michaelschiemer/scripts/ci/monitor-workflow-performance.sh

#!/bin/bash
# Monitor Workflow Performance
# Collects metrics about workflow execution times, Gitea load, and resource usage

set -euo pipefail

SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
REPO_ROOT="$(cd "$SCRIPT_DIR/../.." && pwd)"
OUTPUT_DIR="${REPO_ROOT}/monitoring/workflow-metrics"
TIMESTAMP=$(date +%Y%m%d_%H%M%S)
METRICS_FILE="${OUTPUT_DIR}/workflow_metrics_${TIMESTAMP}.json"

# Create output directory
mkdir -p "$OUTPUT_DIR"

# Colors for output
RED='\033[0;31m'
GREEN='\033[0;32m'
YELLOW='\033[1;33m'
BLUE='\033[0;34m'
NC='\033[0m' # No Color

echo -e "${BLUE}📊 Workflow Performance Monitor${NC}"
echo "=================================="
echo ""

# Check if Gitea API credentials are available
GITEA_URL="${GITEA_URL:-https://git.michaelschiemer.de}"
GITEA_TOKEN="${GITEA_TOKEN:-${CI_TOKEN:-}}"
REPO="${GITHUB_REPOSITORY:-michael/michaelschiemer}"

if [ -z "$GITEA_TOKEN" ]; then
    echo -e "${YELLOW}⚠️  GITEA_TOKEN not set, some metrics will be unavailable${NC}"
fi

# Function to get workflow runs from Gitea API
get_workflow_runs() {
    local workflow_name="$1"
    local limit="${2:-10}"
    
    if [ -z "$GITEA_TOKEN" ]; then
        echo "[]"
        return
    fi
    
    local api_url="${GITEA_URL}/api/v1/repos/${REPO}/actions/runs"
    if [ -n "$workflow_name" ]; then
        api_url="${api_url}?workflow=${workflow_name}&limit=${limit}"
    else
        api_url="${api_url}?limit=${limit}"
    fi
    
    curl -sfL \
        -H "Authorization: token ${GITEA_TOKEN}" \
        -H "Accept: application/json" \
        "$api_url" 2>/dev/null || echo "[]"
}

# Function to calculate average duration
calculate_average_duration() {
    local runs_json="$1"
    local total=0
    local count=0
    
    if [ "$runs_json" = "[]" ] || [ -z "$runs_json" ]; then
        echo "0"
        return
    fi
    
    # Extract durations (in seconds) from workflow runs
    # Note: This is a simplified parser - in production, use jq
    echo "$runs_json" | grep -o '"duration":[0-9]*' | grep -o '[0-9]*' | while read -r duration; do
        if [ -n "$duration" ] && [ "$duration" -gt 0 ]; then
            total=$((total + duration))
            count=$((count + 1))
        fi
    done
    
    if [ "$count" -eq 0 ]; then
        echo "0"
    else
        echo "$((total / count))"
    fi
}

# Collect metrics
echo -e "${BLUE}📥 Collecting workflow metrics...${NC}"

# Get recent workflow runs
BUILD_WORKFLOW_RUNS=$(get_workflow_runs "build-image.yml" 20)
DEPLOY_WORKFLOW_RUNS=$(get_workflow_runs "manual-deploy.yml" 10)

# Calculate metrics
BUILD_AVG_DURATION=$(calculate_average_duration "$BUILD_WORKFLOW_RUNS")
DEPLOY_AVG_DURATION=$(calculate_average_duration "$DEPLOY_WORKFLOW_RUNS")

# Get system metrics (if running on server)
SYSTEM_LOAD="unknown"
DOCKER_CONTAINERS="unknown"
GITEA_RUNNER_STATUS="unknown"

if command -v uptime >/dev/null 2>&1; then
    SYSTEM_LOAD=$(uptime | awk -F'load average:' '{print $2}' | awk '{print $1}' | tr -d ' ')
fi

if command -v docker >/dev/null 2>&1; then
    DOCKER_CONTAINERS=$(docker ps --format '{{.Names}}' | wc -l)
    
    if docker ps --format '{{.Names}}' | grep -q "gitea-runner"; then
        GITEA_RUNNER_STATUS="running"
    else
        GITEA_RUNNER_STATUS="stopped"
    fi
fi

# Create metrics JSON
cat > "$METRICS_FILE" <<EOF
{
  "timestamp": "$(date -u +%Y-%m-%dT%H:%M:%SZ)",
  "workflow_metrics": {
    "build_image": {
      "average_duration_seconds": $BUILD_AVG_DURATION,
      "recent_runs": 20
    },
    "manual_deploy": {
      "average_duration_seconds": $DEPLOY_AVG_DURATION,
      "recent_runs": 10
    }
  },
  "system_metrics": {
    "load_average": "$SYSTEM_LOAD",
    "docker_containers": "$DOCKER_CONTAINERS",
    "gitea_runner_status": "$GITEA_RUNNER_STATUS"
  },
  "optimizations": {
    "repository_artifact_enabled": true,
    "helper_script_caching_enabled": true,
    "combined_deployment_playbook": true,
    "exponential_backoff_health_checks": true,
    "concurrency_groups": true
  }
}
EOF

echo -e "${GREEN}✅ Metrics collected${NC}"
echo ""
echo -e "${BLUE}📈 Summary:${NC}"
echo "  Build Workflow Avg Duration: ${BUILD_AVG_DURATION}s"
echo "  Deploy Workflow Avg Duration: ${DEPLOY_AVG_DURATION}s"
echo "  System Load: $SYSTEM_LOAD"
echo "  Docker Containers: $DOCKER_CONTAINERS"
echo "  Gitea Runner: $GITEA_RUNNER_STATUS"
echo ""
echo -e "${BLUE}💾 Metrics saved to:${NC} $METRICS_FILE"
echo ""

# Display recent workflow runs summary
if [ "$BUILD_WORKFLOW_RUNS" != "[]" ] && [ -n "$BUILD_WORKFLOW_RUNS" ]; then
    echo -e "${BLUE}📋 Recent Build Workflow Runs:${NC}"
    echo "$BUILD_WORKFLOW_RUNS" | grep -o '"status":"[^"]*"' | head -5 | sed 's/"status":"//g' | sed 's/"//g' | while read -r status; do
        case "$status" in
            success)
                echo -e "  ${GREEN}✓${NC} Success"
                ;;
            failure)
                echo -e "  ${RED}✗${NC} Failed"
                ;;
            running)
                echo -e "  ${YELLOW}⟳${NC} Running"
                ;;
            *)
                echo -e "  ${BLUE}?${NC} $status"
                ;;
        esac
    done
    echo ""
fi

echo -e "${GREEN}✅ Monitoring complete${NC}"