A few changes to the FTB notifier interface:

- add an orte ftb notifier help file for more verbose error messages
- check if we can connect to the FTB during component->query and close
  the component, if we cannot.
- make the ftb component interface methods static.
- add mca parameters to set override the default subscription style and
  priority.

This commit was SVN r22011.

orte/mca/notifier/ftb/help-orte-notifier-ftb.txt

@@ -0,0 +1,27 @@
+# -*- text -*-
+#
+# Copyright (c) 2009 The Trustees of Indiana University and Indiana
+#                    University Research and Technology Corporation.
+#                    All rights reserved.
+# $COPYRIGHT$
+# 
+# Additional copyrights may follow
+# 
+# $HEADER$
+#
+# This is the US/English help file for Open MPI's FTB notifier support
+#
+[invalid subscription style]
+Error: the Open MPI FTB component tried to register with an invalid
+FTB client subscription style.
+
+ Subscription style: %s
+#
+[invalid value]
+Error: the Open MPI FTB notifier component tried to register with an
+invalid value in the FTB client information.
+#
+[unable to connect]
+Open MPI's FTB notifier component was unable to establish a connection
+with the FTB backplane.
+#

orte/mca/notifier/ftb/notifier_ftb.h

@@ -28,21 +28,30 @@
 
 BEGIN_C_DECLS
 
-/*
+typedef struct {
- * Component open / close
+    orte_notifier_base_component_t super;
- */
-int orte_notifier_ftb_open(void);
-int orte_notifier_ftb_close(void);
-int orte_notifier_ftb_component_query(mca_base_module_t **module, int *priority);
 
+    /* FTB client subscription style */
+    char *subscription_style;
+
+    /* Priority of this component */
+    int priority;
+} orte_notifier_ftb_component_t;
 
 /*
  * Notifier interfaces
  */
 
-ORTE_MODULE_DECLSPEC extern orte_notifier_base_component_t mca_notifier_ftb_component;
+ORTE_MODULE_DECLSPEC extern orte_notifier_ftb_component_t mca_notifier_ftb_component;
 extern orte_notifier_base_module_t orte_notifier_ftb_module;
 
+/*
+ * FTB client information
+ */
+
+extern FTB_client_t ftb_client_info;
+extern FTB_client_handle_t ftb_client_handle;
+
 END_C_DECLS
 
 #endif

orte/mca/notifier/ftb/notifier_ftb_component.c

@@ -18,6 +18,10 @@
 */
 /** @file:
 *
+* This component proxies notification events to the Fault Tolerant
+* Backplane (See http://www.mcs.anl.gov/research/cifts/).
+* The ORTE notifier severity is translated to the corresponding
+* FTB severity before the event is published to the FTB.  
 */
 
 /*
@@ -26,44 +30,133 @@
 #include "orte_config.h"
 #include "orte/constants.h"
 
+#include <string.h>
+
+#include "orte/util/show_help.h"
+#include "orte/runtime/orte_globals.h"
+#include "opal/mca/base/mca_base_param.h"
 #include "notifier_ftb.h"
 
