feat(ci): add Telegram notifications and Grafana CI/CD dashboard

- Add notify-start job with Telegram notification for build start
- Add notify-complete job with build status and duration notification
- Push CI metrics to Prometheus Pushgateway for Grafana visualization
- Create CI/CD Grafana dashboard with build status, duration, and history
- Add Pushgateway scrape config to Prometheus

Requires TELEGRAM_BOT_TOKEN, TELEGRAM_CHAT_ID, and PUSHGATEWAY_URL secrets.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
This commit is contained in:
Till-JS 2026-01-26 10:31:17 +01:00
parent ebd0e53c9a
commit 618c58c519
3 changed files with 1100 additions and 0 deletions

View file

@ -328,6 +328,150 @@ jobs:
echo "| storage-web | ${{ steps.changes.outputs.storage-web }} |" >> $GITHUB_STEP_SUMMARY
echo "| telegram-stats-bot | ${{ steps.changes.outputs.telegram-stats-bot }} |" >> $GITHUB_STEP_SUMMARY
# ===========================================
# Notify Build Start
# ===========================================
notify-start:
name: Notify Build Start
runs-on: ubuntu-latest
needs: detect-changes
if: needs.detect-changes.outputs.any-changes == 'true'
steps:
- name: Count services to build
id: count
run: |
COUNT=0
SERVICES=""
if [ "${{ needs.detect-changes.outputs.mana-core-auth }}" == "true" ]; then
COUNT=$((COUNT + 1))
SERVICES="${SERVICES}• mana-core-auth\n"
fi
if [ "${{ needs.detect-changes.outputs.manacore-web }}" == "true" ]; then
COUNT=$((COUNT + 1))
SERVICES="${SERVICES}• manacore-web\n"
fi
if [ "${{ needs.detect-changes.outputs.chat-backend }}" == "true" ]; then
COUNT=$((COUNT + 1))
SERVICES="${SERVICES}• chat-backend\n"
fi
if [ "${{ needs.detect-changes.outputs.chat-web }}" == "true" ]; then
COUNT=$((COUNT + 1))
SERVICES="${SERVICES}• chat-web\n"
fi
if [ "${{ needs.detect-changes.outputs.todo-backend }}" == "true" ]; then
COUNT=$((COUNT + 1))
SERVICES="${SERVICES}• todo-backend\n"
fi
if [ "${{ needs.detect-changes.outputs.todo-web }}" == "true" ]; then
COUNT=$((COUNT + 1))
SERVICES="${SERVICES}• todo-web\n"
fi
if [ "${{ needs.detect-changes.outputs.calendar-backend }}" == "true" ]; then
COUNT=$((COUNT + 1))
SERVICES="${SERVICES}• calendar-backend\n"
fi
if [ "${{ needs.detect-changes.outputs.calendar-web }}" == "true" ]; then
COUNT=$((COUNT + 1))
SERVICES="${SERVICES}• calendar-web\n"
fi
if [ "${{ needs.detect-changes.outputs.clock-backend }}" == "true" ]; then
COUNT=$((COUNT + 1))
SERVICES="${SERVICES}• clock-backend\n"
fi
if [ "${{ needs.detect-changes.outputs.clock-web }}" == "true" ]; then
COUNT=$((COUNT + 1))
SERVICES="${SERVICES}• clock-web\n"
fi
if [ "${{ needs.detect-changes.outputs.contacts-backend }}" == "true" ]; then
COUNT=$((COUNT + 1))
SERVICES="${SERVICES}• contacts-backend\n"
fi
if [ "${{ needs.detect-changes.outputs.contacts-web }}" == "true" ]; then
COUNT=$((COUNT + 1))
SERVICES="${SERVICES}• contacts-web\n"
fi
if [ "${{ needs.detect-changes.outputs.presi-backend }}" == "true" ]; then
COUNT=$((COUNT + 1))
SERVICES="${SERVICES}• presi-backend\n"
fi
if [ "${{ needs.detect-changes.outputs.presi-web }}" == "true" ]; then
COUNT=$((COUNT + 1))
SERVICES="${SERVICES}• presi-web\n"
fi
if [ "${{ needs.detect-changes.outputs.storage-backend }}" == "true" ]; then
COUNT=$((COUNT + 1))
SERVICES="${SERVICES}• storage-backend\n"
fi
if [ "${{ needs.detect-changes.outputs.storage-web }}" == "true" ]; then
COUNT=$((COUNT + 1))
SERVICES="${SERVICES}• storage-web\n"
fi
if [ "${{ needs.detect-changes.outputs.telegram-stats-bot }}" == "true" ]; then
COUNT=$((COUNT + 1))
SERVICES="${SERVICES}• telegram-stats-bot\n"
fi
# Estimate build time (roughly 3-5 min per service, parallel execution)
if [ $COUNT -le 3 ]; then
ESTIMATE="~3-5 min"
elif [ $COUNT -le 8 ]; then
ESTIMATE="~5-8 min"
else
ESTIMATE="~8-12 min"
fi
echo "count=$COUNT" >> $GITHUB_OUTPUT
