#!/usr/bin/env bash
# =============================================================================
# run-e2e-test.sh — Test E2E distribué ja4-platform
#
# Architecture :
#   3 VMs endpoint (centos8 / rocky9 / rocky10) : nginx + ja4ebpf
#   1 VM analysis (192.168.42.10) : ClickHouse + bot-detector + dashboard
#   1 VM traffic : curl-impersonate + httpx (génération de trafic externe)
#
# Pipeline testé :
#   trafic host → endpoints → ja4ebpf → ClickHouse central →
#   MV agrégation → bot-detector ML → dashboard API
#
# Usage :
#   make e2e-up && make test-e2e
#   TRAFFIC_COUNT=100 make test-e2e-quick
#   ./run-e2e-test.sh --hits 1000 --http-ratio 0.3 --dns 2 --tls 1.2,1.3
#
# Paramètres :
#   --hits N          Nombre total de requêtes par VM (déf. 500)
#   --http-ratio R    Ratio HTTP/HTTPS : R = part HTTP (0.0–1.0, déf. 0.2)
#   --dns N           Nombre de hostnames SNI utilisés (1–4, déf. 4)
#   --tls VERS        Versions TLS à utiliser, séparées par virgules (déf. 1.2,1.3)
#   --src-ips N       Nombre d'IPs sources par VM (alias sur eth0 dans la VM, déf. 1)
#   --keep-analysis   Conserver la stack analysis (CH + dashboard) après le test
#   --up              Démarrer la stack analysis et les endpoints, puis s'arrêter
#                     (pas de trafic ni vérification — utile pour inspection manuelle)
# =============================================================================
set -euo pipefail

VM_DIR="$(cd "$(dirname "$0")" && pwd)"
PROJECT_ROOT="$(cd "$VM_DIR/../.." && pwd)"
ANALYSIS_IP="192.168.42.10"
ENDPOINT_VMS="centos8 rocky9 rocky10"
TRAFFIC_VM="traffic"
ALL_VMS="centos8 rocky9 rocky10 analysis traffic"
STACK="all-ips"

# ── Paramètres par défaut (surchARGEABLES par CLI ou env) ──
TRAFFIC_COUNT="${TRAFFIC_COUNT:-500}"
HTTP_RATIO="${HTTP_RATIO:-0.2}"
DNS_COUNT="${DNS_COUNT:-4}"
TLS_VERSIONS="${TLS_VERSIONS:-1.2,1.3}"
SRC_IP_COUNT="${SRC_IP_COUNT:-1}"
KEEP_ANALYSIS="${KEEP_ANALYSIS:-false}"
UP_ONLY="${UP_ONLY:-false}"

# ── Parsing des arguments CLI ──
usage() {
    echo "Usage: $0 [OPTIONS]"
    echo ""
    echo "Options :"
    echo "  --hits N         Nombre de requêtes par VM (déf. 500)"
    echo "  --http-ratio R   Ratio HTTP (0=100% HTTPS, 1=100% HTTP, déf. 0.2)"
    echo "  --dns N          Nombre de hostnames SNI utilisés (1-4, déf. 4)"
    echo "  --tls VERS       Versions TLS, séparées par virgules (déf. 1.2,1.3)"
    echo "  --src-ips N      Nombre d'IPs sources par VM (déf. 1)"
    echo "  --keep-analysis  Conserver la stack analysis (CH + dashboard) après le test"
    echo "  --up             Démarrer stack analysis + endpoints, puis s'arrêter"
    echo "  -h, --help       Afficher cette aide"
    exit 0
}

while [[ $# -gt 0 ]]; do
    case "$1" in
        --hits)        TRAFFIC_COUNT="$2"; shift 2 ;;
        --http-ratio)  HTTP_RATIO="$2";   shift 2 ;;
        --dns)         DNS_COUNT="$2";     shift 2 ;;
        --tls)           TLS_VERSIONS="$2";  shift 2 ;;
        --src-ips)       SRC_IP_COUNT="$2";  shift 2 ;;
        --keep-analysis) KEEP_ANALYSIS="true"; shift ;;
        --up)            UP_ONLY="true"; shift ;;
        -h|--help)       usage ;;
        *) echo "Option inconnue : $1"; usage ;;
    esac
done

# ── Validation ──
if ! python3 -c "exit(0 if 0.0 <= ${HTTP_RATIO} <= 1.0 else 1)" 2>/dev/null; then
    echo "ERREUR: --http-ratio doit être entre 0.0 et 1.0" >&2; exit 1
fi
if [ "$DNS_COUNT" -lt 1 ] || [ "$DNS_COUNT" -gt 4 ]; then
    echo "ERREUR: --dns doit être entre 1 et 4" >&2; exit 1
fi

# ── Résolution des versions TLS pour curl ──
CURL_TLS_FLAGS=""
for v in $(echo "$TLS_VERSIONS" | tr ',' ' '); do
    case "$v" in
        1.0) CURL_TLS_FLAGS="$CURL_TLS_FLAGS --tlsv1.0" ;;
        1.1) CURL_TLS_FLAGS="$CURL_TLS_FLAGS --tlsv1.1" ;;
        1.2) CURL_TLS_FLAGS="$CURL_TLS_FLAGS --tlsv1.2" ;;
        1.3) CURL_TLS_FLAGS="$CURL_TLS_FLAGS --tlsv1.3" ;;
        *)   echo "[e2e] WARN: Version TLS '$v' non reconnue, ignorée" ;;
    esac
done
[ -z "$CURL_TLS_FLAGS" ] && CURL_TLS_FLAGS="--tlsv1.2 --tlsv1.3" && TLS_VERSIONS="1.2,1.3"
CURL_TLS_FLAGS="${CURL_TLS_FLAGS# }"  # trim leading space

