github: add github action bot to trigger ci manually.

.github/workflows/action_bot.yaml

@@ -0,0 +1,225 @@
+# Bot command handler for CI permissions.
+# Repository collaborators can comment to control CI:
+#   @xllm-bot run          - Add run-ci label to trigger CI
+#   @xllm-bot rerun        - Cancel and rerun all workflows
+#   @xllm-bot rerun failed - Rerun failed and cancelled jobs
+#   @xllm-bot stop         - Cancel all in-progress workflows
+
+name: Action Bot
+
+on:
+  issue_comment:
+    types: [created]
+
+permissions:
+  contents: read
+  pull-requests: write
+  actions: write
+
+jobs:
+  handle-command:
+    # Only run on PR comments mentioning @xllm-bot.
+    if: |
+      github.event.issue.pull_request &&
+      contains(github.event.comment.body, '@xllm-bot')
+    runs-on: ubuntu-latest
+
+    steps:
+      - name: Check collaborator permission
+        id: check-permission
+        env:
+          GH_TOKEN: ${{ github.token }}
+          REPO: ${{ github.repository }}
+          ACTOR: ${{ github.event.comment.user.login }}
+        run: |
+          echo "Checking if $ACTOR is a collaborator on $REPO..."
+
+          PERMISSION=$(gh api \
+            -H "Accept: application/vnd.github+json" \
+            -H "X-GitHub-Api-Version: 2022-11-28" \
+            "/repos/${REPO}/collaborators/${ACTOR}/permission" \
+            --jq '.permission' 2>/dev/null || true)
+
+          case "$PERMISSION" in
+            admin|maintain|write|triage|read)
+              echo "$ACTOR is authorized with $PERMISSION permission"
+              echo "authorized=true" >> "$GITHUB_OUTPUT"
+              ;;
+            *)
+              echo "$ACTOR is not a collaborator. Repository permission: ${PERMISSION:-none}"
+              echo "authorized=false" >> "$GITHUB_OUTPUT"
+              ;;
+          esac
+
+      - name: Parse command
+        id: parse
+        env:
+          COMMENT_BODY: ${{ github.event.comment.body }}
+        run: |
+          if echo "$COMMENT_BODY" | grep -qi "@xllm-bot rerun failed"; then
+            echo "command=rerun-failed" >> "$GITHUB_OUTPUT"
+          elif echo "$COMMENT_BODY" | grep -qi "@xllm-bot rerun"; then
+            echo "command=rerun" >> "$GITHUB_OUTPUT"
+          elif echo "$COMMENT_BODY" | grep -qi "@xllm-bot stop"; then
+            echo "command=stop" >> "$GITHUB_OUTPUT"
+          elif echo "$COMMENT_BODY" | grep -qi "@xllm-bot run"; then
+            echo "command=run" >> "$GITHUB_OUTPUT"
+          else
+            echo "command=unknown" >> "$GITHUB_OUTPUT"
+          fi
+
+      - name: Handle @xllm-bot run
+        if: steps.check-permission.outputs.authorized == 'true' && steps.parse.outputs.command == 'run'
+        env:
+          GH_TOKEN: ${{ secrets.XLLM_BOT_TOKEN }}
+        run: |
+          echo "Adding run-ci label to PR #${{ github.event.issue.number }}"
+
+          # Add run-ci label.
+          gh pr edit ${{ github.event.issue.number }} \
+            --repo ${{ github.repository }} \
+            --add-label "run-ci"
+
+          # React with thumbs up.
+          gh api \
+            -X POST \
+            "/repos/${{ github.repository }}/issues/comments/${{ github.event.comment.id }}/reactions" \
+            -f content='+1'
+
+          echo "Label added successfully"
+
+      - name: Handle @xllm-bot rerun
+        if: steps.check-permission.outputs.authorized == 'true' && steps.parse.outputs.command == 'rerun'
+        env:
+          GH_TOKEN: ${{ secrets.XLLM_BOT_TOKEN }}
+        run: |
+          echo "Rerunning all jobs for PR #${{ github.event.issue.number }}"
+
+          # Get PR head SHA.
+          PR_SHA=$(gh pr view ${{ github.event.issue.number }} \
+            --repo ${{ github.repository }} \
+            --json headRefOid -q '.headRefOid')
+
+          echo "PR HEAD SHA: $PR_SHA"
+
+          # Cancel in-progress and queued runs first.
+          echo "Cancelling in-progress runs..."
+          gh run list \
+            --repo ${{ github.repository }} \
+            --commit "$PR_SHA" \
+            --json databaseId,status -q '.[] | select(.status == "in_progress" or .status == "queued") | .databaseId' | \
+          while read -r run_id; do
+            if [ -n "$run_id" ]; then
+              echo "Cancelling workflow $run_id..."
