From cb6c404ed6a176ee440c09957a8fd9f34918a29b Mon Sep 17 00:00:00 2001
From: Samuel Lepetit <samuel.lepetit@inria.fr>
Date: Wed, 20 Jun 2012 14:01:52 +0200
Subject: [PATCH] Fix issue when a host failed while executing a task (the
 glass is only half full however, we should do the process kill in simix)

---
 src/msg/msg_gos.c     | 52 +++++++++++++++++++++++++------------------
 src/msg/msg_mailbox.c |  3 ---
 src/simix/smx_host.c  |  8 +------
 3 files changed, 31 insertions(+), 32 deletions(-)

diff --git a/src/msg/msg_gos.c b/src/msg/msg_gos.c
index 0ce5b3ef84..957fcb8c45 100644
--- a/src/msg/msg_gos.c
+++ b/src/msg/msg_gos.c
@@ -23,6 +23,7 @@ XBT_LOG_NEW_DEFAULT_SUBCATEGORY(msg_gos, msg,
  */
 MSG_error_t MSG_task_execute(m_task_t task)
 {
+  xbt_ex_t e;
   simdata_task_t simdata = NULL;
   simdata_process_t p_simdata;
   e_smx_state_t comp_state;
@@ -59,15 +60,16 @@ MSG_error_t MSG_task_execute(m_task_t task)
 #ifdef HAVE_TRACING
   simcall_set_category(simdata->compute, task->category);
 #endif
-
+  
   p_simdata->waiting_action = simdata->compute;
-  comp_state = simcall_host_execution_wait(simdata->compute);
-  p_simdata->waiting_action = NULL;
+  TRY {
+    comp_state = simcall_host_execution_wait(simdata->compute);
+    p_simdata->waiting_action = NULL;
 
-  simdata->isused=0;
+    simdata->isused=0;
+
+    XBT_DEBUG("Execution task '%s' finished in state %d", task->name, (int)comp_state);
 
-  XBT_DEBUG("Execution task '%s' finished in state %d", task->name, (int)comp_state);
-  if (comp_state == SIMIX_DONE) {
     /* action ended, set comm and compute = NULL, the actions is already destroyed in the main function */
     simdata->computation_amount = 0.0;
     simdata->comm = NULL;
@@ -76,22 +78,28 @@ MSG_error_t MSG_task_execute(m_task_t task)
     TRACE_msg_task_execute_end(task);
 #endif
     MSG_RETURN(MSG_OK);
-  } else if (simcall_host_get_state(SIMIX_host_self()) == 0) {
-    /* action ended, set comm and compute = NULL, the actions is already destroyed in the main function */
-    simdata->comm = NULL;
-    simdata->compute = NULL;
-#ifdef HAVE_TRACING
-    TRACE_msg_task_execute_end(task);
-#endif
-    MSG_RETURN(MSG_HOST_FAILURE);
-  } else {
-    /* action ended, set comm and compute = NULL, the actions is already destroyed in the main function */
-    simdata->comm = NULL;
-    simdata->compute = NULL;
-#ifdef HAVE_TRACING
-    TRACE_msg_task_execute_end(task);
-#endif
-    MSG_RETURN(MSG_TASK_CANCELED);
+  }  
+  CATCH(e) {
+    switch (e.category) {
+      case host_error:
+        /* action ended, set comm and compute = NULL, the actions is already  destroyed in the main function */
+        simdata->comm = NULL;
+        simdata->compute = NULL;
+        #ifdef HAVE_TRACING
+          TRACE_msg_task_execute_end(task);
+        #endif
+        MSG_RETURN(MSG_HOST_FAILURE);
+        break;  
+      case cancel_error:
+        /* action ended, set comm and compute = NULL, the actions is already destroyed in the main function */
+        simdata->comm = NULL;
+        simdata->compute = NULL;
+    #ifdef HAVE_TRACING
+        TRACE_msg_task_execute_end(task);
+    #endif
+        MSG_RETURN(MSG_TASK_CANCELED);        
+      break;
+    }
   }
 }
 
diff --git a/src/msg/msg_mailbox.c b/src/msg/msg_mailbox.c
index 5fd54c3b43..997b138659 100644
--- a/src/msg/msg_mailbox.c
+++ b/src/msg/msg_mailbox.c
@@ -156,9 +156,6 @@ MSG_mailbox_put_with_timeout(msg_mailbox_t mailbox, m_task_t task,
 
   CATCH(e) {
     switch (e.category) {
-    case host_error:
-      ret = MSG_HOST_FAILURE;
-      break;
     case network_error:
       ret = MSG_TRANSFER_FAILURE;
       break;
diff --git a/src/simix/smx_host.c b/src/simix/smx_host.c
index 299864a5f9..815d947a5d 100644
--- a/src/simix/smx_host.c
+++ b/src/simix/smx_host.c
@@ -348,13 +348,7 @@ void SIMIX_execution_finish(smx_action_t action)
 
       case SIMIX_FAILED:
         XBT_DEBUG("SIMIX_execution_finished: host '%s' failed", simcall->issuer->smx_host->name);
-        if (simcall->issuer->smx_host == action->execution.host) {
-          // add a reference to the action that will be destroyed when the killed process is cleaned up, and by the end of the current function
-          surf_action_ref(action->execution.surf_exec);
-          SIMIX_process_kill(simcall->issuer);
-        } else {
-          SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
-        }
+        SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
         break;
 
       case SIMIX_CANCELED:
-- 
2.20.1