Retry Telebit client connections

telebitctl

@@ -7,6 +7,8 @@ CLIENT_ENV="${TELEBIT_CLIENT_ENV:-$ROOT_DIR/telebit-client.env}"
 CLIENT_ENV_DIR="${TELEBIT_CLIENT_ENV_DIR:-$ROOT_DIR/.telebit-clients}"
 TELEBIT_BIN="${TELEBIT_BIN:-$ROOT_DIR/bin/telebit}"
 SIGNJWT_BIN="${SIGNJWT_BIN:-$ROOT_DIR/bin/signjwt}"
+TELEBIT_RETRY_DELAY="${TELEBIT_RETRY_DELAY:-5}"
+TELEBIT_RETRY_MAX="${TELEBIT_RETRY_MAX:-0}"
 
 usage() {
   cat <<'USAGE'
@@ -190,6 +192,46 @@ refresh_auth() {
   printf 'Bound %s.%s\n' "$slug" "$DOMAIN"
 }
 
+run_client_with_retries() {
+  env_file="$1"
+  attempts=0
+  stop_requested=false
+  child_pid=""
+
+  case "$TELEBIT_RETRY_DELAY" in
+    ""|*[!0-9]*) die "TELEBIT_RETRY_DELAY must be a non-negative integer" ;;
+  esac
+  case "$TELEBIT_RETRY_MAX" in
+    ""|*[!0-9]*) die "TELEBIT_RETRY_MAX must be a non-negative integer" ;;
+  esac
+
+  trap 'stop_requested=true; if [ -n "$child_pid" ]; then kill "$child_pid" 2>/dev/null || true; fi' INT TERM
+
+  while :; do
+    "$TELEBIT_BIN" --env "$env_file" &
+    child_pid=$!
+    if wait "$child_pid"; then
+      status=0
+    else
+      status=$?
+    fi
+    child_pid=""
+
+    if [ "$stop_requested" = "true" ]; then
+      exit "$status"
+    fi
+
+    attempts=$((attempts + 1))
+    if [ "$TELEBIT_RETRY_MAX" != "0" ] && [ "$attempts" -ge "$TELEBIT_RETRY_MAX" ]; then
+      printf 'Telebit exited with status %s after %s attempt(s); not retrying.\n' "$status" "$attempts" >&2
+      exit "$status"
+    fi
+
+    printf 'Telebit exited with status %s; retrying in %ss.\n' "$status" "$TELEBIT_RETRY_DELAY" >&2
+    sleep "$TELEBIT_RETRY_DELAY" || exit "$?"
+  done
+}
+
 up_client() {
   load_server_env
   need_bin "$TELEBIT_BIN"
@@ -206,7 +248,7 @@ up_client() {
   write_client_env "$slug" "$key" "$port" "$env_file" >/dev/null
   unset SECRET DOMAIN TUNNEL_DOMAIN MGMT_URL POSTGRES_PASSWORD TELEBIT_PORT MGMT_PORT VENDOR_ID
   cd "$HOME"
-  exec "$TELEBIT_BIN" --env "$env_file"
+  run_client_with_retries "$env_file"
 }
 
 build_client() {