# Nombre de requêtes HTTP vs HTTPS dérivés du ratio
HTTPS_COUNT=$(python3 -c "print(int(${TRAFFIC_COUNT} * (1 - ${HTTP_RATIO})))")
HTTP_COUNT=$(python3 -c "print(int(${TRAFFIC_COUNT} * ${HTTP_RATIO}))")

GREEN='\033[0;32m'; RED='\033[0;31m'; YELLOW='\033[1;33m'; CYAN='\033[0;36m'
RESET='\033[0m'; BOLD='\033[1m'
PASS_COUNT=0; FAIL_COUNT=0; WARN_COUNT=0

log()  { echo -e "${BOLD}[e2e]${RESET} $(date +%H:%M:%S) $*"; }
pass() { echo -e "  ${GREEN}PASS${RESET}  $*"; ((PASS_COUNT++)) || true; }
fail() { echo -e "  ${RED}FAIL${RESET}  $*"; ((FAIL_COUNT++)) || true; }
warn() { echo -e "  ${YELLOW}WARN${RESET}   $*"; ((WARN_COUNT++)) || true; }

# Requête ClickHouse via HTTP API sur la VM analysis
ch_query() {
    curl -sf "http://${ANALYSIS_IP}:8123/" -d "$1" 2>/dev/null || echo ""
}

# ═════════════════════════════════════════════════════════════════════════════
# Phase 0 : Setup environnement
# ═════════════════════════════════════════════════════════════════════════════
phase0_setup() {
    echo ""
    echo "╔══════════════════════════════════════════════════════════╗"
    echo "║  Phase 0 : Setup environnement                          ║"
    echo "╚══════════════════════════════════════════════════════════╝"
    echo ""

    cd "$VM_DIR"

    # Vérifier que les VMs existent
    for vm in $ALL_VMS; do
        if ! vagrant status "$vm" 2>/dev/null | grep -q "running"; then
            log "Démarrage de $vm..."
            vagrant up "$vm" 2>&1 | tail -5
        else
            log "$vm déjà en cours d'exécution"
        fi
    done

    # Rsync des fichiers vers toutes les VMs
    log "Synchronisation des fichiers..."
    for vm in $ALL_VMS; do
        vagrant rsync "$vm" 2>&1 | tail -1
    done

    # Découvrir les IPs eth0 des endpoints et filtrer ceux sans ja4ebpf
    declare -A VM_IPS
    ACTIVE_ENDPOINTS=""
    for vm in $ENDPOINT_VMS; do
        # Vérifier que ja4ebpf est installé
        if ! vagrant ssh "$vm" -- 'which ja4ebpf' 2>/dev/null | grep -q ja4ebpf; then
            warn "$vm ignoré — ja4ebpf non installé"
            continue
        fi
        VM_IPS[$vm]=$(vagrant ssh "$vm" -- 'ip -4 addr show eth0' 2>/dev/null \
            | awk '/inet / {sub(/\/.*/, "", $2); print $2; exit}')
        if [ -z "${VM_IPS[$vm]}" ]; then
            fail "Impossible d'obtenir l'IP eth0 de $vm"
        else
            log "  $vm : ${VM_IPS[$vm]}"
            ACTIVE_ENDPOINTS="$ACTIVE_ENDPOINTS $vm"
        fi
    done
    ENDPOINT_VMS=$(echo $ACTIVE_ENDPOINTS)  # overwrite with active VMs only

    # Exporter les IPs pour les phases suivantes
    ENDPOINT_IPS_ARRAY=()
    for vm in $ENDPOINT_VMS; do
        ENDPOINT_IPS_ARRAY+=("${VM_IPS[$vm]}")
    done

    pass "Endpoints actifs : ${ENDPOINT_VMS}"
}

# ═════════════════════════════════════════════════════════════════════════════
# Phase 1 : Démarrage de la stack analysis
# ═════════════════════════════════════════════════════════════════════════════
phase1_analysis() {
    echo ""
    echo "╔══════════════════════════════════════════════════════════╗"
    echo "║  Phase 1 : Stack analysis (ClickHouse + ML + Dashboard) ║"
    echo "╚══════════════════════════════════════════════════════════╝"
    echo ""

    cd "$VM_DIR"

    # Purger les volumes pour garantir une DB vide avant le test
    log "Purge des volumes Docker pour DB vierge..."
    vagrant ssh analysis -- "cd /ja4-platform/tests/vm/analysis && docker compose down -v --remove-orphans" 2>&1 | tail -5

    # Démarrer la stack analysis via docker compose
    log "Démarrage de la stack analysis sur la VM..."
    vagrant ssh analysis -- "cd /ja4-platform/tests/vm/analysis && docker compose up -d --build" 2>&1 | tail -20

    # Attendre ClickHouse
    log "Attente ClickHouse (max 120s)..."
    for i in $(seq 1 60); do
        if curl -sf "http://${ANALYSIS_IP}:8123/ping" >/dev/null 2>&1; then
            pass "ClickHouse prêt (${i}*2s)"
            break
        fi
        sleep 2
    done

    # Vérifier les bases
    local db_count
    db_count=$(ch_query "SELECT count() FROM system.databases WHERE name IN ('ja4_logs','ja4_processing')" | tr -d ' \n')
    [ "$db_count" = "2" ] \
        && pass "Bases ja4_logs + ja4_processing créées" \
        || fail "Bases manquantes (obtenu: $db_count)"

