Fix retry count (#363)

Author: xiaoyeqiannian

dpdispatcher/dp_cloud_server_context.py

@@ -307,7 +307,7 @@ def machine_subfields(cls) -> List[Argument]:
                         "retry_count",
                         [int, type(None)],
                         optional=True,
-                        default=3,
+                        default=2,
                         doc=doc_retry_count,
                     ),
                     Argument(

dpdispatcher/submission.py

@@ -841,7 +841,7 @@ def handle_unexpected_job_state(self):
             retry_count = 3
             assert self.machine is not None
             if hasattr(self.machine, "retry_count") and self.machine.retry_count > 0:
-                retry_count = self.machine.retry_count
+                retry_count = self.machine.retry_count + 1
             if (self.fail_count) > 0 and (self.fail_count % retry_count == 0):
                 raise RuntimeError(
                     f"job:{self.job_hash} {self.job_id} failed {self.fail_count} times.job_detail:{self}"