\`
133+
134+
- 🔄 [${{ matrix.workflow }}](https://github.com/NVIDIA-NeMo/Megatron-Bridge/actions/runs/${RUN_ID}) - Running...
135+
- ⏳ Monitoring every 5 minutes until completion
136+
137+
> **Note:** Tests run without approval when triggered from MCore
138+
EOF
139+
140+
- name: Monitor workflow
141+
if: steps.should_run.outputs.run == 'true'
142+
id: monitor
143+
continue-on-error: true
144+
env:
145+
GH_TOKEN: ${{ secrets.PAT }}
146+
run: |
147+
RUN_ID="${{ steps.get_run_id.outputs.run_id }}"
148+
echo "📊 Monitoring ${{ matrix.workflow }} (Run ID: ${RUN_ID})"
149+
150+
gh run watch ${RUN_ID} --repo NVIDIA-NeMo/Megatron-Bridge --exit-status
151+
152+
CONCLUSION=$(gh run view ${RUN_ID} --repo NVIDIA-NeMo/Megatron-Bridge --json conclusion --jq -r .conclusion)
153+
echo "workflow_status=${CONCLUSION}" >> $GITHUB_ENV
154+
echo "✅ Completed: ${CONCLUSION}"
155+
156+
- name: Report results
157+
if: always() && steps.should_run.outputs.run == 'true'
158+
run: |
159+
CONCLUSION="${{ env.workflow_status || 'unknown' }}"
160+
RUN_ID="${{ steps.get_run_id.outputs.run_id }}"
161+
162+
case "$CONCLUSION" in
163+
"success") ICON="✅"; MSG="passed" ;;
164+
"failure") ICON="❌"; MSG="failed"; EXIT_CODE=1 ;;
165+
"cancelled") ICON="🚫"; MSG="cancelled"; EXIT_CODE=0 ;;
166+
*) ICON="⏳"; MSG="still running or timed out"; EXIT_CODE=0 ;;
167+
esac
168+
169+
cat >> $GITHUB_STEP_SUMMARY << EOF
170+
## 📊 ${{ matrix.name }} Results
171+
172+
### ${ICON} ${{ matrix.workflow }}
173+
**Status:** \`${CONCLUSION}\`
174+
175+
[View full results →](https://github.com/NVIDIA-NeMo/Megatron-Bridge/actions/runs/${RUN_ID})
176+
177+
---
178+
*Triggered from MCore \`${{ needs.get-mcore-info.outputs.short_sha }}\`*
179+
EOF
180+
181+
echo "${ICON} ${{ matrix.name }} ${MSG}"
182+
exit ${EXIT_CODE:-0}
183+

0 commit comments

Comments
 (0)