Merge pull request #2310 from edwardhartnett/ejh_pnetcdf

fix for inq_dimlen() bug with unlimited dimensions on parallel I/O builds

libhdf5/hdf5dim.c

@@ -161,20 +161,24 @@ HDF5_inq_dim(int ncid, int dimid, char *name, size_t *lenp)
     {
         if (dim->unlimited)
         {
+            *lenp = 0;
+
+#ifndef USE_PARALLEL	    
+	    /* Shortcut for non-parallel operation: if the dim->len is
+	     * non-zero, it will be set to the correct size. */
+	    if (dim->len)
+		*lenp = dim->len;
+#endif
+
             /* Since this is an unlimited dimension, go to the file
                and see how many records there are. Take the max number
                of records from all the vars that share this
                dimension. */
-            *lenp = 0;
-            if (dim->len == 0) {
+            if (*lenp == 0)
+	    {
               if ((ret = nc4_find_dim_len(dim_grp, dimid, &lenp)))
                  return ret;
-              if (h5->no_write == NC_TRUE) {
                 dim->len = *lenp;
-              }
-            }
-            else {
-              *lenp = dim->len;
             }
         }
         else

libhdf5/hdf5internal.c

@@ -117,6 +117,8 @@ find_var_dim_max_length(NC_GRP_INFO_T *grp, int varid, int dimid,
 
     *maxlen = 0;
 
+    LOG((3, "find_var_dim_max_length varid %d dimid %d", varid, dimid));    
+
     /* Find this var. */
     var = (NC_VAR_INFO_T*)ncindexith(grp->vars,varid);
     if (!var) return NC_ENOTVAR;
@@ -157,11 +159,27 @@ find_var_dim_max_length(NC_GRP_INFO_T *grp, int varid, int dimid,
                 BAIL(NC_EHDFERR);
             LOG((5, "find_var_dim_max_length: varid %d len %d max: %d",
                  varid, (int)h5dimlen[0], (int)h5dimlenmax[0]));
-            for (d=0; d<dataset_ndims; d++) {
-                if (var->dimids[d] == dimid) {
+            for (d=0; d<dataset_ndims; d++)
+                if (var->dimids[d] == dimid)
                     *maxlen = *maxlen > h5dimlen[d] ? *maxlen : h5dimlen[d];
-                }
-            }
+
+#ifdef USE_PARALLEL
+	    /* If we are doing parallel I/O in collective mode (with
+	     * either pnetcdf or HDF5), then communicate with all
+	     * other tasks in the collective and find out which has
+	     * the max value for the dimension size. */
+	    assert(grp->nc4_info);
+	    LOG((3, "before Allreduce *maxlen %ld grp->nc4_info->parallel %d var->parallel_access %d",
+		 *maxlen, grp->nc4_info->parallel, var->parallel_access));
+	    if (grp->nc4_info->parallel && var->parallel_access == NC_COLLECTIVE)
+	    {
+		if ((MPI_SUCCESS != MPI_Allreduce(MPI_IN_PLACE, maxlen, 1,
+						  MPI_UNSIGNED_LONG_LONG, MPI_MAX,
+						  grp->nc4_info->comm)))
+		    BAIL(NC_EMPI);
+		LOG((3, "after Allreduce *maxlen %ld", *maxlen));
+	    }
+#endif /* USE_PARALLEL */
         }
     }
 

nc_test4/Makefile.am

@@ -93,7 +93,8 @@ endif # BUILD_UTILITIES
 if TEST_PARALLEL4
 check_PROGRAMS += tst_mpi_parallel tst_parallel tst_parallel3		\
 tst_parallel4 tst_parallel5 tst_nc4perf tst_mode tst_simplerw_coll_r	\
-tst_mode tst_parallel_zlib tst_parallel_compress tst_quantize_par
+tst_mode tst_parallel_zlib tst_parallel_compress tst_quantize_par	\
+tst_parallel6
 TESTS += run_par_test.sh
 endif # TEST_PARALLEL4
 

nc_test4/run_par_test.sh.in

@@ -65,3 +65,7 @@ echo
 echo "Parallel I/O test for quantize feature."
 @MPIEXEC@ -n 4 ./tst_quantize_par
 
+echo
+echo "Parallel I/O test contributed by wkliao from pnetcdf."
+@MPIEXEC@ -n 4 ./tst_parallel6
+

nc_test4/tst_parallel6.c

@@ -0,0 +1,74 @@
+/* Copyright 2022, UCAR/Unidata See COPYRIGHT file for copying and
+ * redistribution conditions.
+ *
+ * This parallel I/O test checks the behavior of nc_inq_dimlen() after
+ * parallel I/O writes.
+ *
+ * This program taken from a PNetCDF issue:
+ * https://github.com/Parallel-NetCDF/PnetCDF/issues/72, thanks
+ * wkliao!
+ *
+ * wkliao, Ed Hartnett, 4/11/22
+ */
+
+#include <nc_tests.h>
+#include "err_macros.h"
+#include <stdio.h>
+#include <stdlib.h>
+#include <mpi.h>
+#include <netcdf.h>
+#include <netcdf_par.h>
+
+#define FILENAME "tst_parallel6.nc"
+
+int main(int argc, char** argv)
+{
+    int err = NC_NOERR, rank, nprocs;
+    int ncid, varid, dimid;
+    size_t start[1], count[1], nrecs;
+
+    MPI_Init(&argc, &argv);
+    MPI_Comm_size(MPI_COMM_WORLD, &nprocs);
+    MPI_Comm_rank(MPI_COMM_WORLD, &rank);
+
+    if (!rank)
+        printf("\n*** Testing parallel I/O.\n");
+
+    if (!rank)
+        printf("*** testing record lenth with multiple processes writing records...");
+
+    /* nc_set_log_level(4); */
+    if (nc_create_par(FILENAME, NC_CLOBBER | NC_NETCDF4, MPI_COMM_WORLD,
+		      MPI_INFO_NULL, &ncid)) ERR;
+
+    if (nc_def_dim(ncid, "time", NC_UNLIMITED, &dimid)) ERR;
+    if (nc_def_var(ncid, "var", NC_INT, 1, &dimid, &varid)) ERR;
+    if (nc_var_par_access(ncid, varid, NC_COLLECTIVE)) ERR;
+    if (nc_enddef(ncid)) ERR;
+
+    start[0] = rank;
+    count[0] = 1;
+    if (nc_put_vara_int(ncid, varid, start, count, &rank)) ERR;
+    if (nc_inq_dimlen(ncid, dimid, &nrecs)) ERR;
+    if (nc_close(ncid)) ERR;
+    /* nc_set_log_level(-1); */
+
+    if (nrecs != nprocs)
+    {
+	printf("Rank %d error at line %d of file %s:\n",rank,__LINE__,__FILE__);
+	printf("\tafter writing start=%zd count=%zd\n", start[0], count[0]);
+	printf("\texpecting number of records = %d but got %ld\n",
+	       nprocs, nrecs);
+	ERR;
+    }
+
+    if (!rank)
+        SUMMARIZE_ERR;
+
+    MPI_Finalize();
+
+    if (!rank)
+        FINAL_RESULTS;
+
+    return 0;
+}