Merge pull request #108 from plaans/fix/sched-anytime

Fix: Scheduling problems with Anytime planning

Author: arbimo

planning/unified/plugin/up_aries/solver.py

@@ -8,7 +8,7 @@
 import time
 from fractions import Fraction
 from pathlib import Path
-from typing import IO, Callable, Optional, Iterator
+from typing import IO, Callable, Optional, Iterator, Tuple
 
 import grpc
 import unified_planning as up
@@ -292,15 +292,15 @@ class Aries(AriesEngine, mixins.OneshotPlannerMixin, mixins.AnytimePlannerMixin)
     def name(self) -> str:
         return "aries"
 
-    def _solve(
+    def _prepare_solving(
         self,
         problem: "up.model.AbstractProblem",
         heuristic: Optional[
             Callable[["up.model.state.ROState"], Optional[float]]
         ] = None,
         timeout: Optional[float] = None,
         output_stream: Optional[IO[str]] = None,
-    ) -> "up.engines.results.PlanGenerationResult":
+    ) -> Tuple["_Server", proto.PlanRequest]:
         # Assert that the problem is a valid problem
         assert isinstance(problem, up.model.AbstractProblem)
         if heuristic is not None:
@@ -312,9 +312,15 @@ def _solve(
         # Note: when the `server` object is garbage collected, the process will be killed
         server = _Server(self._executable, output_stream=output_stream)
         proto_problem = self._writer.convert(problem)
-
         req = proto.PlanRequest(problem=proto_problem, timeout=timeout)
-        response = server.planner.planOneShot(req)
+
+        return server, req
+
+    def _process_response(
+        self,
+        response: proto.PlanGenerationResult,
+        problem: "up.model.AbstractProblem",
+    ) -> "up.engines.results.PlanGenerationResult":
         response = self._reader.convert(response, problem)
 
         # if we have a time triggered plan and a recent version of the UP that support setting epsilon-separation,
@@ -330,24 +336,29 @@ def _solve(
 
         return response
 
+    def _solve(
+        self,
+        problem: "up.model.AbstractProblem",
+        heuristic: Optional[
+            Callable[["up.model.state.ROState"], Optional[float]]
+        ] = None,
+        timeout: Optional[float] = None,
+        output_stream: Optional[IO[str]] = None,
+    ) -> "up.engines.results.PlanGenerationResult":
+        server, req = self._prepare_solving(problem, heuristic, timeout, output_stream)
+        response = server.planner.planOneShot(req)
+        return self._process_response(response, problem)
+
     def _get_solutions(
         self,
         problem: "up.model.AbstractProblem",
         timeout: Optional[float] = None,
         output_stream: Optional[IO[str]] = None,
     ) -> Iterator["up.engines.results.PlanGenerationResult"]:
-        # Assert that the problem is a valid problem
-        assert isinstance(problem, up.model.Problem)
-
-        # start a gRPC server in its own process
-        # Note: when the `server` object is garbage collected, the process will be killed
-        server = _Server(self._executable, output_stream=output_stream)
-        proto_problem = self._writer.convert(problem)
-
-        req = proto.PlanRequest(problem=proto_problem, timeout=timeout)
+        server, req = self._prepare_solving(problem, None, timeout, output_stream)
         stream = server.planner.planAnytime(req)
         for response in stream:
-            response = self._reader.convert(response, problem)
+            response = self._process_response(response, problem)
             yield response
             # The parallel solver implementation in aries are such that intermediate answer might arrive late
             if response.status != PlanGenerationResultStatus.INTERMEDIATE: