breakpilot-lehrer/voice-service/scripts/run_bqas.sh

#!/bin/bash
# BQAS Local Runner - Lokale Alternative zu GitHub Actions
# Fuehrt BQAS Tests aus und benachrichtigt bei Fehlern

set -e

# Konfiguration
VOICE_SERVICE_DIR="/Users/benjaminadmin/Projekte/breakpilot-pwa/voice-service"
VOICE_SERVICE_URL="${BQAS_SERVICE_URL:-http://localhost:8091}"
LOG_DIR="/var/log/bqas"
LOG_FILE="${LOG_DIR}/bqas.log"
REGRESSION_THRESHOLD="${BQAS_REGRESSION_THRESHOLD:-0.1}"

# Farben fuer Output
RED='\033[0;31m'
GREEN='\033[0;32m'
YELLOW='\033[0;33m'
BLUE='\033[0;34m'
NC='\033[0m' # No Color

# Argumente
QUICK_MODE=false
GOLDEN_ONLY=false
RAG_ONLY=false
SILENT=false

usage() {
    echo "Usage: $0 [OPTIONS]"
    echo ""
    echo "Options:"
    echo "  --quick       Nur schnelle Golden Tests (fuer Git Hooks)"
    echo "  --golden      Nur Golden Suite"
    echo "  --rag         Nur RAG Suite"
    echo "  --silent      Keine Desktop-Benachrichtigungen"
    echo "  --help        Diese Hilfe anzeigen"
    echo ""
    echo "Umgebungsvariablen:"
    echo "  BQAS_SERVICE_URL         Voice Service URL (default: http://localhost:8091)"
    echo "  BQAS_REGRESSION_THRESHOLD Regression Schwelle (default: 0.1)"
}

while [[ $# -gt 0 ]]; do
    case $1 in
        --quick)
            QUICK_MODE=true
            shift
            ;;
        --golden)
            GOLDEN_ONLY=true
            shift
            ;;
        --rag)
            RAG_ONLY=true
            shift
            ;;
        --silent)
            SILENT=true
            shift
            ;;
        --help)
            usage
            exit 0
            ;;
        *)
            echo "Unbekannte Option: $1"
            usage
            exit 1
            ;;
    esac
done

# Logging-Funktion
log() {
    local level=$1
    local message=$2
    local timestamp=$(date '+%Y-%m-%d %H:%M:%S')

    # Log-Verzeichnis erstellen falls nicht vorhanden
    if [ -d "$LOG_DIR" ]; then
        echo "${timestamp} [${level}] ${message}" >> "$LOG_FILE"
    fi

    # Console Output
    case $level in
        INFO)
            echo -e "${BLUE}[INFO]${NC} ${message}"
            ;;
        SUCCESS)
            echo -e "${GREEN}[SUCCESS]${NC} ${message}"
            ;;
        WARNING)
            echo -e "${YELLOW}[WARNING]${NC} ${message}"
            ;;
        ERROR)
            echo -e "${RED}[ERROR]${NC} ${message}"
            ;;
    esac
}

# Benachrichtigung senden
notify() {
    local title=$1
    local message=$2
    local is_error=${3:-false}

    if [ "$SILENT" = true ]; then
        return
    fi

    # macOS Desktop-Benachrichtigung
    if [ "$is_error" = true ]; then
        osascript -e "display notification \"${message}\" with title \"${title}\" sound name \"Basso\"" 2>/dev/null || true
    else
        osascript -e "display notification \"${message}\" with title \"${title}\"" 2>/dev/null || true
    fi
}

# Python-Notifier aufrufen (falls vorhanden)
notify_python() {
    local status=$1
    local message=$2
    local details=$3

    if [ -f "${VOICE_SERVICE_DIR}/bqas/notifier.py" ]; then
        python3 "${VOICE_SERVICE_DIR}/bqas/notifier.py" \
            --status "$status" \
            --message "$message" \
            --details "$details" 2>/dev/null || true
    fi
}

# Pruefen ob Service laeuft
check_service() {
    log "INFO" "Pruefe Voice Service Verfuegbarkeit..."

    local health_url="${VOICE_SERVICE_URL}/health"
    local response

    response=$(curl -s -o /dev/null -w "%{http_code}" "$health_url" 2>/dev/null) || response="000"

    if [ "$response" = "200" ]; then
        log "SUCCESS" "Voice Service erreichbar"
        return 0
    else
        log "WARNING" "Voice Service nicht erreichbar (HTTP $response)"
        return 1
    fi
}

