ompi_info --show-failed feature

opal/mca/base/base.h

@@ -15,6 +15,7 @@
  *                         reserved.
  * Copyright (c) 2015      Research Organization for Information Science
  *                         and Technology (RIST). All rights reserved.
+ * Copyright (c) 2017 IBM Corporation.  All rights reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -68,6 +69,7 @@ OPAL_DECLSPEC OBJ_CLASS_DECLARATION(mca_base_component_priority_list_item_t);
  */
 OPAL_DECLSPEC extern char *mca_base_component_path;
 OPAL_DECLSPEC extern bool mca_base_component_show_load_errors;
+OPAL_DECLSPEC extern bool mca_base_component_track_load_errors;
 OPAL_DECLSPEC extern bool mca_base_component_disable_dlopen;
 OPAL_DECLSPEC extern char *mca_base_system_default_path;
 OPAL_DECLSPEC extern char *mca_base_user_default_path;

opal/mca/base/mca_base_component_repository.c

@@ -15,6 +15,7 @@
  *                         reserved.
  * Copyright (c) 2015      Research Organization for Information Science
  *                         and Technology (RIST). All rights reserved.
+ * Copyright (c) 2017 IBM Corporation.  All rights reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -55,6 +56,29 @@ OBJ_CLASS_INSTANCE(mca_base_component_repository_item_t, opal_list_item_t,
 
 #endif /* OPAL_HAVE_DL_SUPPORT */
 
+static void clf_constructor(opal_object_t *obj);
+static void clf_destructor(opal_object_t *obj);
+
+OBJ_CLASS_INSTANCE(mca_base_failed_component_t, opal_list_item_t,
+                   clf_constructor, clf_destructor);
+
+
+static void clf_constructor(opal_object_t *obj)
+{
+    mca_base_failed_component_t *cli = (mca_base_failed_component_t *) obj;
+    cli->comp = NULL;
+    cli->error_msg = NULL;
+}
+
+static void clf_destructor(opal_object_t *obj)
+{
+    mca_base_failed_component_t *cli = (mca_base_failed_component_t *) obj;
+    cli->comp = NULL;
+    if( NULL != cli->error_msg ) {
+        free(cli->error_msg);
+        cli->error_msg = NULL;
+    }
+}
 
 /*
  * Private variables
@@ -408,6 +432,14 @@ int mca_base_component_repository_open (mca_base_framework_t *framework,
         }
         opal_output_verbose(vl, 0, "mca_base_component_repository_open: unable to open %s: %s (ignored)",
                             ri->ri_base, err_msg);
+
+        if( mca_base_component_track_load_errors ) {
+            mca_base_failed_component_t *f_comp = OBJ_NEW(mca_base_failed_component_t);
+            f_comp->comp = ri;
+            asprintf(&(f_comp->error_msg), "%s", err_msg);
+            opal_list_append(&framework->framework_failed_components, &f_comp->super);
+        }
+
         return OPAL_ERR_BAD_PARAM;
     }
 

opal/mca/base/mca_base_component_repository.h

@@ -13,6 +13,7 @@
  * Copyright (c) 2015 Cisco Systems, Inc.  All rights reserved.
  * Copyright (c) 2015      Los Alamos National Security, LLC. All rights
  *                         reserved.
+ * Copyright (c) 2017 IBM Corporation.  All rights reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -60,6 +61,17 @@ typedef struct mca_base_component_repository_item_t mca_base_component_repositor
 
 OBJ_CLASS_DECLARATION(mca_base_component_repository_item_t);
 
+/*
+ * Structure to track information about why a component failed to load.
+ */
+struct mca_base_failed_component_t {
+    opal_list_item_t super;
+    mca_base_component_repository_item_t *comp;
+    char *error_msg;
+};
+typedef struct mca_base_failed_component_t mca_base_failed_component_t;
+OPAL_DECLSPEC OBJ_CLASS_DECLARATION(mca_base_failed_component_t);
+
 /**
  * @brief initialize the component repository
  *

opal/mca/base/mca_base_framework.c

@@ -3,6 +3,7 @@
  * Copyright (c) 2012-2015 Los Alamos National Security, LLC. All rights
  *                         reserved.
  * Copyright (c) 2015 Cisco Systems, Inc.  All rights reserved.
+ * Copyright (c) 2017 IBM Corporation.  All rights reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -66,6 +67,7 @@ int mca_base_framework_register (struct mca_base_framework_t *framework,
     }
 
