Fix a bug with server and client number calculation

hpc-io · Nov 15, 2024 · c3b5ecd · c3b5ecd
1 parent 2e2d036
commit c3b5ecd
Show file tree

Hide file tree

Showing 4 changed files with 14 additions and 13 deletions.
diff --git a/src/api/pdc_client_connect.c b/src/api/pdc_client_connect.c
@@ -1638,7 +1638,7 @@ PDC_Client_init()
     // Split the PDC_CLIENT_COMM_WORLD_g communicator, MPI_Comm_split_type requires MPI-3
     /* MPI_Comm_split_type(PDC_CLIENT_COMM_WORLD_g, MPI_COMM_TYPE_SHARED, 0, MPI_INFO_NULL, */
     /*                     &PDC_SAME_NODE_COMM_g); */
-    int sub_comm_color = pdc_client_mpi_rank_g / pdc_server_num_g;
+    int sub_comm_color = pdc_client_mpi_rank_g / (pdc_client_mpi_size_g / pdc_server_num_g);
     MPI_Comm_split(PDC_CLIENT_COMM_WORLD_g, sub_comm_color, pdc_client_mpi_rank_g, &PDC_SAME_NODE_COMM_g);
 
     MPI_Comm_rank(PDC_SAME_NODE_COMM_g, &pdc_client_same_node_rank_g);
@@ -9419,7 +9419,7 @@ PDC_Client_query_kvtag_mpi(const pdc_kvtag_t *kvtag, int *n_res, uint64_t **pdc_
         // Only retrieve the metadata snapshot once
         // TODO: if metadata is changed, need to invalidate the cached snapshot and retrieve a new one
         if (deserializedBulki_g == NULL) {
-            server_rank = pdc_client_mpi_rank_g / pdc_server_num_g;
+            server_rank = pdc_client_mpi_rank_g / (pdc_client_mpi_size_g / pdc_server_num_g);
             // Send one request to each server at the same time
             if (pdc_client_same_node_rank_g == 0)
                 ret_value = PDC_Client_query_kvtag_server(server_rank, kvtag, &nshm, &shm_sizes);

diff --git a/src/server/pdc_server.c b/src/server/pdc_server.c
@@ -721,7 +721,8 @@ PDC_Server_set_close(void)
         start = MPI_Wtime();
 #endif
         char *tmp_env_char = getenv("PDC_DISABLE_CHECKPOINT");
-        if (tmp_env_char != NULL && strcmp(tmp_env_char, "TRUE") == 0) {
+        if (tmp_env_char != NULL && 
+                (strcmp(tmp_env_char, "TRUE") == 0 || atoi(tmp_env_char) == 1)) {
             if (pdc_server_rank_g == 0) {
                 printf("==PDC_SERVER[0]: checkpoint disabled!\n");
             }

diff --git a/src/server/pdc_server_metadata.c b/src/server/pdc_server_metadata.c
@@ -1894,13 +1894,13 @@ PDC_Server_seralize_kvtag_someta_to_shm(uint32_t *n_meta, uint64_t **obj_ids, ui
         goto done;
     }
 
-    if (pdc_server_size_g % pdc_client_num_g != 0) {
+    if (pdc_client_num_g % pdc_server_size_g != 0) {
         printf("==PDC_SERVER[%d]: #servers not divisible by clients!\n", pdc_server_rank_g);
         ret_value = FAIL;
         goto done;
     }
 
-    nclient_per_server = pdc_server_size_g / pdc_client_num_g;
+    nclient_per_server = pdc_client_num_g / pdc_server_size_g;
     nkvtag_per_buf     = ceil(metadata_total_count_g / nclient_per_server);
 
     if (alloc_size < nclient_per_server) {
@@ -1949,7 +1949,7 @@ PDC_Server_seralize_kvtag_someta_to_shm(uint32_t *n_meta, uint64_t **obj_ids, ui
                                                kvtag_list_elt->kvtag->type, PDC_CLS_ITEM);
                         }
 
-                        BULKI_append(kvtag_bulki, key, val);
+                        BULKI_put(kvtag_bulki, key, val);
                         nkvtag_in_buf++;
 
                     } // End for each kvtag in list

diff --git a/src/tests/kvtag_query.c b/src/tests/kvtag_query.c
@@ -79,7 +79,7 @@ main(int argc, char *argv[])
 {
     pdcid_t     pdc, cont_prop, cont, obj_prop;
     pdcid_t *   obj_ids;
-    int         n_obj, n_add_tag, my_obj, my_obj_s, my_add_tag, my_add_tag_s;
+    int         n_obj, n_add_tag, my_obj, my_obj_s, my_add_tag, my_add_tag_s, my_tags, total_tags;
     int         proc_num = 1, my_rank = 0, i, v, iter, round, selectivity, is_using_dart;
     char        obj_name[128];
     double      stime, total_time;
@@ -156,6 +156,7 @@ main(int argc, char *argv[])
     assign_work_to_rank(my_rank, proc_num, n_add_tag, &my_add_tag, &my_add_tag_s);
 
     // This is for adding #rounds tags to the objects.
+    my_tags = 0;
     for (i = 0; i < my_add_tag; i++) {
         for (iter = 0; iter < round; iter++) {
             v = iter;
@@ -172,11 +173,16 @@ main(int argc, char *argv[])
                     printf("fail to add a kvtag to o%d\n", i + my_obj_s);
                 }
             }
+            my_tags++;
         }
         /* if (my_rank == 0) */
         /*     println("Rank %d: Added %d kvtag to the %d th object", my_rank, round, i); */
     }
 
+    MPI_Reduce(&my_tags, &total_tags, 1, MPI_INT, MPI_MAX, 0, MPI_COMM_WORLD);
+    if (my_rank == 0)
+        println("Added %d kvtags", total_tags);
+
 #ifdef ENABLE_MPI
     MPI_Barrier(MPI_COMM_WORLD);
 #endif
@@ -225,20 +231,14 @@ main(int argc, char *argv[])
     // close a container
     if (PDCcont_close(cont) < 0)
         printf("fail to close container c1\n");
-    else
-        printf("successfully close container c1\n");
 
     // close an object property
     if (PDCprop_close(obj_prop) < 0)
         printf("Fail to close property @ line %d\n", __LINE__);
-    else
-        printf("successfully close object property\n");
 
     // close a container property
     if (PDCprop_close(cont_prop) < 0)
         printf("Fail to close property @ line %d\n", __LINE__);
-    else
-        printf("successfully close container property\n");
 
     // close pdc
     if (PDCclose(pdc) < 0)