+              gh run cancel "$run_id" --repo ${{ github.repository }} || true
+            fi
+          done
+
+          # Wait for cancellations to complete.
+          sleep 2
+
+          # Rerun all workflow runs for this commit.
+          echo "Rerunning all workflows..."
+          gh run list \
+            --repo ${{ github.repository }} \
+            --commit "$PR_SHA" \
+            --json databaseId -q '.[].databaseId' | \
+          while read -r run_id; do
+            if [ -n "$run_id" ]; then
+              echo "Rerunning workflow $run_id..."
+              gh run rerun "$run_id" --repo ${{ github.repository }} || true
+            fi
+          done
+
+          # React with thumbs up.
+          gh api \
+            -X POST \
+            "/repos/${{ github.repository }}/issues/comments/${{ github.event.comment.id }}/reactions" \
+            -f content='+1'
+
+          echo "Rerun triggered successfully"
+
+      - name: Handle @xllm-bot rerun failed
+        if: steps.check-permission.outputs.authorized == 'true' && steps.parse.outputs.command == 'rerun-failed'
+        env:
+          GH_TOKEN: ${{ secrets.XLLM_BOT_TOKEN }}
+        run: |
+          echo "Rerunning failed/cancelled jobs for PR #${{ github.event.issue.number }}"
+
+          # Get PR head SHA.
+          PR_SHA=$(gh pr view ${{ github.event.issue.number }} \
+            --repo ${{ github.repository }} \
+            --json headRefOid -q '.headRefOid')
+
+          echo "PR HEAD SHA: $PR_SHA"
+
+          # Rerun failed and cancelled workflow runs for this commit.
+          # Cancelled jobs are common with fail-fast when one job fails.
+          for STATUS in failure cancelled; do
+            gh run list \
+              --repo ${{ github.repository }} \
+              --commit "$PR_SHA" \
+              --status "$STATUS" \
+              --json databaseId -q '.[].databaseId' | \
+            while read -r run_id; do
+              if [ -n "$run_id" ]; then
+                echo "Rerunning $STATUS workflow $run_id..."
+                gh run rerun "$run_id" --repo ${{ github.repository }} --failed || true
+              fi
+            done
+          done
+
+          # React with thumbs up.
+          gh api \
+            -X POST \
+            "/repos/${{ github.repository }}/issues/comments/${{ github.event.comment.id }}/reactions" \
+            -f content='+1'
+
+          echo "Rerun-failed triggered successfully"
+
+      - name: Handle @xllm-bot stop
+        if: steps.check-permission.outputs.authorized == 'true' && steps.parse.outputs.command == 'stop'
+        env:
+          GH_TOKEN: ${{ secrets.XLLM_BOT_TOKEN }}
+        run: |
+          echo "Stopping all workflows for PR #${{ github.event.issue.number }}"
+
+          # Get PR head SHA.
+          PR_SHA=$(gh pr view ${{ github.event.issue.number }} \
+            --repo ${{ github.repository }} \
+            --json headRefOid -q '.headRefOid')
+
+          echo "PR HEAD SHA: $PR_SHA"
+
+          # Cancel all in-progress and queued runs.
+          gh run list \
+            --repo ${{ github.repository }} \
+            --commit "$PR_SHA" \
+            --json databaseId,status -q '.[] | select(.status == "in_progress" or .status == "queued") | .databaseId' | \
+          while read -r run_id; do
+            if [ -n "$run_id" ]; then
+              echo "Cancelling workflow $run_id..."
+              gh run cancel "$run_id" --repo ${{ github.repository }} || true
+            fi
+          done
+
+          # React with thumbs up.
+          gh api \
+            -X POST \
+            "/repos/${{ github.repository }}/issues/comments/${{ github.event.comment.id }}/reactions" \
+            -f content='+1'
+
+          echo "Stop triggered successfully"
+
+      - name: Unauthorized user
+        if: steps.check-permission.outputs.authorized != 'true' && steps.parse.outputs.command != 'unknown'
+        env:
+          GH_TOKEN: ${{ secrets.XLLM_BOT_TOKEN }}
+        run: |
+          echo "User ${{ github.event.comment.user.login }} is not authorized"
+
+          # React with confused emoji.
+          gh api \
+            -X POST \
+            "/repos/${{ github.repository }}/issues/comments/${{ github.event.comment.id }}/reactions" \
+            -f content='confused'