+static int orte_notifier_ftb_close(void);
+static int orte_notifier_ftb_component_query(mca_base_module_t **module, int *priority);
+static int orte_notifier_ftb_register(void);
+
 /*
  * Struct of function pointers that need to be initialized
  */
-orte_notifier_base_component_t mca_notifier_ftb_component = {
+orte_notifier_ftb_component_t mca_notifier_ftb_component = {
     {
-        ORTE_NOTIFIER_BASE_VERSION_1_0_0,
+        {
-        
+            ORTE_NOTIFIER_BASE_VERSION_1_0_0,
-        "ftb", /* MCA module name */
+
-        ORTE_MAJOR_VERSION,  /* MCA module major version */
+            "ftb", /* MCA module name */
-        ORTE_MINOR_VERSION,  /* MCA module minor version */
+            ORTE_MAJOR_VERSION,  /* MCA module major version */
-        ORTE_RELEASE_VERSION,  /* MCA module release version */
+            ORTE_MINOR_VERSION,  /* MCA module minor version */
-        orte_notifier_ftb_open,  /* module open */
+            ORTE_RELEASE_VERSION,  /* MCA module release version */
-        orte_notifier_ftb_close, /* module close */
+
-        orte_notifier_ftb_component_query /* module query */
+            NULL,
+            orte_notifier_ftb_close, /* module close */
+            orte_notifier_ftb_component_query, /* module query */
+            orte_notifier_ftb_register, /* module register */
+        },
+        {
+            /* The component is checkpoint ready */
+            MCA_BASE_METADATA_PARAM_CHECKPOINT
+        }
     },
-    {
+
-        /* The component is checkpoint ready */
+    /* FTB client subscription style */
-        MCA_BASE_METADATA_PARAM_CHECKPOINT
+    "FTB_SUBSCRIPTION_NONE",
-    }
+
+    /* Priority */
+    10,
 };
 
-/* Open the component */
+static int orte_notifier_ftb_close(void)
-int orte_notifier_ftb_open(void)
 {
+
+    if (NULL != mca_notifier_ftb_component.subscription_style) {
+        free(mca_notifier_ftb_component.subscription_style);
+    }
+
+    /* If the FTB client handle is valid, disconnect the client */
+    if (1 == ftb_client_handle.valid) {
+        FTB_Disconnect(ftb_client_handle);
+    }
+
     return ORTE_SUCCESS;
 }
 
-int orte_notifier_ftb_close(void)
+static int orte_notifier_ftb_component_query(mca_base_module_t **module, int *priority)
 {
-    return ORTE_SUCCESS;
+    int ret;
-}
+    *priority = 0;
+    *module = NULL;
 
-int orte_notifier_ftb_component_query(mca_base_module_t **module, int *priority)
+    /* Fill the FTB client information structure */
-{
+    memset(&ftb_client_info, 0, sizeof(ftb_client_info));
-    /* we are a lower-level default, so set a low priority so we can be overridden */
+    strcpy(ftb_client_info.event_space, "ftb.mpi.openmpi");
-    *priority = 1;
+
+    /* We represent each client with a client name of the form
+       openmpi/<hostname>/<PID> as a unique identifier in the
+       FTB client namespace */
+    sprintf(ftb_client_info.client_name, "openmpi/%s/%u",
+            orte_process_info.nodename, orte_process_info.pid);
+
+    sprintf(ftb_client_info.client_jobid, "%u", ORTE_PROC_MY_NAME->jobid);
+
+    strncpy(ftb_client_info.client_subscription_style,
+            mca_notifier_ftb_component.subscription_style,
+            strlen(mca_notifier_ftb_component.subscription_style));
+
+    /* We try to connect to the FTB backplane now, and we abort
+       if we cannot connect for some reason. */
+    if (FTB_SUCCESS != (ret = FTB_Connect(&ftb_client_info, &ftb_client_handle))) {
+        switch (ret) {
+        case FTB_ERR_SUBSCRIPTION_STYLE:
+            orte_show_help("help-orte-notifier-ftb.txt",
+                           "invalid subscription style",
+                           true, ftb_client_info.client_subscription_style);
+
+        case FTB_ERR_INVALID_VALUE:
+            orte_show_help("help-orte-notifier-ftb.txt",
+                           "invalid value",
+                           true);
+
+        default:
+            orte_show_help("help-orte-notifier-ftb.txt",
+                           "unable to connect",
+                           true);
+        }
+        
+        return ORTE_ERR_NOT_FOUND;
+    }
+
+    *priority = 10;
     *module = (mca_base_module_t *)&orte_notifier_ftb_module;
-    return ORTE_SUCCESS;    
+
+    return ORTE_SUCCESS;
+}
+
+static int orte_notifier_ftb_register(void)
+{
+
+    /* FTB client subscription style */
+    mca_base_param_reg_string(&mca_notifier_ftb_component.super.base_version,
+                              "subscription_style",
+                              "FTB client subscription style. "
+                              "Possible values are none, polling, notify and both (polling and notify).",
+                              false, false,
+                              mca_notifier_ftb_component.subscription_style,
+                              &mca_notifier_ftb_component.subscription_style);
+
+    /* Priority */
+    mca_base_param_reg_int(&mca_notifier_ftb_component.super.base_version,
+                           "priority",
+                           "Priority of this component",
+                           false, false, 
+                           mca_notifier_ftb_component.priority,
+                           &mca_notifier_ftb_component.priority);
+
+    return ORTE_SUCCESS;
 }

