Check if communications are successfull.

[loba.git] / communicator.cpp

diff --git a/communicator.cpp b/communicator.cpp
index da5e2b5f8062755727ed07dcd3617e8a07f0a0fb..1bf16cda43b88e475a6d60b9161fb2bacb12ee59 100644 (file)
--- a/communicator.cpp
+++ b/communicator.cpp
@@ -1,92 +1,225 @@
 #include <algorithm>
 #include <cstring>
 #include <algorithm>
 #include <cstring>

+#include <tr1/functional>
+#include <sstream>

 #include <msg/msg.h>
 #include <xbt/log.h>
 #include <msg/msg.h>
 #include <xbt/log.h>

-#include "communicator.h"
-#include "simgrid_features.h"

 
 

-// XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(simu);
-XBT_LOG_NEW_DEFAULT_SUBCATEGORY(comm, simu,
-                                "Messages from asynchronous pipes");
+XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(comm);

 
 

-namespace {
+#include "misc.h"
+#include "options.h"
+#include "simgrid_features.h"
+#include "tracing.h"

 
 

-    bool comm_test_n_destroy(msg_comm_t& comm)
-    {
-        if (MSG_comm_test(comm)) {
-            MSG_comm_destroy(comm);
-            return true;
-        } else
-            return false;
-    }
+#include "communicator.h"

 
 

+std::string message::to_string()
+{
+    static const char* str[] = { "INFO", "CREDIT", "LOAD",
+                                 "CTRL_CLOSE", "DATA_CLOSE" };
+    std::ostringstream oss;
+    oss << str[type] << ": " << amount;
+    return oss.str();

 }
 
 communicator::communicator()
 }
 
 communicator::communicator()

+    : host(static_cast<hostdata*>(MSG_host_get_data(MSG_host_self())))
+    , ctrl_task(NULL)
+    , ctrl_comm(NULL)
+    , data_task(NULL)
+    , data_comm(NULL)
+    , receiver_mutex(xbt_mutex_init())
+    , receiver_cond(xbt_cond_init())

 {
 {

-    const char* hostname = MSG_host_get_name(MSG_host_self());
-    ctrl_mbox = hostname;
-    ctrl_mbox += "_ctrl";
-    ctrl_task = NULL;
-    ctrl_comm = MSG_task_irecv(&ctrl_task, get_ctrl_mbox());
-
-    data_mbox = hostname;
-    data_mbox += "_data";
-    data_task = NULL;
-    data_comm = MSG_task_irecv(&data_task, get_data_mbox());
+    xbt_mutex_acquire(receiver_mutex);
+    receiver_thread =
+        MSG_process_create("receiver", communicator::receiver_wrapper,
+                           this, MSG_host_self());
+    // wait for the receiver to be ready
+    xbt_cond_wait(receiver_cond, receiver_mutex);
+    xbt_mutex_release(receiver_mutex);

 }
 
 communicator::~communicator()
 {
 }
 
 communicator::~communicator()
 {

-    // fixme: don't know how to free pending communications
-    // (data_comm, ctrl_comm and sent_comm)
+    m_task_t task;

 
 

-    flush_sent();
+    XBT_DEBUG("send finalize to receiver/ctrl");
+    task = MSG_task_create("finalize", 0.0, 0, NULL);
+    MSG_task_send(task, get_ctrl_mbox());
+
+    XBT_DEBUG("send finalize to receiver/data");
+    task = MSG_task_create("finalize", 0.0, 0, NULL);
+    MSG_task_send(task, get_data_mbox());
+
+    xbt_mutex_acquire(receiver_mutex);
+    while (receiver_thread) {
+        XBT_DEBUG("waiting for receiver to terminate");
+        xbt_cond_wait(receiver_cond, receiver_mutex);
+    }
+    xbt_mutex_release(receiver_mutex);
+
+    if (ctrl_comm)
+        XBT_WARN("ctrl_comm is pending!");
+    if (data_comm)
+        XBT_WARN("data_comm is pending!");
+    if (!received.empty())
+        XBT_WARN("lost %zu received message%s!",
+                 received.size(), ESSE(received.size()));

     if (!sent_comm.empty())
     if (!sent_comm.empty())