    # Vérifier que la DB est bien vide (nouvelle instance)
    local raw_init
    raw_init=$(ch_query "SELECT count() FROM ja4_logs.http_logs_raw" | tr -d ' \n')
    [ "${raw_init:-0}" -eq 0 ] 2>/dev/null \
        && pass "ClickHouse vierge avant test (http_logs_raw=0)" \
        || warn "ClickHouse non vide (http_logs_raw=${raw_init}) — données résiduelles"

    # Attendre bot-detector
    log "Attente bot-detector (max 120s)..."
    for i in $(seq 1 60); do
        if curl -sf "http://${ANALYSIS_IP}:8080/" >/dev/null 2>&1; then
            pass "bot-detector prêt (${i}*2s)"
            break
        fi
        sleep 2
    done

    # Attendre dashboard
    log "Attente dashboard (max 60s)..."
    for i in $(seq 1 30); do
        if curl -sf "http://${ANALYSIS_IP}:8000/health" >/dev/null 2>&1; then
            pass "dashboard prêt (${i}*2s)"
            break
        fi
        sleep 2
    done
}

# ═════════════════════════════════════════════════════════════════════════════
# Phase 2 : Démarrage des endpoints
# ═════════════════════════════════════════════════════════════════════════════
phase2_endpoints() {
    echo ""
    echo "╔══════════════════════════════════════════════════════════╗"
    echo "║  Phase 2 : Endpoints (nginx+apache+hitch + ja4ebpf)     ║"
    echo "╚══════════════════════════════════════════════════════════╝"
    echo ""

    cd "$VM_DIR"

    # Arrêter les services existants sur les endpoints (session précédente)
    log "Arrêt des services existants sur les endpoints..."
    for vm in $ENDPOINT_VMS; do
        vagrant ssh "$vm" -- "sudo nginx -s stop 2>/dev/null; sudo pkill ja4ebpf 2>/dev/null; sudo pkill httpd 2>/dev/null; sudo pkill hitch 2>/dev/null; sudo pkill varnishd 2>/dev/null; sudo pkill -f 'TCPServer.*8080' 2>/dev/null; sudo pkill -f 'TCPServer.*:80' 2>/dev/null" 2>/dev/null || true
    done
    sleep 2

    # Démarrer chaque endpoint en arrière-plan
    # ja4ebpf pointe vers le ClickHouse de la VM analysis (CH_HOST=192.168.42.10)
    ENDPOINT_PIDS=()
    for vm in $ENDPOINT_VMS; do
        log "Démarrage $vm (nginx + ja4ebpf → ${ANALYSIS_IP})..."
        vagrant ssh "$vm" -- "sudo rm -f /tmp/ja4ebpf-traffic-done" 2>/dev/null || true
        vagrant ssh "$vm" -- "sudo CH_HOST=${ANALYSIS_IP} bash /ja4-platform/tests/vm/run-tests-vm.sh ${STACK} start" &
        ENDPOINT_PIDS+=($!)
    done

    # Attendre que les services soient prêts
    log "Attente démarrage des endpoints (45s)..."
    sleep 45

    # Vérifier la connectivité de chaque endpoint (3 IPs × 2 ports)
    # Les services sont sur le réseau ja4-e2e (eth1, 192.168.42.0/24)
    for vm in $ENDPOINT_VMS; do
        local vm_ip1
        vm_ip1=$(vagrant ssh "$vm" -- 'ip -4 addr show eth1' 2>/dev/null \
            | awk '/inet / {sub(/\/.*/, "", $2); print $2; exit}')
        # Fallback eth0 si eth1 n'existe pas
        if [ -z "$vm_ip1" ]; then
            vm_ip1=$(vagrant ssh "$vm" -- 'ip -4 addr show eth0' 2>/dev/null \
                | awk '/inet / {sub(/\/.*/, "", $2); print $2; exit}')
        fi
        local np bl vm_ip2 vm_ip3
        np=$(echo "$vm_ip1" | awk -F. '{print $1"."$2"."$3}')
        bl=$(echo "$vm_ip1" | awk -F. '{print $4}')
        # Même logique que setup_all_ips : base+50 et base+51
        vm_ip2="${np}.$((bl + 50))"
        vm_ip3="${np}.$((bl + 51))"

        # nginx (IP1)
        if curl -sf --connect-timeout 3 "http://${vm_ip1}/health" >/dev/null 2>&1; then
            pass "$vm nginx:80 OK (${vm_ip1})"
        else
            warn "$vm nginx:80 injoignable (${vm_ip1})"
        fi
        if curl -sf -k --connect-timeout 3 "https://${vm_ip1}/health" >/dev/null 2>&1; then
            pass "$vm nginx:443 OK (${vm_ip1})"
        else
            warn "$vm nginx:443 injoignable (${vm_ip1})"
        fi
        # apache (IP2)
        if curl -sf --connect-timeout 3 "http://${vm_ip2}/health" >/dev/null 2>&1; then
            pass "$vm apache:80 OK (${vm_ip2})"
        else
            warn "$vm apache:80 injoignable (${vm_ip2})"
        fi
        if curl -sf -k --connect-timeout 3 "https://${vm_ip2}/health" >/dev/null 2>&1; then
            pass "$vm apache:443 OK (${vm_ip2})"
        else
            warn "$vm apache:443 injoignable (${vm_ip2})"
        fi
        # hitch+varnish (IP3)
        if curl -sf --connect-timeout 3 "http://${vm_ip3}/health" >/dev/null 2>&1; then
            pass "$vm hitch:80 OK (${vm_ip3})"
        else
            warn "$vm hitch:80 injoignable (${vm_ip3})"
        fi
        if curl -sf -k --connect-timeout 3 "https://${vm_ip3}/health" >/dev/null 2>&1; then
            pass "$vm hitch:443 OK (${vm_ip3})"
        else
            warn "$vm hitch:443 injoignable (${vm_ip3})"
        fi
    done
}