echo "services<<EOF" >> $GITHUB_OUTPUT
echo -e "$SERVICES" >> $GITHUB_OUTPUT
echo "EOF" >> $GITHUB_OUTPUT
echo "estimate=$ESTIMATE" >> $GITHUB_OUTPUT
- name: Send Telegram notification
env:
TELEGRAM_BOT_TOKEN: ${{ secrets.TELEGRAM_BOT_TOKEN }}
TELEGRAM_CHAT_ID: ${{ secrets.TELEGRAM_CHAT_ID }}
run: |
MESSAGE="🔨 *CI Build Started*
📦 *${{ steps.count.outputs.count }} services* building:
${{ steps.count.outputs.services }}
⏱️ Estimated: ${{ steps.count.outputs.estimate }}
🔗 [View Build](https://github.com/${{ github.repository }}/actions/runs/${{ github.run_id }})"
curl -s -X POST "https://api.telegram.org/bot${TELEGRAM_BOT_TOKEN}/sendMessage" \
-d chat_id="${TELEGRAM_CHAT_ID}" \
-d text="${MESSAGE}" \
-d parse_mode="Markdown" \
-d disable_web_page_preview="true" || true
- name: Push build-started metrics to Prometheus
env:
PUSHGATEWAY_URL: ${{ secrets.PUSHGATEWAY_URL }}
run: |
# Skip if no Pushgateway URL configured
if [ -z "$PUSHGATEWAY_URL" ]; then
echo "PUSHGATEWAY_URL not configured, skipping metrics push"
exit 0
fi
START_EPOCH=$(date +%s)
# Push build-in-progress metric
cat <<EOF | curl --data-binary @- "${PUSHGATEWAY_URL}/metrics/job/ci/run_id/${{ github.run_id }}/branch/${{ github.ref_name }}"
# HELP ci_build_in_progress Whether a CI build is currently in progress
# TYPE ci_build_in_progress gauge
ci_build_in_progress 1
# HELP ci_build_start_timestamp_seconds Unix timestamp when build started
# TYPE ci_build_start_timestamp_seconds gauge
ci_build_start_timestamp_seconds ${START_EPOCH}
# HELP ci_build_services_count Number of services being built
# TYPE ci_build_services_count gauge
ci_build_services_count ${{ steps.count.outputs.count }}
EOF
echo "Build-started metrics pushed to Pushgateway"
# ===========================================
# Validation job - runs on PRs
# ===========================================
@ -858,3 +1002,177 @@ jobs:
tags: ${{ steps.meta.outputs.tags }}
cache-from: type=gha
cache-to: type=gha,mode=max
# ===========================================
# Notify Build Complete
# ===========================================
notify-complete:
name: Notify Build Complete
runs-on: ubuntu-latest
needs:
- detect-changes
- notify-start
- build-mana-core-auth
- build-manacore-web
- build-chat-backend
- build-chat-web
- build-todo-backend
- build-todo-web
- build-calendar-backend
- build-calendar-web
- build-clock-backend
- build-clock-web
- build-contacts-backend
- build-contacts-web
- build-presi-backend
- build-presi-web
- build-storage-backend
- build-storage-web
- build-telegram-stats-bot
if: always() && needs.detect-changes.outputs.any-changes == 'true'
steps:
- name: Calculate duration and status
id: result
run: |
# Get workflow start time from GitHub API
START_TIME="${{ github.event.repository.pushed_at }}"
# Calculate duration in minutes
START_EPOCH=$(date -d "${{ github.event.head_commit.timestamp }}" +%s 2>/dev/null || date +%s)
END_EPOCH=$(date +%s)
DURATION_SECONDS=$((END_EPOCH - START_EPOCH))
DURATION_MINUTES=$((DURATION_SECONDS / 60))
DURATION_REMAINDER=$((DURATION_SECONDS % 60))
echo "duration=${DURATION_MINUTES}m ${DURATION_REMAINDER}s" >> $GITHUB_OUTPUT
# Count successful and failed builds
SUCCESS=0
FAILED=0
SKIPPED=0
FAILED_SERVICES=""
SUCCESS_SERVICES=""
check_job() {
local job_name=$1
local result=$2
if [ "$result" == "success" ]; then
SUCCESS=$((SUCCESS + 1))
SUCCESS_SERVICES="${SUCCESS_SERVICES}✅ ${job_name}\n"
elif [ "$result" == "failure" ]; then
FAILED=$((FAILED + 1))
FAILED_SERVICES="${FAILED_SERVICES}❌ ${job_name}\n"
elif [ "$result" == "skipped" ]; then
SKIPPED=$((SKIPPED + 1))
fi
}
check_job "mana-core-auth" "${{ needs.build-mana-core-auth.result }}"
check_job "manacore-web" "${{ needs.build-manacore-web.result }}"
check_job "chat-backend" "${{ needs.build-chat-backend.result }}"
check_job "chat-web" "${{ needs.build-chat-web.result }}"