-        WARN1("Lost %ld send communications!", (long )sent_comm.size());
+        XBT_WARN("lost %zu sent message%s!",
+                 sent_comm.size(), ESSE(sent_comm.size()));
+
+    xbt_cond_destroy(receiver_cond);
+    xbt_mutex_destroy(receiver_mutex);

 }
 
 void communicator::send(const char* dest, message* msg)
 {
 }
 
 void communicator::send(const char* dest, message* msg)
 {

+    XBT_DEBUG("send %s to %s", msg->to_string().c_str(), dest);

     double msg_size = sizeof *msg;
     if (msg->get_type() == message::LOAD)
     double msg_size = sizeof *msg;
     if (msg->get_type() == message::LOAD)

-        msg_size += msg->get_amount();
+        msg_size += opt::comm_cost(msg->get_amount());

     m_task_t task = MSG_task_create("message", 0.0, msg_size, msg);    
     m_task_t task = MSG_task_create("message", 0.0, msg_size, msg);    

-    sent_comm.push_back(MSG_task_isend(task, dest));
-    flush_sent();
+    TRACE_msg_set_task_category(task,
+                                msg->get_type() == message::LOAD ?
+                                TRACE_CAT_DATA : TRACE_CAT_CTRL);
+    msg_comm_t comm = MSG_task_isend(task, dest);
+    sent_comm.push_back(comm);

 }
 
 }
 

-bool communicator::recv(message*& msg, m_host_t& from)
+bool communicator::recv(message*& msg, m_host_t& from, double timeout)

 {
 {

-    msg = NULL;
-
-    if (comm_test_n_destroy(ctrl_comm)) {
-        msg = (message* )MSG_task_get_data(ctrl_task);
-        from = MSG_task_get_source(ctrl_task);
-        MSG_task_destroy(ctrl_task);
-        ctrl_task = NULL;
-        ctrl_comm = MSG_task_irecv(&ctrl_task, get_ctrl_mbox());
-
-    } else if (comm_test_n_destroy(data_comm)) {
-        msg = (message* )MSG_task_get_data(data_task);
-        from = MSG_task_get_source(data_task);
-        MSG_task_destroy(data_task);
-        data_task = NULL;
-        data_comm = MSG_task_irecv(&data_task, get_data_mbox());
+    if (timeout != 0) {
+        volatile double deadline =
+            timeout > 0 ? MSG_get_clock() + timeout : 0.0;
+        xbt_mutex_acquire(receiver_mutex);
+        while (received.empty() && (!deadline || deadline > MSG_get_clock())) {
+            xbt_ex_t e;
+            XBT_DEBUG("waiting for a message to come");
+            TRY {
+                if (deadline)
+                    xbt_cond_timedwait(receiver_cond, receiver_mutex,
+                                       deadline - MSG_get_clock());
+                else
+                    xbt_cond_wait(receiver_cond, receiver_mutex);
+            }
+            CATCH (e) {
+                if (e.category != timeout_error)
+                    RETHROW;
+                xbt_ex_free(e);
+            }
+        }
+        xbt_mutex_release(receiver_mutex);