# ═════════════════════════════════════════════════════════════════════════════
# Phase 3 : Génération de trafic
# ═════════════════════════════════════════════════════════════════════════════
phase3_traffic() {
    echo ""
    echo "╔══════════════════════════════════════════════════════════╗"
    echo "║  Phase 3 : Génération de trafic → endpoints             ║"
    echo "╚══════════════════════════════════════════════════════════╝"
    echo ""

    local total_ok=0

    # Domaines pour SNI (cert CN=platform.test, nginx accepte tout via server_name _)
    local DOMAIN_SUFFIXES_ALL=("platform.test" "api.platform.test" "www.platform.test" "app.platform.test")
    local DOMAIN_SUFFIXES=("${DOMAIN_SUFFIXES_ALL[@]:0:${DNS_COUNT}}")

    # ── Collecter les IPs de tous les endpoints (3 IPs par VM sur réseau ja4-e2e/eth1) ──
    local -A VM_IPS_MAP
    for vm in $ENDPOINT_VMS; do
        local base_ip
        base_ip=$(vagrant ssh "$vm" -- 'ip -4 addr show eth1' 2>/dev/null \
            | awk '/inet / {sub(/\/.*/, "", $2); print $2; exit}')
        # Fallback eth0
        if [ -z "$base_ip" ]; then
            base_ip=$(vagrant ssh "$vm" -- 'ip -4 addr show eth0' 2>/dev/null \
                | awk '/inet / {sub(/\/.*/, "", $2); print $2; exit}')
        fi
        VM_IPS_MAP[$vm]="$base_ip"
        # Calculer IP2 et IP3 (identique à setup_all_ips : base+50 et base+51)
        local np bl
        np=$(echo "$base_ip" | awk -F. '{print $1"."$2"."$3}')
        bl=$(echo "$base_ip" | awk -F. '{print $4}')
        VM_IPS_MAP["${vm}_ip2"]="${np}.$((bl + 50))"
        VM_IPS_MAP["${vm}_ip3"]="${np}.$((bl + 51))"
    done

    # ── Ajouter des IPs alias sur la VM traffic pour diversifier les sources ──
    # Utiliser eth1 (réseau ja4-e2e, 192.168.42.x) car les cibles sont sur ce réseau.
    local TRAFFIC_IP
    TRAFFIC_IP=$(vagrant ssh "$TRAFFIC_VM" -- 'ip -4 addr show eth1' 2>/dev/null \
        | awk '/inet / {sub(/\/.*/, "", $2); print $2; exit}')
    # Fallback eth0 si eth1 n'existe pas
    if [ -z "$TRAFFIC_IP" ]; then
        TRAFFIC_IP=$(vagrant ssh "$TRAFFIC_VM" -- 'ip -4 addr show eth0' 2>/dev/null \
            | awk '/inet / {sub(/\/.*/, "", $2); print $2; exit}')
    fi
    local TRAFFIC_IFACE
    if [ -n "$TRAFFIC_IP" ]; then
        case "$TRAFFIC_IP" in
            192.168.42.*) TRAFFIC_IFACE="eth1" ;;
            *) TRAFFIC_IFACE="eth0" ;;
        esac
    fi

    if [ "$SRC_IP_COUNT" -gt 1 ] && [ -n "$TRAFFIC_IP" ]; then
        log "Ajout de ${SRC_IP_COUNT} IPs sources sur ${TRAFFIC_VM} (${TRAFFIC_IFACE})..."
        local net_prefix
        net_prefix=$(echo "$TRAFFIC_IP" | awk -F. '{print $1"."$2"."$3}')
        local base_last
        base_last=$(echo "$TRAFFIC_IP" | awk -F. '{print $4}')
        for i in $(seq 1 $((SRC_IP_COUNT - 1))); do
            local alias_last=$((base_last + 100 + i))
            [ "$alias_last" -gt 254 ] && alias_last=$((10 + i))
            local alias_ip="${net_prefix}.${alias_last}"
            vagrant ssh "$TRAFFIC_VM" -- "sudo ip addr add ${alias_ip}/24 dev ${TRAFFIC_IFACE} 2>/dev/null || true" 2>/dev/null || true
        done
    fi

