ompi/instance: fix cleanup function registration order

- Append PML cleanup into the finalize of the instance domain ('ompi_instance_common_domain')
  before RTE/OPAL init.

- The reason is RTE init (ompi_rte_init()) will call opal_init(), which in turn
  will set the internal tracking domain to OPAL's one ('opal_init_domain'), and
  this PML cleanup function would be mis-registered as belonging to 'opal_init_domain'
  instead of the current 'ompi_instance_common_domain'.

- The consequence of such mis-registration is that: at MPI_Finalize(), this
  PML cleanup (*_del_procs()) will be executed by RTE; and, depending on
  their registration order, this may cut the grass under the feet of other
  running components (*_progress())

- This may be the root cause of issue #10117

Signed-off-by: Minh Quan Ho <[email protected]>

Author: Minh Quan Ho

ompi/instance/instance.c

@@ -8,6 +8,7 @@
  *                         reserved.
  * Copyright (c) 2023      Jeffrey M. Squyres.  All rights reserved.
  * Copyright (c) 2024      NVIDIA Corporation.  All rights reserved.
+ * Copyright (c) 2025      SiPearl.  All rights reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -381,6 +382,10 @@ static int ompi_mpi_instance_init_common (int argc, char **argv)
     opal_finalize_domain_init (&ompi_instance_common_domain, "ompi_mpi_instance_init_common");
     opal_finalize_set_domain (&ompi_instance_common_domain);
 
+    /* Append PML cleanup into the finalize of this domain ('ompi_instance_common_domain')
+       before RTE init */
+    ompi_mpi_instance_append_finalize (ompi_mpi_instance_cleanup_pml);
+
     if (OPAL_SUCCESS != (ret = opal_arch_set_fortran_logical_size(sizeof(ompi_fortran_logical_t)))) {
         return ompi_instance_print_error ("ompi_mpi_init: opal_arch_set_fortran_logical_size failed", ret);
     }
@@ -638,8 +643,6 @@ static int ompi_mpi_instance_init_common (int argc, char **argv)
         return ompi_instance_print_error ("ompi_group_init() failed", ret);
     }
 
-    ompi_mpi_instance_append_finalize (ompi_mpi_instance_cleanup_pml);
-
     /* initialize communicator subsystem */
     if (OMPI_SUCCESS != (ret = ompi_comm_init ())) {
         opal_mutex_unlock (&instance_lock);
@@ -906,8 +909,6 @@ static int ompi_mpi_instance_finalize_common (void)
         mca_mpool_base_tree_print (ompi_debug_show_mpi_alloc_mem_leaks);
     }
 
-    opal_finalize_cleanup_domain (&ompi_instance_common_domain);
-
     if (NULL != ompi_mpi_main_thread) {
         OBJ_RELEASE(ompi_mpi_main_thread);
         ompi_mpi_main_thread = NULL;
@@ -936,6 +937,9 @@ static int ompi_mpi_instance_finalize_common (void)
 
     ompi_rte_initialized = false;
 
+    /* Should be called in reverse order of init, i.e. after RTE finalize */
+    opal_finalize_cleanup_domain (&ompi_instance_common_domain);
+
     for (int i = 0 ; ompi_lazy_frameworks[i] ; ++i) {
         if (0 < ompi_lazy_frameworks[i]->framework_refcnt) {
             /* May have been "opened" multiple times. We want it closed now! */