     OBJ_CONSTRUCT(&framework->framework_components, opal_list_t);
+    OBJ_CONSTRUCT(&framework->framework_failed_components, opal_list_t);
 
     if (framework->framework_flags & MCA_BASE_FRAMEWORK_FLAG_NO_DSO) {
         flags |= MCA_BASE_REGISTER_STATIC_ONLY;
@@ -228,12 +230,16 @@ int mca_base_framework_close (struct mca_base_framework_t *framework) {
                                       framework->framework_output);
             OBJ_RELEASE(item);
         }
+        while (NULL != (item = opal_list_remove_first (&framework->framework_failed_components))) {
+            OBJ_RELEASE(item);
+        }
         ret = OPAL_SUCCESS;
     }
 
     framework->framework_flags &= ~(MCA_BASE_FRAMEWORK_FLAG_REGISTERED | MCA_BASE_FRAMEWORK_FLAG_OPEN);
 
     OBJ_DESTRUCT(&framework->framework_components);
+    OBJ_DESTRUCT(&framework->framework_failed_components);
 
     framework_close_output (framework);
 

opal/mca/base/mca_base_framework.h

@@ -2,6 +2,7 @@
 /*
  * Copyright (c) 2012-2015 Los Alamos National Security, LLC. All rights
  *                         reserved.
+ * Copyright (c) 2017 IBM Corporation.  All rights reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -154,6 +155,8 @@ typedef struct mca_base_framework_t {
     /** List of selected components (filled in by mca_base_framework_register()
         or mca_base_framework_open() */
     opal_list_t                              framework_components;
+    /** List of components that failed to load */
+    opal_list_t                              framework_failed_components;
 } mca_base_framework_t;
 
 

opal/mca/base/mca_base_open.c

@@ -49,6 +49,7 @@ int mca_base_opened = 0;
 char *mca_base_system_default_path = NULL;
 char *mca_base_user_default_path = NULL;
 bool mca_base_component_show_load_errors = true;
+bool mca_base_component_track_load_errors = false;
 bool mca_base_component_disable_dlopen = false;
 
 static char *mca_base_verbose = NULL;
@@ -111,6 +112,14 @@ int mca_base_open(void)
     (void) mca_base_var_register_synonym(var_id, "opal", "mca", NULL, "component_show_load_errors",
                                          MCA_BASE_VAR_SYN_FLAG_DEPRECATED);
 
+    mca_base_component_track_load_errors = false;
+    var_id = mca_base_var_register("opal", "mca", "base", "component_track_load_errors",
+                                   "Whether to track errors for components that failed to load or not",
+                                   MCA_BASE_VAR_TYPE_BOOL, NULL, 0, 0,
+                                   OPAL_INFO_LVL_9,
+                                   MCA_BASE_VAR_SCOPE_READONLY,
+                                   &mca_base_component_track_load_errors);
+
     mca_base_component_disable_dlopen = false;
     var_id = mca_base_var_register("opal", "mca", "base", "component_disable_dlopen",
                                    "Whether to attempt to disable opening dynamic components or not",

opal/runtime/opal_info_support.c

@@ -15,6 +15,7 @@
  *                         reserved.
  * Copyright (c) 2011-2012 University of Houston. All rights reserved.
  * Copyright (c) 2016      Intel, Inc.  All rights reserved.
+ * Copyright (c) 2017 IBM Corporation.  All rights reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -50,6 +51,7 @@
 #include "opal/mca/installdirs/installdirs.h"
 
 #include "opal/runtime/opal_info_support.h"
+#include "opal/mca/base/mca_base_component_repository.h"
 
 const char *opal_info_path_prefix = "prefix";
 const char *opal_info_path_bindir = "bindir";
@@ -109,6 +111,9 @@ OBJ_CLASS_INSTANCE(opal_info_component_map_t,
                    component_map_construct,
                    component_map_destruct);
 
