фикс фикса

backend/app/services/discovery/db_resolver.py

@@ -27,12 +27,9 @@ async def autodetect_db_from_server(
     """
     from configparser import RawConfigParser
 
-    try:
-        config_text = await ssh_run_multi(
-            server_ip, ssh_port, ssh_options, f"cat {server_config_path}"
+    config_text = await _read_server_config_with_retry(
+        server_ip, ssh_port, ssh_options, server_config_path
     )
-    except Exception:
-        return None
 
     parser = RawConfigParser(strict=False)
     parser.read_string(config_text)
@@ -52,6 +49,34 @@ async def autodetect_db_from_server(
     return None
 
 
+async def _read_server_config_with_retry(
+    server_ip: str,
+    ssh_port: int,
+    ssh_options: list,
+    server_config_path: str,
+    attempts: int = 2,
+) -> str:
+    """Read server config, retrying once for transient command timeouts."""
+    last_exc: Exception | None = None
+    for attempt in range(1, attempts + 1):
+        try:
+            return await ssh_run_multi(
+                server_ip, ssh_port, ssh_options, f"cat {server_config_path}", timeout=30
+            )
+        except TimeoutError as exc:
+            last_exc = exc
+            if attempt < attempts:
+                logger.warning(
+                    "[discovery] Reading server config from %s timed out; retrying (%d/%d)",
+                    server_ip, attempt + 1, attempts,
+                )
+                continue
+            raise
+        except Exception:
+            raise
+    raise last_exc or RuntimeError("server config read failed")
+
+
 async def get_rack_hosts(
     host: str,
     port: int,

backend/app/services/discovery/orchestrator.py

@@ -257,9 +257,16 @@ async def _resolve_db_connection(
             return None
     elif obj.server_ip:
         await _emit("disc_action", {"message": f"Чтение конфигурации сервера {obj.server_ip}..."})
+        try:
             detected = await autodetect_db_from_server(
                 server_ip=obj.server_ip, ssh_port=ssh_port, ssh_options=ssh_options,
             )
+        except Exception as exc:
+            result.errors.append(
+                f"Could not read server config for auto-detection: {exc}. "
+                f"Fill in DB fields manually on the object."
+            )
+            return None
         if not detected or not all([detected.get("host"), detected.get("user"), detected.get("dbname")]):
             await _emit_db_autodetect_error(obj, ssh_port, ssh_options, result)
             return None
@@ -283,13 +290,25 @@ async def _resolve_db_connection(
 
 async def _emit_db_autodetect_error(obj, ssh_port, ssh_options, result: DiscoveryResult) -> None:
     try:
-        raw = await ssh_run_multi(obj.server_ip, ssh_port, ssh_options, "cat /etc/caps/config.ini")
+        raw = await ssh_run_multi(
+            obj.server_ip, ssh_port, ssh_options, "cat /etc/caps/config.ini", timeout=30
+        )
         diag_parser = RawConfigParser(strict=False)
         diag_parser.read_string(raw)
         found_sections = diag_parser.sections()
         found_keys = {s: list(diag_parser.options(s)) for s in found_sections}
+        found_db_uri_sections = [
+            s for s in found_sections
+            if diag_parser.get(s, "db_uri", fallback=None)
+        ]
+        db_uri_note = (
+            f" db_uri keys were found in sections {found_db_uri_sections}, "
+            f"but none could be parsed as a PostgreSQL URI."
+            if found_db_uri_sections else ""
+        )
         result.errors.append(
             f"db_uri not found in server config. "
+            f"{db_uri_note}"
             f"Sections found: {found_sections}. "
             f"Keys per section: {found_keys}. "
             f"Fill in DB fields manually on the object."
@@ -468,11 +487,30 @@ async def _ssh_fetch_rack_config(
     Returns (ssh_ok, config_text).
     """
     await _emit("disc_action", {"message": f"SSH → {rack_host} ({rack_name})"})
+    last_exc: Exception | None = None
+    for attempt in range(1, 3):
         try:
-        config_text = await ssh_run_multi(rack_host, ssh_port, ssh_options, f"cat {config_path}")
+            config_text = await ssh_run_multi(
+                rack_host, ssh_port, ssh_options, f"cat {config_path}", timeout=30
+            )
             return True, config_text
+        except TimeoutError as exc:
+            last_exc = exc
+            if attempt < 2:
+                logger.warning(
+                    "[discovery] Reading rack config from %s timed out; retrying (%d/2)",
+                    rack_host, attempt + 1,
+                )
+                await _emit("disc_action", {
+                    "message": f"Повторное чтение конфигурации {rack_host} ({rack_name})..."
+                })
+                continue
+            break
         except Exception as exc:
-        result.errors.append(f"SSH to {rack_host} failed: {exc}")
+            last_exc = exc
+            break
+
+    result.errors.append(f"SSH to {rack_host} failed: {last_exc}")
     return False, None
 
 

backend/app/services/discovery/ssh_client.py

@@ -141,9 +141,21 @@ async def fetch_slave_config(
     host: str, port: int, options: list, config_path: str,
 ) -> str | None:
     """Read a remote config file from a slave candidate. Returns None on failure."""
+    for attempt in range(1, 3):
         try:
-        return await ssh_run_multi(host, port, options, f"cat {config_path}", timeout=15)
-    except Exception:
+            return await ssh_run_multi(host, port, options, f"cat {config_path}", timeout=30)
+        except TimeoutError as exc:
+            if attempt < 2:
+                logger.warning(
+                    "[discovery] Reading slave config from %s timed out; retrying (%d/2)",
+                    host, attempt + 1,
+                )
+                continue
+            logger.warning("[discovery] Reading slave config from %s failed: %s", host, exc)
+            return None
+        except Exception as exc:
+            logger.warning("[discovery] Reading slave config from %s failed: %s", host, exc)
+            return None
     return None