# Regression Check durchfuehren
check_regression() {
    log "INFO" "Pruefe auf Score-Regression..."

    local regression_url="${VOICE_SERVICE_URL}/api/v1/bqas/regression-check?threshold=${REGRESSION_THRESHOLD}"
    local response

    response=$(curl -s "$regression_url" 2>/dev/null) || {
        log "WARNING" "Regression-Check fehlgeschlagen"
        return 1
    }

    local is_regression
    is_regression=$(echo "$response" | python3 -c "import sys,json; print(json.load(sys.stdin).get('is_regression', False))" 2>/dev/null) || is_regression="False"

    if [ "$is_regression" = "True" ]; then
        local delta
        delta=$(echo "$response" | python3 -c "import sys,json; print(json.load(sys.stdin).get('delta', 0))" 2>/dev/null) || delta="unknown"
        log "ERROR" "Regression erkannt! Score-Abfall: ${delta}"
        return 1
    else
        log "SUCCESS" "Keine Regression erkannt"
        return 0
    fi
}

# Tests ausfuehren
run_tests() {
    local test_type=$1
    local test_path=$2
    local exit_code=0

    log "INFO" "Starte ${test_type} Tests..."

    cd "$VOICE_SERVICE_DIR"

    # Aktiviere venv falls vorhanden
    if [ -f "venv/bin/activate" ]; then
        source venv/bin/activate
    fi

    # pytest ausfuehren
    if python3 -m pytest "$test_path" -v --tb=short 2>&1 | tee -a "$LOG_FILE"; then
        log "SUCCESS" "${test_type} Tests bestanden"
        exit_code=0
    else
        log "ERROR" "${test_type} Tests fehlgeschlagen"
        exit_code=1
    fi

    return $exit_code
}

# Hauptlogik
main() {
    local start_time=$(date +%s)
    local golden_exit=0
    local rag_exit=0
    local regression_exit=0
    local service_available=false

    log "INFO" "=========================================="
    log "INFO" "BQAS Local Runner gestartet"
    log "INFO" "=========================================="

    # Service-Check (optional, Tests koennen auch offline laufen)
    if check_service; then
        service_available=true
    fi

    # Quick Mode: Nur schnelle Tests
    if [ "$QUICK_MODE" = true ]; then
        log "INFO" "Quick Mode - nur schnelle Golden Tests"
        run_tests "Golden (Quick)" "tests/bqas/test_golden.py -k 'not slow'" || golden_exit=1
    else
        # Vollstaendige Test-Ausfuehrung
        if [ "$RAG_ONLY" = false ]; then
            run_tests "Golden" "tests/bqas/test_golden.py" || golden_exit=1
        fi

        if [ "$GOLDEN_ONLY" = false ]; then
            run_tests "RAG" "tests/bqas/test_rag.py" || rag_exit=1
        fi

        # Regression-Check nur wenn Service verfuegbar
        if [ "$service_available" = true ]; then
            check_regression || regression_exit=1
        fi
    fi

    # Zusammenfassung
    local end_time=$(date +%s)
    local duration=$((end_time - start_time))

    log "INFO" "=========================================="
    log "INFO" "BQAS Run abgeschlossen (${duration}s)"
    log "INFO" "=========================================="

    # Ergebnis ermitteln
    local total_failures=$((golden_exit + rag_exit + regression_exit))

    if [ $total_failures -eq 0 ]; then
        log "SUCCESS" "Alle Tests bestanden!"
        notify "BQAS" "Alle Tests bestanden" false
        notify_python "success" "Alle Tests bestanden" "Dauer: ${duration}s"
        return 0
    else
        local failure_details=""
        [ $golden_exit -ne 0 ] && failure_details="${failure_details}Golden Tests fehlgeschlagen. "
        [ $rag_exit -ne 0 ] && failure_details="${failure_details}RAG Tests fehlgeschlagen. "
        [ $regression_exit -ne 0 ] && failure_details="${failure_details}Regression erkannt. "

        log "ERROR" "Tests fehlgeschlagen: ${failure_details}"
        notify "BQAS Alert" "$failure_details" true
        notify_python "failure" "Tests fehlgeschlagen" "$failure_details"
        return 1
    fi
}

# Script ausfuehren
main