fixup checkpoint timing race

wthrowe · Feb 7, 2025 · a895f4e · a895f4e
1 parent 81bf640
commit a895f4e
Show file tree

Hide file tree

Showing 2 changed files with 8 additions and 2 deletions.
diff --git a/src/Parallel/PhaseControl/CheckpointAndExitAfterWallclock.hpp b/src/Parallel/PhaseControl/CheckpointAndExitAfterWallclock.hpp
@@ -157,6 +157,11 @@ struct CheckpointAndExitAfterWallclock : public PhaseChange {
 
  private:
   std::optional<double> wallclock_hours_for_checkpoint_and_exit_ = std::nullopt;
+  // This flag is set during arbitration when the class decides to
+  // halt the run.  As it is not checkpointed, this distinguishes the
+  // state immediately after writing the checkpoint from that
+  // immediately after reading it during the restart.
+  //
   // Phase arbitration is only run from Main, so there are no
   // threading issues here.
   // NOLINTNEXTLINE(spectre-mutable)

diff --git a/tests/Unit/Parallel/PhaseControl/Test_CheckpointAndExitAfterWallclock.cpp b/tests/Unit/Parallel/PhaseControl/Test_CheckpointAndExitAfterWallclock.cpp
@@ -8,6 +8,7 @@
 
 #include "DataStructures/DataBox/DataBox.hpp"
 #include "Framework/TestCreation.hpp"
+#include "Framework/TestHelpers.hpp"
 #include "Options/Protocols/FactoryCreation.hpp"
 #include "Parallel/ExitCode.hpp"
 #include "Parallel/GlobalCache.hpp"
@@ -100,8 +101,8 @@ SPECTRE_TEST_CASE("Unit.Parallel.PhaseControl.CheckpointAndExitAfterWallclock",
   {
     INFO("Restarting from checkpoint");
     // Check behavior following the checkpoint phase
-    const PhaseControl::CheckpointAndExitAfterWallclock phase_change_restart(
-        0.0);
+    const PhaseControl::CheckpointAndExitAfterWallclock phase_change_restart =
+        serialize_and_deserialize(phase_change0);
     PhaseChangeDecisionData phase_change_decision_data{
         Parallel::Phase::Execute, false, true, Parallel::ExitCode::Complete};
     auto decision_result = phase_change_restart.arbitrate_phase_change(