+static void opal_info_show_failed_component(const mca_base_component_repository_item_t* ri,
+                                            const char *error_msg);
+
 int opal_info_init(int argc, char **argv,
                    opal_cmd_line_t *opal_info_cmd_line)
 {
@@ -157,6 +162,8 @@ int opal_info_init(int argc, char **argv,
                             "Show only variables with at most this level (1-9)");
     opal_cmd_line_make_opt3(opal_info_cmd_line, 's', NULL, "selected-only", 0,
                             "Show only variables from selected components");
+    opal_cmd_line_make_opt3(opal_info_cmd_line, '\0', NULL, "show-failed", 0,
+                            "Show the components that failed to load along with the reason why they failed.");
 
     /* set our threading level */
     opal_set_using_threads(false);
@@ -223,6 +230,10 @@ int opal_info_init(int argc, char **argv,
         opal_info_register_flags = MCA_BASE_REGISTER_DEFAULT;
     }
 
+    if( opal_cmd_line_is_taken(opal_info_cmd_line, "show-failed") ) {
+        mca_base_component_track_load_errors = true;
+    }
+
     return OPAL_SUCCESS;
 }
 
@@ -245,6 +256,7 @@ static int info_register_framework (mca_base_framework_t *framework, opal_pointe
         map = OBJ_NEW(opal_info_component_map_t);
         map->type = strdup(framework->framework_name);
         map->components = &framework->framework_components;
+        map->failed_components = &framework->framework_failed_components;
         opal_pointer_array_add(component_map, map);
     }
 
@@ -1012,6 +1024,7 @@ void opal_info_show_component_version(opal_pointer_array_t *mca_types,
     bool want_all_types = false;
     bool found;
     mca_base_component_list_item_t *cli;
+    mca_base_failed_component_t *cli_failed;
     int j;
     char *pos;
     opal_info_component_map_t *map;
@@ -1057,6 +1070,15 @@ void opal_info_show_component_version(opal_pointer_array_t *mca_types,
                 }
             }
 
+            /* found it! */
+            OPAL_LIST_FOREACH(cli_failed, map->failed_components, mca_base_failed_component_t) {
+                mca_base_component_repository_item_t *ri = cli_failed->comp;
+                if (want_all_components ||
+                    0 == strcmp(component_name, ri->ri_name) ) {
+                    opal_info_show_failed_component(ri, cli_failed->error_msg);
+                }
+            }
+
             if (!want_all_types) {
                 break;
             }
@@ -1065,6 +1087,30 @@ void opal_info_show_component_version(opal_pointer_array_t *mca_types,
 }
 
 
+static void opal_info_show_failed_component(const mca_base_component_repository_item_t* ri,
+                                            const char *error_msg)
+{
+    char *message, *content;
+
+    if (opal_info_pretty) {
+        asprintf(&message, "MCA %s", ri->ri_type);
+        asprintf(&content, "%s (failed to load) %s", ri->ri_name, error_msg);
+
+        opal_info_out(message, NULL, content);
+
+        free(message);
+        free(content);
+    } else {
+        asprintf(&message, "mca:%s:%s:failed", ri->ri_type, ri->ri_name);
+        asprintf(&content, "%s", error_msg);
+
+        opal_info_out(NULL, message, content);
+
+        free(message);
+        free(content);
+    }
+}
+
 /*
  * Given a component, display its relevant version(s)
  */

opal/runtime/opal_info_support.h

@@ -2,6 +2,7 @@
  * Copyright (c) 2012-2013 Los Alamos National Security, LLC.
  *                         All rights reserved.
 * Copyright (c) 2014 Cisco Systems, Inc.  All rights reserved.
+* Copyright (c) 2017 IBM Corporation.  All rights reserved.
 * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -50,6 +51,7 @@ typedef struct {
     opal_list_item_t super;
     char *type;
     opal_list_t *components;
+    opal_list_t *failed_components;
 } opal_info_component_map_t;
 OPAL_DECLSPEC OBJ_CLASS_DECLARATION(opal_info_component_map_t);