Fix a bug occuring when a host involved in a communication fails

[simgrid.git] / src / simix / smx_network.c

diff --git a/src/simix/smx_network.c b/src/simix/smx_network.c
index 91eef6b..4987f8f 100644 (file)
--- a/src/simix/smx_network.c
+++ b/src/simix/smx_network.c
@@ -1,7 +1,5 @@
-/*     $Id$     */
-
-/* Copyright (c) 2009 Cristian Rosa.
-   All rights reserved.                                          */
+/* Copyright (c) 2009, 2010. The SimGrid Team.
+ * All rights reserved.                                                     */
 
 /* This program is free software; you can redistribute it and/or modify it
  * under the terms of the license (GNU LGPL) which comes with this package. */
@@ -220,7 +218,7 @@ static inline void SIMIX_communication_start(smx_comm_t comm)
        it will be restarted when the sender process resume */
     if(SIMIX_process_is_suspended(comm->src_proc) || 
        SIMIX_process_is_suspended(comm->dst_proc)) {
-      SIMIX_action_set_priority(comm->act, 0);
+      SIMIX_action_suspend(comm->act);
     }
     
     /* Add the communication as user data of the action */
@@ -239,9 +237,9 @@ static inline void SIMIX_communication_start(smx_comm_t comm)
  *  \param timeout The max amount of time to wait for the communication to finish
  *
  *  Throws:
- *   - host_error if peer failed
- *   - timeout_error if communication reached the timeout specified
- *   - network_error if network failed or peer issued a timeout
+ *   - host_error if local peer failed
+ *   - timeout_error if communication reached the timeout specified (either because of local peer or remote peer)
+ *   - network_error if network failed or remote peer failed
  */
 static inline void SIMIX_communication_wait_for_completion(smx_comm_t comm, double timeout)
 {
@@ -292,7 +290,7 @@ static inline void SIMIX_communication_wait_for_completion(smx_comm_t comm, doub
   SIMIX_sem_release_forever(comm->sem);
   
   /* Check for errors other than timeouts (they are catched above) */
-  if(!SIMIX_host_get_state(SIMIX_host_self())){
+  if (!SIMIX_host_get_state(SIMIX_host_self())){
     if(comm->rdv)
       SIMIX_rdv_remove(comm->rdv, comm);
     SIMIX_communication_destroy(comm);
@@ -300,6 +298,16 @@ static inline void SIMIX_communication_wait_for_completion(smx_comm_t comm, doub
   } else if (SIMIX_action_get_state(comm->act) == SURF_ACTION_FAILED){
     SIMIX_communication_destroy(comm);
     THROW0(network_error, 0, "Link failure");
+  } else if (!SIMIX_host_get_state(SIMIX_process_get_host(comm->dst_proc)) ||
+      !SIMIX_host_get_state(SIMIX_process_get_host(comm->src_proc))) {
+    /* We test both src&dst because we dunno who we are today, and we already tested myself above.
+     *    So, at the end, we test the remote peer only
+     * Moreover, we have to test it because if the remote peer fails, the action comm->act is not done nor failed.
+     *    In that case, we got awaken by the little endless actions created in the SIMIX_sem_acquire(comm->sem)
+     *    at the beginning of this function. */
+    SIMIX_communication_destroy(comm);
+    THROW0(network_error, 0, "Remote peer failed");
+
   }
   SIMIX_communication_destroy(comm);
 }
@@ -347,12 +355,19 @@ void SIMIX_network_copy_buffer_callback(smx_comm_t comm, size_t buff_size) {
  */
 void SIMIX_network_copy_data(smx_comm_t comm)
 {
+  size_t buff_size = comm->src_buff_size;
+
+  DEBUG6("Copying comm %p data from %s (%p) -> %s (%p) (%zu bytes)",
+      comm,
+      comm->src_proc->smx_host->name, comm->src_buff,
+      comm->dst_proc->smx_host->name, comm->dst_buff,
+      buff_size);
+
   /* If there is no data to be copy then return */
   if(!comm->src_buff || !comm->dst_buff)
     return;
   
   /* Copy at most dst_buff_size bytes of the message to receiver's buffer */
-  size_t buff_size = comm->src_buff_size;
   if (comm->dst_buff_size)
     buff_size = MIN(buff_size,*(comm->dst_buff_size));
   
@@ -362,11 +377,6 @@ void SIMIX_network_copy_data(smx_comm_t comm)
 
   if(buff_size == 0)
     return;
-  DEBUG6("Copying comm %p data from %s (%p) -> %s (%p) (%zu bytes)",
-      comm,
-      comm->src_proc->smx_host->name, comm->src_buff,
-      comm->dst_proc->smx_host->name, comm->dst_buff,
-      buff_size);
   (*SIMIX_network_copy_data_callback)(comm, buff_size);
 
   /* pimple to display the message sizes */
@@ -384,12 +394,11 @@ void SIMIX_network_copy_data(smx_comm_t comm)
 /* pimple to display the message sizes */
 void SIMIX_message_sizes_output(const char *filename) {
   FILE * out = fopen(filename,"w");
-  INFO1("Output message sizes to %s",filename);
   xbt_assert1(out,"Cannot open file %s",filename);
   uintptr_t key,data;
   xbt_dict_cursor_t cursor;
   xbt_dict_foreach(msg_sizes,cursor,key,data) {
-    fprintf(out,"%ld %ld\n",key,data);
+    fprintf(out,"%zu %zu\n",key,data);
   }
   fclose(out);
 }