metrics/scripts/run.sh

#!/bin/bash
set -e

SCRIPT_DIR="$(dirname "$0")"
METRICS_PORT="${METRICS_PORT:-9090}"
NODE_EXPORTER_PORT="${NODE_EXPORTER_PORT:-9100}"
TARGETS_FILE="/tmp/metrics_targets.txt"

# Start node_exporter in background
start_node_exporter() {
    echo "Starting node_exporter on port $NODE_EXPORTER_PORT..."
    node_exporter --web.listen-address=":$NODE_EXPORTER_PORT" &
    NODE_EXPORTER_PID=$!
    sleep 2

    if ! kill -0 $NODE_EXPORTER_PID 2>/dev/null; then
        echo "Failed to start node_exporter"
        exit 1
    fi
    echo "node_exporter running (PID $NODE_EXPORTER_PID)"
}

# Create the aggregator CGI script
create_aggregator_script() {
    cat > /tmp/aggregate.sh << 'SCRIPT'
#!/bin/bash
NODE_EXPORTER_PORT="${NODE_EXPORTER_PORT:-9100}"
TARGETS_FILE="/tmp/metrics_targets.txt"

# Collect all metrics
{
    # System metrics
    echo "# HELP vibestack_up Whether the metrics aggregator is up"
    echo "# TYPE vibestack_up gauge"
    echo "vibestack_up 1"
    echo ""

    # node_exporter metrics
    curl -s "http://localhost:$NODE_EXPORTER_PORT/metrics" 2>/dev/null || true

    # Skill metrics
    if [ -f "$TARGETS_FILE" ]; then
        while IFS=' ' read -r skill_name target || [ -n "$skill_name" ]; do
            [[ "$skill_name" =~ ^#.*$ ]] && continue
            [ -z "$skill_name" ] && continue
            [ -z "$target" ] && continue

            echo ""
            echo "# Metrics from skill: $skill_name"
            curl -s "http://${target}/metrics" 2>/dev/null || true
        done < "$TARGETS_FILE"
    fi
}
SCRIPT
    chmod +x /tmp/aggregate.sh
}

# Serve metrics using socat
serve_metrics() {
    echo "Starting metrics aggregator on port $METRICS_PORT..."

    if ! command -v socat &>/dev/null; then
        echo "Installing socat..."
        apt-get update && apt-get install -y socat
    fi

    # Use socat to handle HTTP requests
    exec socat TCP-LISTEN:$METRICS_PORT,reuseaddr,fork EXEC:"/tmp/metrics_handler.sh"
}

# Create health check script
create_health_script() {
    cat > /tmp/health.sh << 'SCRIPT'
#!/bin/bash
SKILLS_DIR="${SKILLS_DIR:-/skills}"

# Health check endpoints for known skills
declare -A HEALTH_PORTS=(
    ["caddy"]="2019"
    ["ttyd"]="7681"
    ["loki"]="3100"
    ["duckdb"]="8432"
    ["claude"]="8888"
    ["metrics"]="9090"
)

result='{"status":"healthy","skills":{'
first=true

for skill_dir in "$SKILLS_DIR"/*/; do
    skill_name=$(basename "$skill_dir")
    [ "$skill_name" = "supervisor" ] && continue
    [ "$skill_name" = "metrics" ] && continue

    port="${HEALTH_PORTS[$skill_name]}"
    [ -z "$port" ] && continue

    # Check health
    if curl -sf --max-time 2 "http://localhost:$port/health" >/dev/null 2>&1 || \
       curl -sf --max-time 2 "http://localhost:$port/" >/dev/null 2>&1; then
        status="up"
    else
        status="down"
    fi

    $first || result+=','
    first=false
    result+="\"$skill_name\":\"$status\""
done

result+='}}'

# Check if any are down
if echo "$result" | grep -q '"down"'; then
    result=$(echo "$result" | sed 's/"healthy"/"degraded"/')
fi

echo "$result"
SCRIPT
    chmod +x /tmp/health.sh
}

# Create HTTP handler script
create_handler_script() {
    cat > /tmp/metrics_handler.sh << 'HANDLER'
#!/bin/bash
# Read HTTP request
read -r request_line
path=$(echo "$request_line" | cut -d' ' -f2)

while read -r header; do
    [ "$header" = $'\r' ] && break
done

send_response() {
    local content_type="$1"
    local body="$2"
    local body_length=${#body}

    printf "HTTP/1.1 200 OK\r\n"
    printf "Content-Type: %s\r\n" "$content_type"
    printf "Content-Length: %d\r\n" "$body_length"
    printf "Connection: close\r\n"
    printf "\r\n"
    printf "%s" "$body"
}

case "$path" in
    /health)
        health=$(/tmp/health.sh)
        send_response "application/json" "$health"
        ;;
    /metrics|*)
        metrics=$(/tmp/aggregate.sh)
        send_response "text/plain; charset=utf-8" "$metrics"
        ;;
esac
HANDLER
    chmod +x /tmp/metrics_handler.sh
}

# Cleanup on exit
cleanup() {
    echo "Shutting down..."
    [ -n "$NODE_EXPORTER_PID" ] && kill $NODE_EXPORTER_PID 2>/dev/null
    exit 0
}
trap cleanup SIGTERM SIGINT

start_node_exporter
create_aggregator_script
create_health_script
create_handler_script
serve_metrics