Fix issue with Subfiling VFD and multiple opens of same file

HDFGroup · Mar 20, 2024 · bdc1d52 · bdc1d52
1 parent 87fb6ed
commit bdc1d52
Show file tree

Hide file tree

Showing 5 changed files with 215 additions and 178 deletions.
diff --git a/release_docs/RELEASE.txt b/release_docs/RELEASE.txt
@@ -663,6 +663,16 @@ Bug Fixes since HDF5-1.14.0 release
 
     Library
     -------
+    - Fixed an issue with the Subfiling VFD and multiple opens of a
+      file
+
+      An issue with the way the Subfiling VFD handles multiple opens
+      of the same file caused the file structures for the extra opens
+      to occasionally get mapped to an incorrect subfiling context
+      object. The VFD now correctly maps the file structures for
+      additional opens of an already open file to the same context
+      object.
+
     - Fixed a bug that causes the library to incorrectly identify
       the endian-ness of 16-bit and smaller C floating-point datatypes
 

diff --git a/src/H5FDsubfiling/H5FDioc.c b/src/H5FDsubfiling/H5FDioc.c
@@ -843,12 +843,17 @@ H5FD__ioc_open(const char *name, unsigned flags, hid_t fapl_id, haddr_t maxaddr)
         H5_SUBFILING_GOTO_ERROR(H5E_FILE, H5E_CANTOPENFILE, NULL, "unable to open subfiles for file '%s'",
                                 name);
 
-    /* Initialize I/O concentrator threads if this MPI rank is an I/O concentrator */
+    /*
+     * Initialize I/O concentrator threads if this MPI rank is an I/O
+     * concentrator and the threads haven't already been initialized by
+     * a different open of this file
+     */
     sf_context = H5_get_subfiling_object(file_ptr->context_id);
-    if (sf_context && sf_context->topology->rank_is_ioc) {
+    if (sf_context && sf_context->topology->rank_is_ioc && !sf_context->threads_inited) {
         if (initialize_ioc_threads(sf_context) < 0)
             H5_SUBFILING_GOTO_ERROR(H5E_FILE, H5E_CANTINIT, NULL,
                                     "unable to initialize I/O concentrator threads");
+        sf_context->threads_inited = true;
     }
 
     ret_value = (H5FD_t *)file_ptr;
@@ -917,14 +922,22 @@ H5FD__ioc_close_int(H5FD_ioc_t *file_ptr)
             if (MPI_SUCCESS != (mpi_code = MPI_Barrier(file_ptr->comm)))
                 H5_SUBFILING_MPI_GOTO_ERROR(FAIL, "MPI_Barrier failed", mpi_code);
 
-        if (sf_context && sf_context->topology->rank_is_ioc) {
-            if (finalize_ioc_threads(sf_context) < 0)
-                /* Note that closing of subfiles is collective */
-                H5_SUBFILING_DONE_ERROR(H5E_VFL, H5E_CANTCLOSEFILE, FAIL, "unable to finalize IOC threads");
+        /* Only finalize IOC threads and close subfiles if this is
+         * the last file holding a reference to the context
+         */
+        if (sf_context && sf_context->file_ref == 1) {
+            if (sf_context->topology->rank_is_ioc && sf_context->threads_inited) {
+                if (finalize_ioc_threads(sf_context) < 0)
+                    /* Note that closing of subfiles is collective */
+                    H5_SUBFILING_DONE_ERROR(H5E_VFL, H5E_CANTCLOSEFILE, FAIL,
+                                            "unable to finalize IOC threads");
+            }
+
+            if (H5_close_subfiles(file_ptr->context_id, file_ptr->comm) < 0)
+                H5_SUBFILING_GOTO_ERROR(H5E_VFL, H5E_CANTCLOSEFILE, FAIL,
+                                        "unable to close subfiling file(s)");
         }
 
-        if (H5_close_subfiles(file_ptr->context_id, file_ptr->comm) < 0)
-            H5_SUBFILING_GOTO_ERROR(H5E_VFL, H5E_CANTCLOSEFILE, FAIL, "unable to close subfiling file(s)");
         file_ptr->context_id = -1;
     }