From 01110cd322b5de0c4a4b8611cca2a5a57bb48179 Mon Sep 17 00:00:00 2001
From: Nicolas Cantu <nicolas.cantu@pm.me>
Date: Tue, 6 Jan 2026 14:36:41 +0100
Subject: [PATCH] Fix root cause of ControlSocket already exists errors
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

**Motivations:**
- SSH errors 'ControlSocket already exists, disabling multiplexing' were occurring
- Socket could become invalid between check and execution
- ControlMaster=auto detects invalid socket but doesn't remove it, just disables multiplexing
- Dead socket remains and causes issues for subsequent commands

**Root causes:**
- ControlMaster socket can exist but connection can be dead
- Socket can become invalid between check_ssh_connection() and ssh_exec() execution
- ControlMaster=auto detects invalid socket but doesn't remove it, leaving dead socket
- Dead socket causes 'ControlSocket already exists' errors on subsequent commands
- Previous cleanup was not aggressive enough to remove dead sockets

**Correctifs:**
- Improved check_ssh_connection() to test actual connection with 'true' command instead of just 'ssh -O check'
- Enhanced cleanup_dead_ssh() to remove entire directory instead of just socket file
- This ensures socket is truly removed even if process is still holding it
- Removed complex multi-step cleanup, using simple directory removal and recreation
- Socket validation now tests actual connection usability, not just socket existence

**Evolutions:**
- More robust socket cleanup that guarantees removal of dead sockets
- Better detection of invalid sockets before command execution

**Pages affectées:**
- deploy.sh: Improved cleanup_dead_ssh() and check_ssh_connection() functions
---
 deploy.sh | 23 +++++++++++++++++------
 1 file changed, 17 insertions(+), 6 deletions(-)

diff --git a/deploy.sh b/deploy.sh
index 44eff2a..944d9a4 100644
--- a/deploy.sh
+++ b/deploy.sh
@@ -22,20 +22,29 @@ cleanup_dead_ssh() {
         # Attendre un peu pour que la fermeture se termine
         sleep 0.5
     fi
-    # Forcer la suppression du socket même s'il existe encore
-    rm -f "${SSH_CONTROL_PATH}" 2>/dev/null || true
-    # Nettoyer aussi le répertoire parent s'il est vide
-    rmdir "${SSH_CONTROL_DIR}" 2>/dev/null || true
+    # Forcer la suppression du socket et du répertoire parent
+    # Supprimer le répertoire entier garantit que le socket est vraiment supprimé
+    rm -rf "${SSH_CONTROL_DIR}" 2>/dev/null || true
+    # Recréer le répertoire pour les prochaines connexions
+    mkdir -p "${SSH_CONTROL_DIR}" 2>/dev/null || true
 }
 
 # Fonction pour vérifier si la connexion SSH maître est valide
 check_ssh_connection() {
-    ssh -O check -o ControlPath="${SSH_CONTROL_PATH}" ${SERVER} 2>/dev/null || return 1
+    # Vérifier si le socket existe
+    if [ ! -S "${SSH_CONTROL_PATH}" ]; then
+        return 1
+    fi
+    # Tester la connexion en essayant de l'utiliser avec une commande simple
+    # Cela détecte si le socket existe mais la connexion est morte
+    ssh -o ControlPath="${SSH_CONTROL_PATH}" ${SERVER} "true" 2>/dev/null || return 1
+    return 0
 }
 
 # Fonction pour exécuter une commande SSH avec connexion persistante
 ssh_exec() {
-    # Nettoyer le socket s'il existe mais est invalide
+    # Toujours vérifier et nettoyer le socket avant chaque commande
+    # pour éviter les sockets morts qui causent "ControlSocket already exists, disabling multiplexing"
     if [ -S "${SSH_CONTROL_PATH}" ]; then
         if ! check_ssh_connection; then
             # Connexion morte, nettoyer avant d'exécuter
@@ -44,6 +53,8 @@ ssh_exec() {
     fi
 
     # Exécuter la commande SSH (une seule tentative, pas de retry)
+    # ControlMaster=auto va créer une nouvelle connexion si le socket n'existe pas,
+    # ou réutiliser s'il est valide
     ssh -o ControlMaster=auto \
         -o ControlPath="${SSH_CONTROL_PATH}" \
         -o ControlPersist=300 \