.github/workflows/build_x86_64_cuda.yaml

@@ -2,21 +2,9 @@ name: xLLM Build x86_64 CUDA
 
 on: 
   workflow_dispatch:
-  push:
-    branches: [main]
-    paths-ignore:
-      - '.github/**'
-      - 'cibuild/**'
-      - 'cmake/**'
-      - 'docs/**'
-      - 'third_party/**'
-      - 'tools/**'
-      - '*.md'
-      - '*.txt'
-      - '*.yml'
   pull_request:
     branches: [main]
-    types: [opened, synchronize, reopened]
+    types: [labeled]
     paths-ignore:
       - 'cmake/**'
       - 'docs/**'
@@ -25,13 +13,6 @@ on:
       - '*.md'
       - '*.txt'
       - '*.yml'
-  pull_request_review:
-    types: [submitted]
-    paths:
-      - '.github/**.yaml'
-      - 'cibuild/**.sh'
-      - 'setup.py'
-      - 'examples/generate.py'
 
 env:
   JOBNAME: xllm-x86_64-cuda-cibuild-${{ github.run_id }}
@@ -41,71 +22,13 @@ concurrency:
   cancel-in-progress: ${{ startsWith(github.ref, 'refs/pull/') }}
 
 jobs:
-  # need to review code first when sensitive files are modified.
-  check-sensitive:
-    runs-on: [self-hosted]
-    outputs:
-      requires_approval: ${{ steps.check_sensitive.outputs.requires_approval }}
-      do_build: ${{ steps.decide.outputs.do_build }}
-    steps:
-      - name: Checkout Code
-        uses: actions/checkout@v4
-        with:
-          fetch-depth: 0  # Ensure we can compare commits
-
-      - name: Install jq
-        run: yum install -y jq
-
-      - name: Check if sensitive files were changed
-        id: check_sensitive
-        run: |
-          sensitive_files=(
-            ".github/**.yaml"
-            "cibuild/**.sh"
-            "setup.py"
-          )
-          changed_files=$(git diff --name-only ${{ github.event.pull_request.base.sha }} ${{ github.sha }})
-          requires_approval="false"
-          while IFS= read -r changed_file; do
-            [[ -z "$changed_file" ]] && continue
-            for pattern in "${sensitive_files[@]}"; do
-              if [[ "$changed_file" == $pattern ]]; then
-                requires_approval="true"
-                break 2
-              fi
-          done
-          done < <(git diff --name-only "${{ github.event.pull_request.base.sha }}" "${{ github.sha }}")
-          echo "requires_approval=$requires_approval" >> $GITHUB_OUTPUT
-
-      - name: Decide whether to check build
-        id: decide
-        run: |
-          event="${{ github.event_name }}"
-          if [[ "$event" == "workflow_dispatch" || "$event" == "push" ]]; then
-            echo "do_build=true" >> $GITHUB_OUTPUT
-          elif [[ "$event" == "pull_request" ]]; then
-            if  [ "${{ steps.check_sensitive.outputs.requires_approval }}" == "true" ]; then
-              echo "do_build=false" >> $GITHUB_OUTPUT
-            else
-              echo "do_build=true" >> $GITHUB_OUTPUT
-            fi
-          elif [[ "$event" == "pull_request_review" ]]; then
-            # Since pull_request_review now only triggers when sensitive files are modified,
-            # we only need to check if the review is approved
-            if [[ "${{ github.event.review.state }}" == "approved" ]]; then
-              echo "do_build=true" >> $GITHUB_OUTPUT
-            else
-              echo "do_build=false" >> $GITHUB_OUTPUT
-            fi
-          else
-            echo "do_build=false" >> $GITHUB_OUTPUT
-          fi
-
   build:
-    needs: check-sensitive
     if: >
-      (github.event_name == 'workflow_dispatch' || github.event_name == 'push') ||
-      needs.check-sensitive.outputs.do_build == 'true'
+      github.event_name == 'workflow_dispatch' ||
+      (
+        github.event_name == 'pull_request' &&
+        github.event.label.name == 'run-ci'
+      )
     runs-on: [self-hosted]
     steps:
     - name: Checkout Code