    # ── Construire les listes d'IPs cibles et SNI appariées ──
    # Chaque entrée (VM, service, domaine) a un SNI unique : <vm>-<service>-<domaine>
    # TARGET_IPS et SNI_HOSTS sont des listes parallèles de même longueur.
    local TARGET_IPS=""
    local SNI_HOSTS_STR=""
    # Noms de services par clé IP
    local SVC_NAMES_ip1="nginx" SVC_NAMES_ip2="apache" SVC_NAMES_ip3="hitch"
    for target_vm in $ENDPOINT_VMS; do
        local vm_base="${VM_IPS_MAP[$target_vm]}"
        local vm_ip2="${VM_IPS_MAP[${target_vm}_ip2]}"
        local vm_ip3="${VM_IPS_MAP[${target_vm}_ip3]}"
        for ip_key in ip1 ip2 ip3; do
            local svc_name_var="SVC_NAMES_${ip_key}"
            local svc_name="${!svc_name_var}"
            local map_key="${target_vm}_${ip_key}"
            local target_ip="${VM_IPS_MAP[$map_key]:-$vm_base}"
            for domain in "${DOMAIN_SUFFIXES[@]}"; do
                local sni="${target_vm}-${svc_name}-${domain}"
                TARGET_IPS="$TARGET_IPS $target_ip"
                SNI_HOSTS_STR="$SNI_HOSTS_STR $sni"
            done
        done
    done
    TARGET_IPS=$(echo $TARGET_IPS)
    SNI_HOSTS_STR=$(echo $SNI_HOSTS_STR)
    log "Cibles : $(echo "$TARGET_IPS" | wc -w) paires IP/SNI ($(echo "$SNI_HOSTS_STR" | tr ' ' '\n' | sort -u | wc -l) uniques)"

    # ── Synchroniser generate-traffic.sh vers la VM traffic ──
    log "Synchronisation du script de trafic vers ${TRAFFIC_VM}..."
    vagrant rsync "$TRAFFIC_VM" 2>&1 | tail -1

    # ── Écrire le config et lancer le trafic depuis la VM traffic ──
    log "Génération depuis ${TRAFFIC_VM} : ${HTTPS_COUNT} HTTPS + ${HTTP_COUNT} HTTP..."

    vagrant ssh "$TRAFFIC_VM" -- "cat > /tmp/e2e-traffic.env << 'ENVEOF'
export HITS=${HTTPS_COUNT}
export HITS_HTTP=${HTTP_COUNT}
export TARGET_IPS='${TARGET_IPS}'
export SNI_HOSTS='${SNI_HOSTS_STR}'
export TLS_FLAGS='${CURL_TLS_FLAGS}'
export SRC_IP_COUNT=${SRC_IP_COUNT}
ENVEOF"

    # Lancer le générateur de trafic (curl-impersonate + curl)
    vagrant ssh "$TRAFFIC_VM" -- \
        "source /tmp/e2e-traffic.env && bash /ja4-platform/tests/vm/generate-traffic.sh" \
        > /tmp/e2e-traffic.out 2>&1 &

    # Attendre que le processus se termine (max 300s)
    for i in $(seq 1 300); do
        if [ -f /tmp/e2e-traffic.out ] && ! pgrep -f "vagrant ssh $TRAFFIC_VM.*generate-traffic" >/dev/null 2>&1; then
            break
        fi
        sleep 1
    done
    local result
    result=$(tail -1 /tmp/e2e-traffic.out 2>/dev/null || echo "0/${HTTPS_COUNT} 0/${HTTP_COUNT}")

    # Format: "ok_https/hits_https ok_http/hits_http"
    local https_result http_result
    https_result=$(echo "$result" | awk '{print $1}')
    http_result=$(echo "$result" | awk '{print $2}')
    local ok_https ok_http
    ok_https=$(echo "$https_result" | cut -d/ -f1)
    ok_http=$(echo "${http_result:-0/0}" | cut -d/ -f1)

    log "  ${TRAFFIC_VM} HTTPS : ${https_result}  HTTP : ${http_result:-0/0}"
    total_ok=$((total_ok + ok_https + ok_http))

    # ── HTTP/2 massif depuis la VM traffic si httpx est disponible ──
    if vagrant ssh "$TRAFFIC_VM" -- 'python3 -c "import httpx"' 2>/dev/null; then
        log "Génération HTTP/2 depuis ${TRAFFIC_VM} (${HTTPS_COUNT} requêtes, TLS=${TLS_VERSIONS}, DNS=${DNS_COUNT})..."

        # Écrire le script httpx sur la VM traffic
        vagrant ssh "$TRAFFIC_VM" -- "cat > /tmp/e2e-h2-traffic.py << 'PYEOF'
import httpx, ssl as _ssl, warnings, random, os
warnings.filterwarnings('ignore')
paths = ['/', '/health', '/data', '/api/users', '/api/v1/status', '/login', '/api/search']
sni_hosts = os.environ.get('SNI_HOSTS', 'platform.test').split()
target_ips = os.environ.get('TARGET_IPS', '127.0.0.1').split()
# Les listes sont appariées : même index = même cible (vm-service-domaine)
assert len(sni_hosts) == len(target_ips), f'SNI_HOSTS ({len(sni_hosts)}) != TARGET_IPS ({len(target_ips)})'
targets = list(zip(target_ips, sni_hosts))
tls_versions = [v.strip() for v in os.environ.get('TLS_VERSIONS', '1.2,1.3').split(',')]
uas_browser = [
    'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 Chrome/131.0.0.0 Safari/537.36',
    'Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/605.1.15 Safari/605.1.15',
    'Mozilla/5.0 (X11; Linux x86_64; rv:133.0) Gecko/20100101 Firefox/133.0',
]
uas_bot = ['python-httpx/0.28.1', 'Googlebot/2.1', 'Go-http-client/2.0']
supported = {'1.2': _ssl.TLSVersion.TLSv1_2, '1.3': _ssl.TLSVersion.TLSv1_3}
tls_map = [supported[v] for v in tls_versions if v in supported]
ctx = _ssl.SSLContext(_ssl.PROTOCOL_TLS_CLIENT)
ctx.check_hostname = False
ctx.verify_mode = _ssl.CERT_NONE
if tls_map:
    ctx.minimum_version = min(tls_map)
    ctx.maximum_version = max(tls_map)
hits = int(os.environ.get('HITS', '100'))
with httpx.Client(http2=True, verify=ctx) as c:
    for i in range(hits):
        p = random.choice(paths)
        target_ip, sni = random.choice(targets)
        ua = random.choice(uas_browser if random.random() < 0.7 else uas_bot)
        try:
            c.get(f'https://{target_ip}' + p, headers={'User-Agent': ua, 'Host': sni})
        except:
            pass
PYEOF"

