Merge pull request #2460 from eisenhauer/Mismatch

Detect and warn of SST version mismatch
ornladios · Sep 12, 2020 · b046337 · b046337
2 parents c789adb + 27e7010
commit b046337
Show file tree

Hide file tree

Showing 4 changed files with 38 additions and 1 deletion.
diff --git a/source/adios2/toolkit/sst/cp/cp_common.c b/source/adios2/toolkit/sst/cp/cp_common.c
@@ -1318,6 +1318,18 @@ extern char *CP_GetContactString(SstStream Stream, attr_list DPAttrs)
     return ret;
 }
 
+static void CP_versionError(CMConnection conn, char *formatName)
+{
+    fprintf(stderr,
+            " * An invalid message of type \"%s\" has been received on an "
+            "incoming connection.\n",
+            formatName);
+    fprintf(stderr, " * In ADIOS2/SST this likely means a version mismatch "
+                    "between stream participants.\n");
+    fprintf(stderr, " * Please ensure that all writers and readers are built "
+                    "with the same version of ADIOS2.\n");
+}
+
 extern CP_Info CP_getCPInfo(CP_DP_Interface DPInfo, char *ControlModule)
 {
     CP_Info StreamCP;
@@ -1346,6 +1358,7 @@ extern CP_Info CP_getCPInfo(CP_DP_Interface DPInfo, char *ControlModule)
         }
 
         CMlisten(SharedCMInfo->cm);
+        CMregister_invalid_message_handler(SharedCMInfo->cm, CP_versionError);
 
         if (!CP_SstParamsList)
         {

diff --git a/thirdparty/EVPath/EVPath/cm.c b/thirdparty/EVPath/EVPath/cm.c
@@ -2552,7 +2552,11 @@ timeout_conn(CManager cm, void *client_data)
      local_format = FFS_target_from_encode(conn->cm->FFScontext, data_buffer);
      original_format = FFSTypeHandle_from_encode(conn->cm->FFScontext, data_buffer);
      if (local_format == NULL) {
-	 fprintf(stderr, "invalid format in incoming buffer\n");
+	 if (conn->cm->unregistered_format_handler) {
+	     conn->cm->unregistered_format_handler(conn, name_of_FMformat(FMFormat_of_original(original_format)));
+	 } else {
+	     fprintf(stderr, "No conversion found for incoming CM message\n");
+	 }
 	 return 0;
      }
      CMtrace_out(cm, CMDataVerbose, "CM - Receiving record of type %s, FFSformat %p\n", 
@@ -2716,6 +2720,12 @@ timeout_conn(CManager cm, void *client_data)
      }
  }
 
+extern void
+INT_CMregister_invalid_message_handler(CManager cm, CMUnregCMHandler handler)
+{
+    cm->unregistered_format_handler = handler;
+}
+
  extern int
  INT_CMwrite(CMConnection conn, CMFormat format, void *data)
  {

diff --git a/thirdparty/EVPath/EVPath/cm_internal.h b/thirdparty/EVPath/EVPath/cm_internal.h
@@ -136,6 +136,7 @@ typedef struct _CManager {
 
     func_entry *shutdown_functions;
     CMperf_upcall perf_upcall;
+    CMUnregCMHandler unregistered_format_handler;
 
     struct _event_path_data *evp;
     FILE * CMTrace_file;
@@ -566,3 +567,6 @@ extern void wait_for_pending_write(CMConnection conn);
 extern int
 INT_CMinstall_pull_schedule(CManager cm, struct timeval *base_time, 
 			    struct timeval *period, CMavail_period_ptr avail);
+extern void
+INT_CMregister_invalid_message_handler(CManager cm, CMUnregCMHandler handler);
+
diff --git a/thirdparty/EVPath/EVPath/evpath.h b/thirdparty/EVPath/EVPath/evpath.h
@@ -614,6 +614,16 @@ typedef int (*CMNonCMHandler) (CMConnection conn, CMTransport transport,
 extern void
 CMregister_non_CM_message_handler (int header, CMNonCMHandler handler);
 
+/*!
+ * register a handler for CM messages that don't match registered handlers.
+ *
+ */
+typedef void (*CMUnregCMHandler) (CMConnection conn, char *format_name);
+
+/*NOLOCK*/
+extern void
+CMregister_invalid_message_handler (CManager cm, CMUnregCMHandler handler);
+
 /*!
  * return the pointer to the static transport services structure.
  *