Try to failover when we get an async error from the lower layer (BTL)..

This commit was SVN r12420.

ompi/mca/pml/dr/pml_dr.c

@@ -233,5 +233,10 @@ int mca_pml_dr_dump(
 void mca_pml_dr_error_handler(
         struct mca_btl_base_module_t* btl,
         int32_t flags) { 
-    orte_errmgr.abort();
+    /* try failover ! */
+    mca_pml_dr_sendreq_cleanup_active(btl);
+    mca_bml.bml_del_btl(btl);
+    /* orte_errmgr.abort(); */
 }
+
+

ompi/mca/pml/dr/pml_dr_sendreq.c

@@ -1145,3 +1145,18 @@ void mca_pml_dr_send_request_frag_ack(
 }
 
 
+void mca_pml_dr_sendreq_cleanup_active(mca_btl_base_module_t* btl) {  
+   opal_list_item_t* item;
+    
+    for (item = opal_list_get_first(&mca_pml_dr.send_active) ;
+         item != opal_list_get_end(&mca_pml_dr.send_active) ;
+         item = opal_list_get_next(item)) {
+        mca_pml_dr_send_request_t* sendreq = (mca_pml_dr_send_request_t*) item;
+        mca_btl_base_descriptor_t* des = sendreq->req_descriptor;
+        mca_bml_base_btl_t* bml_btl = des->des_context;
+        if( bml_btl && bml_btl->btl == btl) { 
+            des->des_context = NULL;
+        }
+            
+    }
+}

ompi/mca/pml/dr/pml_dr_sendreq.h

@@ -484,6 +484,8 @@ void mca_pml_dr_send_request_frag_ack(
     mca_btl_base_module_t* btl,
     mca_pml_dr_ack_hdr_t*);
 
+void mca_pml_dr_sendreq_cleanup_active(mca_btl_base_module_t* btl);
+
  
 #if defined(c_plusplus) || defined(__cplusplus)
 }

ompi/mca/pml/dr/pml_dr_vfrag.c

@@ -24,7 +24,6 @@
 
 static void mca_pml_dr_vfrag_wdog_timeout(int fd, short event, void* vfrag); 
 static void mca_pml_dr_vfrag_ack_timeout(int fd, short event, void* vfrag);
-static void mca_pml_dr_vfrag_cleanup_active_desc(mca_bml_base_btl_t* bml_btl);
 
 static void mca_pml_dr_vfrag_construct(mca_pml_dr_vfrag_t* vfrag)
 {
@@ -82,7 +81,7 @@ static void mca_pml_dr_vfrag_wdog_timeout(int fd, short event, void* data)
         /* declare btl dead */
         opal_output(0, "%s:%d:%s: failing BTL: %s", __FILE__, __LINE__, __func__, 
                     vfrag->bml_btl->btl->btl_component->btl_version.mca_component_name);
-        mca_pml_dr_vfrag_cleanup_active_desc(vfrag->bml_btl);
+        mca_pml_dr_sendreq_cleanup_active(vfrag->bml_btl->btl);
         mca_bml.bml_del_btl(vfrag->bml_btl->btl);
         mca_pml_dr_vfrag_reset(vfrag);
     } 
@@ -119,7 +118,7 @@ static void mca_pml_dr_vfrag_ack_timeout(int fd, short event, void* data)
         /* declare btl dead */
         opal_output(0, "%s:%d:%s: failing BTL: %s", __FILE__, __LINE__, __func__, 
             vfrag->bml_btl->btl->btl_component->btl_version.mca_component_name);
-        mca_pml_dr_vfrag_cleanup_active_desc(vfrag->bml_btl);
+        mca_pml_dr_sendreq_cleanup_active(vfrag->bml_btl->btl);
         mca_bml.bml_del_btl(vfrag->bml_btl->btl);
         mca_pml_dr_vfrag_reset(vfrag);
     }
@@ -198,18 +197,3 @@ void mca_pml_dr_vfrag_reschedule(mca_pml_dr_vfrag_t* vfrag)
     }
 }
 
-
-static void mca_pml_dr_vfrag_cleanup_active_desc(mca_bml_base_btl_t* bml_btl) {  
-   opal_list_item_t* item;
-    
-    for (item = opal_list_get_first(&mca_pml_dr.send_active) ;
-         item != opal_list_get_end(&mca_pml_dr.send_active) ;
-         item = opal_list_get_next(item)) {
-        mca_pml_dr_send_request_t* sendreq = (mca_pml_dr_send_request_t*) item;
-        mca_btl_base_descriptor_t* des = sendreq->req_descriptor;
-        if( des->des_context == bml_btl) { 
-            des->des_context = NULL;
-        }
-            
-    }
-}