fix : t 021 a 033 fait

RecetteScripts/backup-bdd-recette.sh

@@ -162,17 +162,22 @@ TS="$(date +'%Y-%m-%d_%H-%M-%S')"
 BACKUP_DIR_NAME="backup_${TS}"
 LOG_FILE="${LOG_DIR}/${BACKUP_DIR_NAME}.log"
 
+exec > >(tee -a "$LOG_FILE") 2>&1
+
 TMP_DIR="$(mktemp -d /tmp/pg_dump_XXXXXX)" || {
   echo "ERROR: impossible de créer le dossier temporaire" >&2
   exit 1
 }
 
-exec > >(tee -a "$LOG_FILE") 2>&1
+log() { echo "[$(date '+%Y-%m-%d %H:%M:%S')] $*"; }
 
-log() { echo "---- $(date +'%Y-%m-%d %H:%M:%S') ---- $*"; }
+fail() {
+  log "ERROR: $*"
+  exit 1
+}
 
 require_cmd() {
-  command -v "$1" >/dev/null 2>&1
+  command -v "$1" >/dev/null 2>&1 || fail "commande manquante : $1"
 }
 
 safe_remove_dir() {
@@ -192,10 +197,7 @@ export PGPASSWORD
 #######################################
 
 for cmd in ssh scp curl jq pg_dump pg_dumpall mktemp; do
-  require_cmd "$cmd" || {
-    echo "ERROR: commande manquante : $cmd" >&2
-    exit 1
-  }
+  require_cmd "$cmd"
 done
 
 [[ -f "$SSH_KEY" ]] || {
@@ -222,14 +224,14 @@ chmod 600 "$SSH_KEY" || true
 DISCORD_WEBHOOK_URL="${DISCORD_WEBHOOK_URL:-}"
 DISCORD_PING="${DISCORD_PING:-@here}"
 
-discord_send() {
+send_discord() {
   local msg="$1"
+  local payload
   [[ -z "${DISCORD_WEBHOOK_URL:-}" ]] && return 0
 
-  local payload
   payload="$(jq -n --arg content "$msg" '{content: $content}')" || {
     log "ERROR: impossible de construire le payload JSON Discord"
-    return 1
+    return 0
   }
 
   curl -fsS \
@@ -251,7 +253,7 @@ Dumps transfer: ✅
 Users transfer: ✅
 EOF
 )"
-  discord_send "$msg"
+  send_discord "$msg"
 }
 
 #######################################
@@ -265,7 +267,7 @@ discord_msg_users_ok_simple() {
 Users backup validé
 EOF
 )"
-  discord_send "$msg"
+  send_discord "$msg"
 }
 
 discord_msg_users_error() {
@@ -297,7 +299,7 @@ EOF
 )"
   fi
 
-  discord_send "$msg"
+  send_discord "$msg"
 }
 
 #######################################
@@ -312,7 +314,7 @@ discord_msg_db_ok_simple() {
 Backup validé : ${db}
 EOF
 )"
-  discord_send "$msg"
+  send_discord "$msg"
 }
 
 discord_msg_db_error() {
@@ -347,7 +349,7 @@ EOF
 )"
   fi
 
-  discord_send "$msg"
+  send_discord "$msg"
 }
 
 #######################################
@@ -370,11 +372,36 @@ declare -A DB_DETAILS
 #######################################
 
 LOCK_DIR="/tmp/pg_multi_dump_stream.lock.d"
+LOCK_PID_FILE="${LOCK_DIR}/pid"
 
 if ! mkdir "$LOCK_DIR" 2>/dev/null; then
-  log "ERROR: Backup déjà en cours"
-  discord_msg_users_error "" "" "Lock already exists"
-  exit 1
+  stale_lock="no"
+  existing_pid=""
+
+  if [[ -f "$LOCK_PID_FILE" ]]; then
+    existing_pid="$(<"$LOCK_PID_FILE")"
+  fi
+
+  if [[ "$existing_pid" =~ ^[0-9]+$ ]] && kill -0 "$existing_pid" 2>/dev/null; then
+    log "ERROR: Backup déjà en cours (PID ${existing_pid})"
+    discord_msg_users_error "" "" "Lock already exists (PID ${existing_pid})"
+    exit 1
+  fi
+
+  stale_lock="yes"
+  log "WARNING: lock périmé détecté, nettoyage en cours"
+  rm -rf -- "$LOCK_DIR"
+
+  mkdir "$LOCK_DIR" 2>/dev/null || fail "impossible de recréer le lock après nettoyage"
+fi
+
+echo $$ > "$LOCK_PID_FILE" || {
+  rm -rf -- "$LOCK_DIR"
+  fail "impossible d'écrire le PID du lock"
+}
+
+if [[ "${stale_lock:-no}" == "yes" ]]; then
+  log "Lock périmé nettoyé."
 fi
 
 cleanup() {
@@ -406,18 +433,18 @@ fi
 
 ROLES_FILE="${TMP_DIR}/user_${TS}.sql"
 