check_job "todo-backend" "${{ needs.build-todo-backend.result }}"
check_job "todo-web" "${{ needs.build-todo-web.result }}"
check_job "calendar-backend" "${{ needs.build-calendar-backend.result }}"
check_job "calendar-web" "${{ needs.build-calendar-web.result }}"
check_job "clock-backend" "${{ needs.build-clock-backend.result }}"
check_job "clock-web" "${{ needs.build-clock-web.result }}"
check_job "contacts-backend" "${{ needs.build-contacts-backend.result }}"
check_job "contacts-web" "${{ needs.build-contacts-web.result }}"
check_job "presi-backend" "${{ needs.build-presi-backend.result }}"
check_job "presi-web" "${{ needs.build-presi-web.result }}"
check_job "storage-backend" "${{ needs.build-storage-backend.result }}"
check_job "storage-web" "${{ needs.build-storage-web.result }}"
check_job "telegram-stats-bot" "${{ needs.build-telegram-stats-bot.result }}"
echo "success=$SUCCESS" >> $GITHUB_OUTPUT
echo "failed=$FAILED" >> $GITHUB_OUTPUT
if [ $FAILED -gt 0 ]; then
echo "status=❌ FAILED" >> $GITHUB_OUTPUT
echo "emoji=🔴" >> $GITHUB_OUTPUT
else
echo "status=✅ SUCCESS" >> $GITHUB_OUTPUT
echo "emoji=🟢" >> $GITHUB_OUTPUT
fi
echo "failed_services<<EOF" >> $GITHUB_OUTPUT
echo -e "$FAILED_SERVICES" >> $GITHUB_OUTPUT
echo "EOF" >> $GITHUB_OUTPUT
- name: Send Telegram notification
env:
TELEGRAM_BOT_TOKEN: ${{ secrets.TELEGRAM_BOT_TOKEN }}
TELEGRAM_CHAT_ID: ${{ secrets.TELEGRAM_CHAT_ID }}
run: |
FAILED_INFO=""
if [ "${{ steps.result.outputs.failed }}" != "0" ]; then
FAILED_INFO="
❌ *Failed:*
${{ steps.result.outputs.failed_services }}"
fi
MESSAGE="${{ steps.result.outputs.emoji }} *CI Build Complete*
${{ steps.result.outputs.status }}
📊 *Results:*
✅ Success: ${{ steps.result.outputs.success }}
❌ Failed: ${{ steps.result.outputs.failed }}
⏱️ Duration: ${{ steps.result.outputs.duration }}
${FAILED_INFO}
🔗 [View Build](https://github.com/${{ github.repository }}/actions/runs/${{ github.run_id }})"
curl -s -X POST "https://api.telegram.org/bot${TELEGRAM_BOT_TOKEN}/sendMessage" \
-d chat_id="${TELEGRAM_CHAT_ID}" \
-d text="${MESSAGE}" \
-d parse_mode="Markdown" \
-d disable_web_page_preview="true" || true
- name: Push metrics to Prometheus
if: always()
env:
PUSHGATEWAY_URL: ${{ secrets.PUSHGATEWAY_URL }}
run: |
# Skip if no Pushgateway URL configured
if [ -z "$PUSHGATEWAY_URL" ]; then
echo "PUSHGATEWAY_URL not configured, skipping metrics push"
exit 0
fi
# Calculate duration in seconds for metrics
START_EPOCH=$(date -d "${{ github.event.head_commit.timestamp }}" +%s 2>/dev/null || date +%s)
END_EPOCH=$(date +%s)
DURATION_SECONDS=$((END_EPOCH - START_EPOCH))
# Determine build status (1 = success, 0 = failed)
if [ "${{ steps.result.outputs.failed }}" == "0" ]; then
BUILD_STATUS=1
else
BUILD_STATUS=0
fi
# Push metrics to Pushgateway
cat <<EOF | curl --data-binary @- "${PUSHGATEWAY_URL}/metrics/job/ci/run_id/${{ github.run_id }}/branch/${{ github.ref_name }}"
# HELP ci_build_duration_seconds Duration of CI build in seconds
# TYPE ci_build_duration_seconds gauge
ci_build_duration_seconds ${DURATION_SECONDS}
# HELP ci_build_status Build status (1=success, 0=failed)
# TYPE ci_build_status gauge
ci_build_status ${BUILD_STATUS}
# HELP ci_builds_success_total Number of successful service builds
# TYPE ci_builds_success_total gauge
ci_builds_success_total ${{ steps.result.outputs.success }}
# HELP ci_builds_failed_total Number of failed service builds
# TYPE ci_builds_failed_total gauge
ci_builds_failed_total ${{ steps.result.outputs.failed }}
# HELP ci_build_timestamp_seconds Unix timestamp of build completion
# TYPE ci_build_timestamp_seconds gauge
ci_build_timestamp_seconds ${END_EPOCH}
# HELP ci_build_in_progress Whether a CI build is currently in progress
# TYPE ci_build_in_progress gauge
ci_build_in_progress 0
EOF
echo "Metrics pushed to Pushgateway successfully"