orte/mca/notifier/ftb/notifier_ftb_module.c

@@ -51,14 +51,11 @@ orte_notifier_base_module_t orte_notifier_ftb_module = {
     mypeerlog
 };
 
-/* Module "global" variables */
+/* FTB client information */
-static FTB_client_t cinfo = {
+FTB_client_t ftb_client_info;
-    .event_space = "ftb.mpi.openmpi",
+
-    .client_name = "",
+/* FTB client handle */
-    .client_jobid = "",
+FTB_client_handle_t ftb_client_handle;
-    .client_subscription_style = "FTB_SUBSCRIPTION_NONE"
-};
-static FTB_client_handle_t chandle;
 
 static FTB_event_info_t ftb_event_info[] = {
 /* 0 */    {"UNKNOWN_ERROR",	"error"},
@@ -102,23 +99,11 @@ static int orte_err2ftb(int errnum)
 static int init(void) {
     int ret;
 
-/*    snprintf(cinfo.client_name, FTB_MAX_CLIENT_NAME, "%s", argv[0]); 
+    ret = FTB_Declare_publishable_events(ftb_client_handle, 0, ftb_event_info, ftb_event_info_count);
- * How to obtain argv[0] at this point?  I don't know...
- * similarly, how do we obtain client_jobid?
- *    snprintf(cinfo.client_jobid, FTB_MAX_CLIENT_JOBID, "%s", orte_jobid???); 
- */
-    
-    if (FTB_SUCCESS != (ret = FTB_Connect(&cinfo, &chandle))) {
-        opal_output(orte_notifier_base_output,
-            "notifier:ftb:init FTB_Connect failed ret=%d\n", ret);
-        return ORTE_ERROR;
-    }
-
-    ret = FTB_Declare_publishable_events(chandle, 0, ftb_event_info, ftb_event_info_count);
     if (FTB_SUCCESS != ret) {
         opal_output(orte_notifier_base_output,
             "notifier:ftb:init FTB_Declare_publishable_events failed ret=%d\n", ret);
-        FTB_Disconnect(chandle);
+        FTB_Disconnect(ftb_client_handle);
         return ORTE_ERROR;
     }
 
@@ -126,7 +111,7 @@ static int init(void) {
 }
 
 static void finalize(void) {
-    FTB_Disconnect(chandle);
+    FTB_Disconnect(ftb_client_handle);
 }
 
 static void convert2ftb(int errcode, char *payload)
@@ -138,7 +123,7 @@ static void convert2ftb(int errcode, char *payload)
     snprintf(eprop.event_payload, FTB_MAX_PAYLOAD_DATA, "%s", (payload != NULL) ? payload : "");
 
     event_id = orte_err2ftb(errcode);
-    ret = FTB_Publish(chandle, ftb_event_info[event_id].event_name, &eprop, &ehandle);
+    ret = FTB_Publish(ftb_client_handle, ftb_event_info[event_id].event_name, &eprop, &ehandle);
     if (FTB_SUCCESS != ret) {
         opal_output(orte_notifier_base_output,
             "notifier:ftb:convert2ftb(%d,'%s') FTB_Publish failed ret=%d\n", errcode, eprop.event_payload, ret);