Доведен до ума и стабилизирован модуль ZFS

modules/zfs_backup_ops.py

@@ -10,6 +10,9 @@ import re
 from datetime import datetime, timedelta
 from typing import Any, Dict, List, Optional, Tuple
 
+# Результат batch-операции: (успешно, список ошибок)
+# ошибка = (server/dataset_key, сообщение) — например "gwo2.mps.cln.su/containers/smb2"
+
 from .logger import get_logger
 from .protocols import SSHProtocol
 from .ssh_base import SSHBase
@@ -130,14 +133,18 @@ def backup_server(
     server: Dict[str, Any],
     ssh_defaults: Dict[str, Any],
     log: logging.Logger,
-) -> bool:
+    collect_errors: bool = False,
+) -> Tuple[int, List[Tuple[str, str]]]:
     """
     Выполнить бэкап для одного сервера: снапшоты, send/recv, очистка.
     Конфиг v1.1: server['pools'] — список пулов с source_pool, datasets, target_pool.
+    Возвращает (успешно, список_ошибок), где ошибка = (ключ "server/dataset", сообщение).
     """
     name = server["name"]
     pools = server["pools"]
     retention_days = int(server.get("retention_days", 30))
+    success_count = 0
+    errors: List[Tuple[str, str]] = []
 
     port = ssh_defaults.get("port", 22)
     username = ssh_defaults.get("username", "root")
@@ -148,27 +155,35 @@ def backup_server(
     log.info("Сервер %s: дата снапшотов %s", name, date_str)
 
     ssh = SSHBase(hostname=name, port=port, username=username, pkey_file=pkey_file, host_keys=host_keys)
-    ssh.connect()
-
-    pool_counts: Dict[str, int] = {}
-    total_datasets = 0
+    try:
+        ssh.connect()
+    except Exception as e:
+        for pool_config in pools:
+            for dataset in pool_config["datasets"]:
+                errors.append((f"{name}/{dataset}", str(e)))
+        return 0, errors
 
     try:
         for pool_config in pools:
             source_pool = pool_config["source_pool"]
             datasets = pool_config["datasets"]
             target_pool = pool_config["target_pool"]
-            pool_counts[source_pool] = len(datasets)
-            total_datasets += len(datasets)
 
             for dataset in datasets:
                 full_dataset = f"{source_pool}/{dataset}"
+                key = f"{name}/{dataset}"
 
                 def do_snapshot(fd=full_dataset, d=date_str):
                     if not _create_snapshot(ssh, fd, d, log):
                         raise RuntimeError(f"Snapshot {fd}@{d} failed")
 
-                _run_with_retry(log, name, f"snapshot {full_dataset}", do_snapshot)
+                try:
+                    _run_with_retry(log, name, f"snapshot {full_dataset}", do_snapshot)
+                except Exception as e:
+                    errors.append((key, str(e)))
+                    if not collect_errors:
+                        raise
+                    continue
 
         for pool_config in pools:
             source_pool = pool_config["source_pool"]
@@ -178,6 +193,7 @@ def backup_server(
             for dataset in datasets:
                 full_dataset = f"{source_pool}/{dataset}"
                 target_dataset = f"{target_pool}/{dataset}"
+                key = f"{name}/{dataset}"
 
                 def do_send_recv(fd=full_dataset, td=target_dataset, d=date_str):
                     prev_date = _get_previous_snapshot_date(ssh, td)
@@ -188,14 +204,20 @@ def backup_server(
                     if not _replicate_snapshot(ssh, fd, td, d, prev_date, log):
                         raise RuntimeError(f"Replicate {fd}@{d} → {td} failed")
 
-                _run_with_retry(log, name, f"send/recv {full_dataset}", do_send_recv)
+                try:
+                    _run_with_retry(log, name, f"send/recv {full_dataset}", do_send_recv)
+                    success_count += 1
+                except Exception as e:
+                    errors.append((key, str(e)))
+                    if not collect_errors:
+                        raise
+                    continue
 
         cutoff = datetime.now() - timedelta(days=retention_days)
         for pool_config in pools:
             source_pool = pool_config["source_pool"]
             datasets = pool_config["datasets"]
             target_pool = pool_config["target_pool"]
-
             for dataset in datasets:
                 for pool, ds in [(source_pool, dataset), (target_pool, dataset)]:
                     full_ds = f"{pool}/{ds}"
@@ -218,9 +240,9 @@ def backup_server(
                         except ValueError:
                             continue
 
-        parts = ", ".join(f"{p}:{c}" for p, c in sorted(pool_counts.items()))
-        log.info("✅ %s: %s datasets backed up (%s)", name, total_datasets, parts)
-        return True
+        if not collect_errors:
+            success_count = sum(len(pc["datasets"]) for pc in pools)
+        return success_count, errors
 
     finally:
         ssh.close()