feat: allow unrestricted independent job concurrency

src/archive_clients/jobs.py

@@ -86,6 +86,8 @@ class ClientJobExecutor:
         self.verification_timeout = verification_timeout
         self.free_space_reserve_bytes = free_space_reserve_bytes
         self._cancel_events: dict[str, threading.Event] = {}
+        self._execution_locks: dict[str, threading.Lock] = {}
+        self._execution_locks_guard = threading.Lock()
 
     def request_cancel(self, job_id: str) -> None:
         self._cancel_events.setdefault(job_id, threading.Event()).set()
@@ -115,6 +117,22 @@ class ClientJobExecutor:
         self,
         command: control_pb2.ExecuteStepCommand,
         event_callback: Callable[[control_pb2.JobEvent], None] | None = None,
+    ) -> list[control_pb2.JobEvent]:
+        # asyncio cancellation of a connection-bound task cannot stop the
+        # synchronous filesystem/qB operation already running in its worker
+        # thread. A replay after reconnect therefore waits for that operation
+        # and then reads its durable event journal instead of executing twice.
+        with self._execution_lock(command.job_id):
+            return self._execute_locked(command, event_callback)
+
+    def _execution_lock(self, job_id: str) -> threading.Lock:
+        with self._execution_locks_guard:
+            return self._execution_locks.setdefault(job_id, threading.Lock())
+
+    def _execute_locked(
+        self,
+        command: control_pb2.ExecuteStepCommand,
+        event_callback: Callable[[control_pb2.JobEvent], None] | None = None,
     ) -> list[control_pb2.JobEvent]:
         definition = self._definition(command.job_id)
         replay = self._replay(