        vagrant ssh "$TRAFFIC_VM" -- \
            "source /tmp/e2e-traffic.env && TLS_VERSIONS='${TLS_VERSIONS}' python3 /tmp/e2e-h2-traffic.py" \
            2>/dev/null || true
        pass "HTTP/2 généré depuis ${TRAFFIC_VM}"
    else
        warn "httpx non disponible sur ${TRAFFIC_VM} — HTTP/2 ignoré"
    fi

    pass "Trafic total : ${total_ok} requêtes réussies"
}

# ═════════════════════════════════════════════════════════════════════════════
# Phase 4 : Attente du pipeline
# ═════════════════════════════════════════════════════════════════════════════
phase4_wait() {
    echo ""
    echo "╔══════════════════════════════════════════════════════════╗"
    echo "║  Phase 4 : Attente du pipeline                           ║"
    echo "╚══════════════════════════════════════════════════════════╝"
    echo ""

    # Flush ja4ebpf
    log "Attente flush ja4ebpf (15s)..."
    sleep 15

    # Signaler aux endpoints que le trafic est terminé
    cd "$VM_DIR"
    for vm in $ENDPOINT_VMS; do
        vagrant ssh "$vm" -- 'sudo touch /tmp/ja4ebpf-traffic-done' 2>/dev/null || true
    done

    # Attendre que les données brutes arrivent dans ClickHouse
    log "Attente données dans ClickHouse (max 60s)..."
    for i in $(seq 1 30); do
        local raw_count
        raw_count=$(ch_query "SELECT count() FROM ja4_logs.http_logs_raw" 2>/dev/null | tr -d ' \n')
        if [ "${raw_count:-0}" -gt 0 ] 2>/dev/null; then
            pass "Données brutes reçues : ${raw_count} lignes (${i}*2s)"
            break
        fi
        sleep 2
    done

    # Attendre au moins un cycle bot-detector
    log "Attente cycle bot-detector (max 120s)..."
    for i in $(seq 1 60); do
        local scores
        scores=$(ch_query "SELECT count() FROM ja4_processing.ml_all_scores" 2>/dev/null | tr -d ' \n')
        if [ "${scores:-0}" -gt 0 ] 2>/dev/null; then
            pass "bot-detector a complété au moins 1 cycle ($scores scores)"
            break
        fi
        sleep 2
    done

    # Attendre les processus endpoint en arrière-plan
    log "Attente fin des processus endpoint..."
    for pid in "${ENDPOINT_PIDS[@]:-}"; do
        wait "$pid" 2>/dev/null || true
    done
}

# ═════════════════════════════════════════════════════════════════════════════
# Phase 5 : Vérification complète
# ═════════════════════════════════════════════════════════════════════════════
phase5_verify() {
    echo ""
    echo "╔══════════════════════════════════════════════════════════╗"
    echo "║  Phase 5 : Vérification du pipeline complet              ║"
    echo "╚══════════════════════════════════════════════════════════╝"
    echo ""

    # ── Layer 1 : Données brutes ─────────────────────────────────────────────
    log "── Layer 1 : Données brutes ──"

    local raw_count
    raw_count=$(ch_query "SELECT count() FROM ja4_logs.http_logs_raw" | tr -d ' \n')
    if [ "${raw_count:-0}" -gt 0 ] 2>/dev/null; then
        pass "http_logs_raw : ${raw_count} lignes"
    else
        fail "http_logs_raw vide"
    fi

    # Multi-source : données de plusieurs endpoints
    local host_count
    host_count=$(ch_query "SELECT uniqExact(host) FROM ja4_logs.http_logs" | tr -d ' \n')
    if [ "${host_count:-0}" -ge 2 ] 2>/dev/null; then
        pass "Multi-source : ${host_count} hôtes distincts"
    else
        warn "Multi-source : ${host_count:-0} hôte(s) distinct(s)"
    fi

    # ── Layer 2 : Pipeline ClickHouse ────────────────────────────────────────
    log "── Layer 2 : Pipeline ClickHouse (MVs) ──"

    local ja4_count
    ja4_count=$(ch_query "SELECT count() FROM ja4_logs.http_logs WHERE ja4 != ''" | tr -d ' \n')
    [ "${ja4_count:-0}" -gt 0 ] 2>/dev/null \
        && pass "JA4 fingerprints : ${ja4_count}" \
        || fail "JA4 fingerprints absents"

    local method_count
    method_count=$(ch_query "SELECT count() FROM ja4_logs.http_logs WHERE method != ''" | tr -d ' \n')
    [ "${method_count:-0}" -gt 0 ] 2>/dev/null \
        && pass "L7 HTTP : ${method_count} requêtes capturées" \
        || fail "L7 HTTP absent"

    local agg_count
    agg_count=$(ch_query "SELECT count() FROM ja4_processing.agg_host_ip_ja4_1h" | tr -d ' \n')
    [ "${agg_count:-0}" -gt 0 ] 2>/dev/null \
        && pass "Agrégation agg_host_ip_ja4_1h : ${agg_count} entrées" \
        || warn "Agrégation agg_host_ip_ja4_1h vide (volume insuffisant pour la fenêtre horaire)"

    local feat_count
    feat_count=$(ch_query "SELECT count() FROM ja4_processing.view_ai_features_1h" | tr -d ' \n')
    [ "${feat_count:-0}" -gt 0 ] 2>/dev/null \
        && pass "AI features : ${feat_count} lignes" \
        || warn "AI features vides (agrégation horaire pas encore calculée)"

    # ── Layer 3 : ML bot-detector ────────────────────────────────────────────
    log "── Layer 3 : ML bot-detector ──"

    local scores_count
    scores_count=$(ch_query "SELECT count() FROM ja4_processing.ml_all_scores" | tr -d ' \n')
    [ "${scores_count:-0}" -gt 0 ] 2>/dev/null \
        && pass "ml_all_scores : ${scores_count} classifications" \
        || fail "ml_all_scores vide — bot-detector n'a pas produit de résultats"

    local anomaly_count
    anomaly_count=$(ch_query "SELECT count() FROM ja4_processing.ml_detected_anomalies" | tr -d ' \n')
    if [ "${anomaly_count:-0}" -gt 0 ] 2>/dev/null; then
        pass "ml_detected_anomalies : ${anomaly_count} anomalies détectées"
    else
        warn "ml_detected_anomalies vide (pas d'anomalies dans le trafic de test)"
    fi

    # bot-detector health
    if curl -sf "http://${ANALYSIS_IP}:8080/" >/dev/null 2>&1; then
        pass "bot-detector health OK"
    else
        fail "bot-detector health KO"
    fi

    # ── Layer 4 : Dashboard ──────────────────────────────────────────────────
    log "── Layer 4 : Dashboard ──"

    if curl -sf "http://${ANALYSIS_IP}:8000/health" >/dev/null 2>&1; then
        pass "dashboard /health OK"
    else
        fail "dashboard /health KO"
    fi

    local overview
    overview=$(curl -sf "http://${ANALYSIS_IP}:8000/api/overview" 2>/dev/null || echo "")
    if [ -n "$overview" ] && [ "$overview" != "{}" ] && [ "$overview" != "null" ]; then
        pass "dashboard /api/overview retourne des données"
    else
        warn "dashboard /api/overview vide"
    fi

    local detections
    detections=$(curl -sf "http://${ANALYSIS_IP}:8000/api/detections" 2>/dev/null || echo "")
    if [ -n "$detections" ]; then
        pass "dashboard /api/detections accessible"
    else
        warn "dashboard /api/detections pas de réponse"
    fi
}

# ═════════════════════════════════════════════════════════════════════════════
# Résumé
# ═════════════════════════════════════════════════════════════════════════════
phase_summary() {
    local total=$((PASS_COUNT + FAIL_COUNT + WARN_COUNT))
    echo ""
    echo "╔══════════════════════════════════════════════════════════╗"
    echo "║  RÉSULTATS E2E                                           ║"
    echo "╚══════════════════════════════════════════════════════════╝"
    echo -e "  ${GREEN}Réussis : $PASS_COUNT${RESET} / $total"
    [ "$WARN_COUNT" -gt 0 ] && echo -e "  ${YELLOW}Avertissements : $WARN_COUNT${RESET} / $total"
    [ "$FAIL_COUNT" -gt 0 ] && echo -e "  ${RED}Échoués : $FAIL_COUNT${RESET} / $total"
    echo ""

    if [ "$FAIL_COUNT" -eq 0 ]; then
        echo -e "  ${GREEN}${BOLD}Test E2E distribué : SUCCÈS${RESET}"
    else
        echo -e "  ${RED}${BOLD}Test E2E distribué : $FAIL_COUNT ÉCHECS${RESET}"
    fi
    echo ""
}

# ═════════════════════════════════════════════════════════════════════════════
# Nettoyage
# ═════════════════════════════════════════════════════════════════════════════
cleanup() {
    # Supprimer les IPs alias sur la VM traffic (sur eth1, réseau ja4-e2e)
    if [ "${SRC_IP_COUNT:-1}" -gt 1 ]; then
        log "Suppression des IPs alias sur ${TRAFFIC_VM}..."
        vagrant ssh "$TRAFFIC_VM" -- "
            base_ip=\$(ip -4 addr show eth1 2>/dev/null | awk '/inet / {sub(/\/.*/, \"\", \\\$2); print \\\$2; exit}')
            if [ -z \"\$base_ip\" ]; then
                base_ip=\$(ip -4 addr show eth0 | awk '/inet / {sub(/\/.*/, \"\", \\\$2); print \\\$2; exit}')
                iface=eth0
            else
                iface=eth1
            fi
            net_prefix=\$(echo \$base_ip | awk -F. '{print \\\$1\".\"\\\$2\".\"\\\$3}')
            base_last=\$(echo \$base_ip | awk -F. '{print \\\$4}')
            for i in \$(seq 1 $((SRC_IP_COUNT - 1))); do
                alias_last=\$((base_last + 100 + i))
                [ \$alias_last -gt 254 ] && alias_last=\$((10 + i))
                sudo ip addr del \${net_prefix}.\${alias_last}/24 dev \$iface 2>/dev/null || true
            done
        " 2>/dev/null || true
    fi

    # Toujours arrêter les endpoints (sauf --keep global)
    if [ "${KEEP_RUNNING:-false}" != "true" ]; then
        log "Nettoyage des endpoints..."
        cd "$VM_DIR"
        for vm in $ENDPOINT_VMS; do
            vagrant ssh "$vm" -- "sudo pkill ja4ebpf 2>/dev/null; sudo nginx -s stop 2>/dev/null; sudo pkill httpd 2>/dev/null; sudo pkill hitch 2>/dev/null; sudo pkill varnishd 2>/dev/null; sudo pkill -f 'TCPServer.*8080' 2>/dev/null; sudo pkill -f 'TCPServer.*:80' 2>/dev/null" 2>/dev/null || true
            # Supprimer les IPs alias (IP2, IP3) sur eth1
            local vm_base_ip
            vm_base_ip=$(vagrant ssh "$vm" -- 'ip -4 addr show eth1' 2>/dev/null \
                | awk '/inet / {sub(/\/.*/, "", $2); print $2; exit}')
            if [ -n "$vm_base_ip" ]; then
                local np bl ip2 ip3
                np=$(echo "$vm_base_ip" | awk -F. '{print $1"."$2"."$3}')
                bl=$(echo "$vm_base_ip" | awk -F. '{print $4}')
                ip2="${np}.$((bl + 50))"
                ip3="${np}.$((bl + 51))"
                vagrant ssh "$vm" -- "sudo ip addr del ${ip2}/24 dev eth1 2>/dev/null || true; sudo ip addr del ${ip3}/24 dev eth1 2>/dev/null || true" 2>/dev/null || true
            fi
        done
    fi

    # Stack analysis : conserver si --keep-analysis ou --keep, sinon arrêter
    if [ "${KEEP_RUNNING:-false}" = "true" ] || [ "$KEEP_ANALYSIS" = "true" ]; then
        log "Stack analysis conservée :"
        log "  ClickHouse : http://${ANALYSIS_IP}:8123/play"
        log "  Dashboard  : http://${ANALYSIS_IP}:8000"
        log "  Bot-detector : http://${ANALYSIS_IP}:8080"
    else
        log "Arrêt de la stack analysis..."
        vagrant ssh analysis -- "cd /ja4-platform/tests/vm/analysis && docker compose down -v" 2>/dev/null || true
    fi

    if [ "${KEEP_RUNNING:-false}" = "true" ]; then
        log "VMs endpoints conservées :"
        for vm in $ENDPOINT_VMS; do
            local vm_ip
            vm_ip=$(vagrant ssh "$vm" -- 'ip -4 addr show eth1' 2>/dev/null \
                | awk '/inet / {sub(/\/.*/, "", $2); print $2; exit}')
            if [ -z "$vm_ip" ]; then
                vm_ip=$(vagrant ssh "$vm" -- 'ip -4 addr show eth0' 2>/dev/null \
                    | awk '/inet / {sub(/\/.*/, "", $2); print $2; exit}')
            fi
            local np bl
            np=$(echo "$vm_ip" | awk -F. '{print $1"."$2"."$3}')
            bl=$(echo "$vm_ip" | awk -F. '{print $4}')
            log "  $vm : ${vm_ip} (nginx)  ${np}.$((bl + 50)) (apache)  ${np}.$((bl + 51)) (hitch+varnish)"
        done
    fi
}

# ═════════════════════════════════════════════════════════════════════════════
# Main
# ═════════════════════════════════════════════════════════════════════════════

KEEP_RUNNING="${KEEP_RUNNING:-false}"

trap cleanup EXIT

phase0_setup
phase1_analysis
phase2_endpoints

if [ "$UP_ONLY" = "true" ]; then
    echo ""
    echo "╔══════════════════════════════════════════════════════════╗"
    echo "║  Stack prête — mode --up (pas de trafic ni vérification) ║"
    echo "╚══════════════════════════════════════════════════════════╝"
    echo ""
    log "ClickHouse  : http://${ANALYSIS_IP}:8123/play"
    log "Dashboard   : http://${ANALYSIS_IP}:8000"
    log "Bot-detector : http://${ANALYSIS_IP}:8080"
    for vm in $ENDPOINT_VMS; do
        local vm_ip
        vm_ip=$(vagrant ssh "$vm" -- 'ip -4 addr show eth1' 2>/dev/null \
            | awk '/inet / {sub(/\/.*/, "", $2); print $2; exit}')
        if [ -z "$vm_ip" ]; then
            vm_ip=$(vagrant ssh "$vm" -- 'ip -4 addr show eth0' 2>/dev/null \
                | awk '/inet / {sub(/\/.*/, "", $2); print $2; exit}')
        fi
        local np bl vm_ip2 vm_ip3
        np=$(echo "$vm_ip" | awk -F. '{print $1"."$2"."$3}')
        bl=$(echo "$vm_ip" | awk -F. '{print $4}')
        vm_ip2="${np}.$((bl + 50))"
        vm_ip3="${np}.$((bl + 51))"
        log "$vm nginx         : http://${vm_ip}:80   https://${vm_ip}:443"
        log "$vm apache        : http://${vm_ip2}:80  https://${vm_ip2}:443"
        log "$vm hitch+varnish : http://${vm_ip3}:80  https://${vm_ip3}:443"
    done
    log ""
    log "Pour générer du trafic :"
    log "  ./run-e2e-test.sh --hits 200 --dns 3 --tls 1.2,1.3 --keep-analysis"
    log "  curl -sk https://<endpoint_ip>/health"
    KEEP_ANALYSIS=true
    exit 0
fi

phase3_traffic
phase4_wait
phase5_verify
phase_summary

[ "$FAIL_COUNT" -eq 0 ] && exit 0 || exit 1