     }
 
     }
 

-    return msg != NULL;
+    if (received.empty())
+        return false;
+
+    m_task_t task = received.front();
+    received.pop();
+    msg = static_cast<message*>(MSG_task_get_data(task));
+    from = MSG_task_get_source(task);
+    MSG_task_destroy(task);
+
+    XBT_DEBUG("received %s from %s",
+           msg->to_string().c_str(), MSG_host_get_name(from));
+
+    return true;
+}
+
+void communicator::flush(bool wait)
+{
+    sent_comm.remove_if(comm_test_n_destroy);
+    if (wait && !sent_comm.empty()) {
+        msg_comm_t comms[sent_comm.size()];
+        std::copy(sent_comm.begin(), sent_comm.end(), comms);
+        MSG_comm_waitall(comms, sent_comm.size(), -1.0);
+        if (!MSG_WAIT_DESTROYS_COMMS)
+            std::for_each(sent_comm.begin(), sent_comm.end(),
+                          comm_check_n_destroy);
+        sent_comm.clear();
+    }
+}
+
+int communicator::receiver_wrapper(int, char* [])
+{
+    communicator* comm;
+    comm = static_cast<communicator*>(MSG_process_get_data(MSG_process_self()));
+    comm->receiver();
+
+    XBT_DEBUG("terminate");
+    xbt_mutex_acquire(comm->receiver_mutex);
+    comm->receiver_thread = NULL;
+    xbt_cond_signal(comm->receiver_cond);
+    xbt_mutex_release(comm->receiver_mutex);
+
+    return 0;
+}
+
+void communicator::receiver1(msg_comm_t& comm, m_task_t& task, const char* mbox)
+{
+    comm_check_n_destroy(comm);
+    if (strcmp(MSG_task_get_name(task), "finalize")) {
+        XBT_DEBUG("received message on %s", mbox);
+        xbt_mutex_acquire(receiver_mutex);
+        received.push(task);
+        xbt_cond_signal(receiver_cond);
+        xbt_mutex_release(receiver_mutex);
+        task = NULL;
+        comm = MSG_task_irecv(&task, mbox);
+    } else {
+        XBT_DEBUG("received finalize on %s", mbox);
+        MSG_task_destroy(task);
+        task = NULL;
+        comm = NULL;
+    }
+}
+
+void communicator::receiver()
+{
+    ctrl_comm = MSG_task_irecv(&ctrl_task, get_ctrl_mbox());
+    data_comm = MSG_task_irecv(&data_task, get_data_mbox());
+    XBT_DEBUG("receiver ready");
+    xbt_mutex_acquire(receiver_mutex);
+    xbt_cond_signal(receiver_cond); // signal master that we are ready
+    xbt_mutex_release(receiver_mutex);
+
+    xbt_dynar_t comms = xbt_dynar_new(sizeof(msg_comm_t), NULL);
+    while (ctrl_comm || data_comm) {
+
+        if (ctrl_comm)
+            xbt_dynar_push(comms, &ctrl_comm);
+        if (data_comm)
+            xbt_dynar_push(comms, &data_comm);
+        int index = MSG_comm_waitany(comms);
+        msg_comm_t finished_comm = xbt_dynar_get_as(comms, index, msg_comm_t);
+        xbt_dynar_reset(comms);
+
+        if (finished_comm == ctrl_comm)
+            receiver1(ctrl_comm, ctrl_task, get_ctrl_mbox());
+        else if (finished_comm == data_comm)
+            receiver1(data_comm, data_task, get_data_mbox());
+        else
+            THROW1(0, 0, "Cannot handle unknown comm -- %p", finished_comm);
+    }
+    xbt_dynar_free(&comms);

 }
 
 }
 

-int communicator::send_backlog()
+void communicator::comm_check_n_destroy(msg_comm_t comm)

 {
 {

-    flush_sent();
-    return sent_comm.size();
+    xbt_assert(MSG_comm_get_status(comm) == MSG_OK);
+    MSG_comm_destroy(comm);

 }
 
 }
 

-void communicator::flush_sent()
+bool communicator::comm_test_n_destroy(msg_comm_t comm)

 {
 {

-    std::remove_if(sent_comm.begin(), sent_comm.end(), comm_test_n_destroy);
+    if (MSG_comm_test(comm)) {
+        comm_check_n_destroy(comm);
+        return true;
+    } else
+        return false;

 }
 
 // Local variables:
 }
 
 // Local variables: