Merge pull request #28062 from pbehne/missing_subapp_checkpoint

Ensure duplicates are not added to list of checkpoint files.
idaholab · Jul 11, 2024 · 366d996 · 366d996
2 parents f1f4c31 + 5167b9e
commit 366d996
Show file tree

Hide file tree

Showing 5 changed files with 40 additions and 3 deletions.
diff --git a/framework/include/outputs/Checkpoint.h b/framework/include/outputs/Checkpoint.h
@@ -43,6 +43,12 @@ struct CheckpointFileNames
 
   /// Filenames for restartable data
   std::vector<std::filesystem::path> restart;
+
+  bool operator==(const CheckpointFileNames & rhs) const
+  {
+    // Compare the relevant members for equality
+    return (this->checkpoint == rhs.checkpoint) && (this->restart == rhs.restart);
+  }
 };
 
 /**

diff --git a/framework/src/outputs/Checkpoint.C b/framework/src/outputs/Checkpoint.C
@@ -183,7 +183,23 @@ Checkpoint::output()
 void
 Checkpoint::updateCheckpointFiles(CheckpointFileNames file_struct)
 {
-  // Update the list of stored files
+  // It is possible to have already written a checkpoint with the same file
+  // names contained in file_struct. If this is the case, file_struct will
+  // already be stored in _file_names. When this happens, the current state of
+  // the simulation is likely different than the state when the duplicately
+  // named checkpoint was last written. Because of this, we want to go ahead and
+  // rewrite the duplicately named checkpoint, overwritting the files
+  // representing the old state. For accurate bookkeeping, we will delete the
+  // existing instance of file_struct from _file_names and re-append it to the
+  // end of _file_names (to keep the order in which checkpoints are written
+  // accurate).
+
+  const auto it = std::find(_file_names.begin(), _file_names.end(), file_struct);
+  // file_struct was found in _file_names.
+  // Delete it so it can be re-added as the last element.
+  if (it != _file_names.end())
+    _file_names.erase(it);
+
   _file_names.push_back(file_struct);
 
   // Remove the file and the corresponding directory if it's empty

diff --git a/test/tests/outputs/checkpoint/checkpoint_child.i b/test/tests/outputs/checkpoint/checkpoint_child.i
@@ -40,7 +40,7 @@
 
 [Executioner]
   type = Transient
-  num_steps = 11
+  num_steps = 2
   dt = 0.1
   solve_type = PJFNK
   petsc_options_iname = '-pc_type -pc_hypre_type'

diff --git a/test/tests/outputs/checkpoint/checkpoint_parent.i b/test/tests/outputs/checkpoint/checkpoint_parent.i
@@ -40,7 +40,7 @@
 
 [MultiApps]
   [sub_app]
-    type = TransientMultiApp
+    type = FullSolveMultiApp
     input_files = "checkpoint_child.i"
     positions = '0 0 0'
   []

diff --git a/test/tests/outputs/checkpoint/tests b/test/tests/outputs/checkpoint/tests
@@ -349,3 +349,18 @@
     expect_err = "Shortcut checkpoint syntax cannot be used with another Checkpoint object in the 'Outputs' block"
   []
 []
+
+[subapp_only_cp]
+  issues = "#24777"
+  type = CheckFiles
+  recover = false
+  input = checkpoint_parent.i
+  cli_args = 'Outputs/checkpoint=false sub_app:Outputs/cp/type=Checkpoint sub_app:Outputs/cp/num_files=1 sub_app:Outputs/cp/execute_on=final'
+  check_files = "checkpoint_parent_out_sub_app0_cp_cp/0003-restart-0.rd/data "
+                "checkpoint_parent_out_sub_app0_cp_cp/0003-restart-0.rd/header "
+                "checkpoint_parent_out_sub_app0_cp_cp/0003-mesh.cpa.gz/1/header.gz "
+                "checkpoint_parent_out_sub_app0_cp_cp/0003-mesh.cpa.gz/1/split-1-0.gz "
+                "checkpoint_parent_out_sub_app0_cp_cp/0003-mesh.cpa.gz/meta_data_mesh.rd/data "
+                "checkpoint_parent_out_sub_app0_cp_cp/0003-mesh.cpa.gz/meta_data_mesh.rd/header"
+  requirement = "The system shall support outputting checkpoint files at the subapp level with checkpoints disabled in the main app"
+[]