-set +e
-
 log "Export des rôles PostgreSQL"
 
-pg_dumpall \
+if pg_dumpall \
   -h "$PGHOST" \
   -p "$PGPORT" \
   -U "$PGUSER" \
   --globals-only \
-  > "$ROLES_FILE"
-
-RET=$?
+  > "$ROLES_FILE"; then
+  RET=0
+else
+  RET=$?
+fi
 
 if [[ $RET -ne 0 ]]; then
   USERS_OK=
@@ -428,8 +455,11 @@ else
 fi
 
 if [[ -n "${USERS_EXPORT_OK:-}" ]]; then
-  scp "${SCP_OPTS[@]}" "$ROLES_FILE" "$IA_SSH:${BACKUP_REMOTE_DIR}/user/"
-  RET=$?
+  if scp "${SCP_OPTS[@]}" "$ROLES_FILE" "$IA_SSH:${BACKUP_REMOTE_DIR}/user/"; then
+    RET=0
+  else
+    RET=$?
+  fi
 
   if [[ $RET -ne 0 ]]; then
     USERS_OK=
@@ -444,14 +474,10 @@ if [[ -n "${USERS_EXPORT_OK:-}" ]]; then
   fi
 fi
 
-set -e
-
 #######################################
 # Dump des bases
 #######################################
 
-set +e
-
 for DB in "${DBS_ARRAY[@]}"; do
   FILE="${TMP_DIR}/${DB}_${TS}.dump"
 
@@ -461,8 +487,11 @@ for DB in "${DBS_ARRAY[@]}"; do
 
   log "Dump $DB"
 
-  pg_dump -h "$PGHOST" -p "$PGPORT" -U "$PGUSER" -Fc -d "$DB" -f "$FILE"
-  RET=$?
+  if pg_dump -h "$PGHOST" -p "$PGPORT" -U "$PGUSER" -Fc -d "$DB" -f "$FILE"; then
+    RET=0
+  else
+    RET=$?
+  fi
 
   if [[ $RET -ne 0 ]]; then
     DUMPS_OK=
@@ -472,8 +501,11 @@ for DB in "${DBS_ARRAY[@]}"; do
     continue
   fi
 
-  scp "${SCP_OPTS[@]}" "$FILE" "$IA_SSH:${BACKUP_REMOTE_DIR}/${DB}/"
-  RET=$?
+  if scp "${SCP_OPTS[@]}" "$FILE" "$IA_SSH:${BACKUP_REMOTE_DIR}/${DB}/"; then
+    RET=0
+  else
+    RET=$?
+  fi
 
   if [[ $RET -ne 0 ]]; then
     DUMPS_OK=
@@ -482,18 +514,17 @@ for DB in "${DBS_ARRAY[@]}"; do
   fi
 done
 
-set -e
-
 #######################################
 # Rotation distante
 #######################################
 
 log "Starting remote rotation: delete backups older than ${RETENTION_DAYS} days"
 
-set +e
-
-ssh "${SSH_OPTS[@]}" "$IA_SSH" "find '${BACKUP_REMOTE_DIR}/user' -type f -name 'user_*.sql' -mtime +${RETENTION_DAYS} -delete"
-RET=$?
+if ssh "${SSH_OPTS[@]}" "$IA_SSH" "find '${BACKUP_REMOTE_DIR}/user' -type f -name 'user_*.sql' -mtime +${RETENTION_DAYS} -delete"; then
+  RET=0
+else
+  RET=$?
+fi
 
 if [[ $RET -ne 0 ]]; then
   log "ERROR: remote rotation failed for users"
@@ -502,8 +533,11 @@ else
 fi
 
 for DB in "${DBS_ARRAY[@]}"; do
-  ssh "${SSH_OPTS[@]}" "$IA_SSH" "find '${BACKUP_REMOTE_DIR}/${DB}' -type f -name '${DB}_*.dump' -mtime +${RETENTION_DAYS} -delete"
-  RET=$?
+  if ssh "${SSH_OPTS[@]}" "$IA_SSH" "find '${BACKUP_REMOTE_DIR}/${DB}' -type f -name '${DB}_*.dump' -mtime +${RETENTION_DAYS} -delete"; then
+    RET=0
+  else
+    RET=$?
+  fi
 
   if [[ $RET -ne 0 ]]; then
     log "ERROR: remote rotation failed for ${DB}"
@@ -512,8 +546,6 @@ for DB in "${DBS_ARRAY[@]}"; do
   fi
 done
 
-set -e
-
 log "Remote rotation finished"
 
 #######################################