Merge pull request #186 from openforcefield/compute-n_retries

Added `n_retries` usage for `SynchronousComputeService`

alchemiscale/compute/service.py

@@ -83,6 +83,7 @@ def __init__(
         scratch_basedir: os.PathLike,
         keep_shared: bool = False,
         keep_scratch: bool = False,
+        n_retries: int = 3,
         sleep_interval: int = 30,
         heartbeat_interval: int = 300,
         scopes: Optional[List[Scope]] = None,
@@ -118,6 +119,8 @@ def __init__(
         keep_scratch
             If True, don't remove scratch directories for `ProtocolUnit`s after
             completion.
+        n_retries
+            Number of times to attempt a given Task on failure.
         sleep_interval
             Time in seconds to sleep if no Tasks claimed from compute API.
         heartbeat_interval
@@ -177,6 +180,8 @@ def __init__(
         self.scratch_basedir.mkdir(exist_ok=True)
         self.keep_scratch = keep_scratch
 
+        self.n_retries = n_retries
+
         self.scheduler = sched.scheduler(time.monotonic, time.sleep)
 
         self.compute_service_id = ComputeServiceID(f"{self.name}-{uuid4()}")
@@ -336,6 +341,7 @@ def execute(self, task: ScopedKey) -> ScopedKey:
                 scratch_basedir=scratch,
                 keep_scratch=self.keep_scratch,
                 raise_error=False,
+                n_retries=self.n_retries,
             )
         finally:
             if not self.keep_shared:

devtools/configs/synchronous-compute-settings.yaml

@@ -30,6 +30,9 @@ init:
   # completion.
   keep_scratch: False
 
+  # Number of times to attempt a given Task on failure
+  n_retries: 3
+
   # Time in seconds to sleep if no Tasks claimed from compute API.
   sleep_interval: 30