Recover transient Syncthing and SQLite job failures

src/archive_clients/jobs.py

@@ -671,6 +671,15 @@ class ClientJobExecutor:
                 # Syncthing can expose the per-job directory before the
                 # ready marker and manifest have arrived atomically as a set.
                 pass
+            except RouteSetupError as exc:
+                # A network interruption can temporarily make the local
+                # Syncthing REST API unavailable.  The staged payload is
+                # durable and the job must remain recoverable; keep polling
+                # instead of turning a transient outage into a failed job.
+                logger.warning("syncthing_status_deferred", extra={
+                    "job_id": definition.job_id,
+                    "error_type": type(exc).__name__,
+                })
             time.sleep(self.poll_interval)
 
     def _target_materialize(

src/archive_clients/state.py

@@ -7,6 +7,8 @@ import json
 import os
 import sqlite3
 import stat
+import threading
+from contextlib import contextmanager
 from dataclasses import dataclass
 from pathlib import Path
 
@@ -14,6 +16,12 @@ from pathlib import Path
 SCHEMA_VERSION = 3
 
 
+# A client daemon executes unrelated jobs concurrently.  SQLite still permits
+# only one writer, so serialize this process's short state transactions rather
+# than allowing an otherwise healthy job to fail after its busy timeout.
+_DATABASE_LOCK = threading.RLock()
+
+
 class CommandConflict(RuntimeError):
     pass
 
@@ -557,14 +565,30 @@ class ClientStore:
             ).fetchall()
         return [dict(row) for row in rows]
 
-    def _connect(self) -> sqlite3.Connection:
-        connection = sqlite3.connect(self.database, isolation_level=None, timeout=5)
-        connection.row_factory = sqlite3.Row
-        connection.execute("PRAGMA foreign_keys = ON")
-        connection.execute("PRAGMA journal_mode = WAL")
-        connection.execute("PRAGMA synchronous = FULL")
-        connection.execute("PRAGMA busy_timeout = 5000")
-        return connection
+    @contextmanager
+    def _connect(self):
+        """Yield one connection while serializing local SQLite writers.
+
+        The longer SQLite timeout also covers a short lock held by a separate
+        maintenance process such as a backup.  The lock is deliberately held
+        for the full transaction, including ``BEGIN IMMEDIATE``.
+        """
+        with _DATABASE_LOCK:
+            connection = sqlite3.connect(
+                self.database, isolation_level=None, timeout=30
+            )
+            try:
+                connection.row_factory = sqlite3.Row
+                connection.execute("PRAGMA foreign_keys = ON")
+                connection.execute("PRAGMA journal_mode = WAL")
+                connection.execute("PRAGMA synchronous = FULL")
+                connection.execute("PRAGMA busy_timeout = 30000")
+                # Preserve the original ``with connection`` commit/rollback
+                # behavior used by every store operation.
+                with connection:
+                    yield connection
+            finally:
+                connection.close()
 
 
 def _canonical(value: object) -> str: