/**
* \brief Blocks the current process until a communication is finished.
* \param L a Lua state
* \return number of values returned to Lua
*
* - Argument 1 (comm): a comm (previously created by isend or irecv)
* - Argument 2 (number, optional): timeout (default is no timeout)
* - Return values (task or nil + string): in case of success, returns the task
* received if you are the receiver and nil if you are the sender. In case of
* failure, returns nil plus an error string.
*/
static int l_comm_wait(lua_State* L) {
msg_comm_t comm = sglua_check_comm(L, 1);
double timeout = -1;
if (lua_gettop(L) >= 2) {
timeout = luaL_checknumber(L, 2);
}
/* comm ... */
msg_error_t res = MSG_comm_wait(comm, timeout);
if (res == MSG_OK) {
msg_task_t task = MSG_comm_get_task(comm);
if (MSG_task_get_sender(task) == MSG_process_self()) {
/* I'm the sender */
return 0;
}
else {
/* I'm the receiver: find the Lua task from the C task */
sglua_task_unregister(L, task);
/* comm ... task */
return 1;
}
}
else {
/* the communication has failed */
lua_pushnil(L);
/* comm ... nil */
lua_pushstring(L, sglua_get_msg_error(res));
/* comm ... nil error */
return 2;
}
}
msg_error_t peer_wait_for_message(peer_t peer)
{
msg_error_t status;
msg_comm_t comm = NULL;
msg_task_t task = NULL;
int idx = -1;
int done = 0;
while (!done) {
comm = MSG_task_irecv(&task, peer->me);
queue_pending_connection(comm, peer->pending_recvs);
if ((idx = MSG_comm_waitany(peer->pending_recvs)) != -1) {
comm = xbt_dynar_get_as(peer->pending_recvs, idx, msg_comm_t);
status = MSG_comm_get_status(comm);
XBT_DEBUG("peer_wait_for_message: error code = %d", status);
xbt_assert(status == MSG_OK, "peer_wait_for_message() failed");
task = MSG_comm_get_task(comm);
MSG_comm_destroy(comm);
xbt_dynar_cursor_rm(peer->pending_recvs, (unsigned int*)&idx);
done = peer_execute_task(peer, task);
task_message_delete(task);
task = NULL;
}
process_pending_connections(peer->pending_sends);
}
return status;
}
/** Receiver function */
int receiver(int argc, char *argv[])
{
int id = -1;
int i;
char mailbox[80];
xbt_dynar_t comms = xbt_dynar_new(sizeof(msg_comm_t), NULL);
int tasks = atof(argv[2]);
msg_task_t *task = xbt_new(msg_task_t, tasks);
_XBT_GNUC_UNUSED int read;
read = sscanf(argv[1], "%d", &id);
xbt_assert(read, "Invalid argument %s\n", argv[1]);
sprintf(mailbox, "receiver-%d", id);
MSG_process_sleep(10);
msg_comm_t res_irecv;
for (i = 0; i < tasks; i++) {
XBT_INFO("Wait to receive task %d", i);
task[i] = NULL;
res_irecv = MSG_task_irecv(&task[i], mailbox);
xbt_dynar_push_as(comms, msg_comm_t, res_irecv);
}
/* Here we are waiting for the receiving of all communications */
msg_task_t task_com;
while (!xbt_dynar_is_empty(comms)) {
_XBT_GNUC_UNUSED msg_error_t err;
xbt_dynar_remove_at(comms, MSG_comm_waitany(comms), &res_irecv);
task_com = MSG_comm_get_task(res_irecv);
MSG_comm_destroy(res_irecv);
XBT_INFO("Processing \"%s\"", MSG_task_get_name(task_com));
MSG_task_execute(task_com);
XBT_INFO("\"%s\" done", MSG_task_get_name(task_com));
err = MSG_task_destroy(task_com);
xbt_assert(err == MSG_OK, "MSG_task_destroy failed");
}
xbt_dynar_free(&comms);
xbt_free(task);
/* Here we tell to sender that all tasks are done */
sprintf(mailbox, "finalize");
res_irecv = MSG_task_isend(MSG_task_create(NULL, 0, 0, NULL), mailbox);
MSG_comm_wait(res_irecv, -1);
MSG_comm_destroy(res_irecv);
XBT_INFO("I'm done. See you!");
return 0;
} /* end_of_receiver */
static int receiver(int argc, char *argv[])
{
xbt_assert(argc==3, "This function expects 2 parameters from the XML deployment file");
int id = xbt_str_parse_int(argv[1], "ID should be numerical, not %s");
int task_amount = xbt_str_parse_int(argv[2], "Invalid amount of tasks: %s");
msg_task_t *tasks = xbt_new(msg_task_t, task_amount);
xbt_dynar_t comms = xbt_dynar_new(sizeof(msg_comm_t), NULL);
char mailbox[80];
snprintf(mailbox,79, "receiver-%d", id);
MSG_process_sleep(10);
for (int i = 0; i < task_amount; i++) {
XBT_INFO("Wait to receive task %d", i);
tasks[i] = NULL;
msg_comm_t comm = MSG_task_irecv(&tasks[i], mailbox);
xbt_dynar_push_as(comms, msg_comm_t, comm);
}
/* Here we are waiting for the receiving of all communications */
while (!xbt_dynar_is_empty(comms)) {
msg_comm_t comm;
// MSG_comm_waitany returns the rank of the comm that just ended. Remove it.
xbt_dynar_remove_at(comms, MSG_comm_waitany(comms), &comm);
msg_task_t task = MSG_comm_get_task(comm);
MSG_comm_destroy(comm);
XBT_INFO("Processing \"%s\"", MSG_task_get_name(task));
MSG_task_execute(task);
XBT_INFO("\"%s\" done", MSG_task_get_name(task));
msg_error_t err = MSG_task_destroy(task);
xbt_assert(err == MSG_OK, "MSG_task_destroy failed");
}
xbt_dynar_free(&comms);
xbt_free(tasks);
/* Here we tell to sender that all tasks are done */
MSG_task_send(MSG_task_create(NULL, 0, 0, NULL), "finalize");
XBT_INFO("I'm done. See you!");
return 0;
}
/**
* @brief Returns whether a communication is finished.
*
* Unlike wait(), This function always returns immediately.
*
* - Argument 1 (comm): a comm (previously created by isend or irecv)
* - Return values (task/boolean or nil + string): if the communication is not
* finished, return false. If the communication is finished and was successful,
* returns the task received if you are the receiver or true if you are the
* sender. If the communication is finished and has failed, returns nil
* plus an error string.
*/
static int l_comm_test(lua_State* L) {
msg_comm_t comm = sglua_check_comm(L, 1);
/* comm ... */
if (!MSG_comm_test(comm)) {
/* not finished yet */
lua_pushboolean(L, 0);
/* comm ... false */
return 1;
}
else {
/* finished but may have failed */
msg_error_t res = MSG_comm_get_status(comm);
if (res == MSG_OK) {
msg_task_t task = MSG_comm_get_task(comm);
if (MSG_task_get_sender(task) == MSG_process_self()) {
/* I'm the sender */
lua_pushboolean(L, 1);
/* comm ... true */
return 1;
}
else {
/* I'm the receiver: find the Lua task from the C task*/
sglua_task_unregister(L, task);
/* comm ... task */
return 1;
}
}
else {
/* the communication has failed */
lua_pushnil(L);
/* comm ... nil */
lua_pushstring(L, sglua_get_msg_error(res));
/* comm ... nil error */
return 2;
}
}
}
void jcomm_bind_task(JNIEnv *env, jobject jcomm) {
msg_comm_t comm = (msg_comm_t) (uintptr_t) env->GetLongField(jcomm, jcomm_field_Comm_bind);
//test if we are receiving or sending a task.
jboolean jreceiving = env->GetBooleanField(jcomm, jcomm_field_Comm_receiving);
if (jreceiving == JNI_TRUE) {
//bind the task object.
msg_task_t task = MSG_comm_get_task(comm);
xbt_assert(task != nullptr, "Task is nullptr");
jobject jtask_global = static_cast<jobject>(MSG_task_get_data(task));
//case where the data has already been retrieved
if (jtask_global == nullptr) {
return;
}
//Make sure the data will be correctly gc.
jobject jtask_local = env->NewLocalRef(jtask_global);
env->DeleteGlobalRef(jtask_global);
env->SetObjectField(jcomm, jtask_field_Comm_task, jtask_local);
MSG_task_set_data(task, nullptr);
}
}
/**
* \brief Node Function
* Arguments:
* - my id
* - the id of a guy I know in the system (except for the first node)
* - the time to sleep before I join (except for the first node)
* - the deadline time
*/
static int node(int argc, char *argv[])
{
double init_time = MSG_get_clock();
msg_task_t task_received = NULL;
int join_success = 0;
double deadline;
xbt_assert(argc == 3 || argc == 5, "Wrong number of arguments for this node");
s_node_t node = {0};
node.id = xbt_str_parse_int(argv[1], "Invalid ID: %s");
node.known_id = -1;
node.ready = -1;
node.pending_tasks = xbt_dynar_new(sizeof(msg_task_t), NULL);
get_mailbox(node.id, node.mailbox);
XBT_DEBUG("New node with id %s (%08x)", node.mailbox, (unsigned)node.id);
for (int i=0; i<LEVELS_COUNT; i++){
int d = domain(node.id, i);
for (int j=0; j<LEVEL_SIZE; j++)
node.routing_table[i][j] = (d==j) ? node.id : -1;
}
for (int i=0; i<NEIGHBORHOOD_SIZE; i++)
node.neighborhood_set[i] = -1;
for (int i=0; i<NAMESPACE_SIZE; i++)
node.namespace_set[i] = -1;
if (argc == 3) { // first ring
XBT_DEBUG("Hey! Let's create the system.");
deadline = xbt_str_parse_double(argv[2], "Invalid deadline: %s");
node.ready = 0;
XBT_DEBUG("Create a new Pastry ring...");
join_success = 1;
} else {
node.known_id = xbt_str_parse_int(argv[2], "Invalid known ID: %s");
double sleep_time = xbt_str_parse_double(argv[3], "Invalid sleep time: %s");
deadline = xbt_str_parse_double(argv[4], "Invalid deadline: %s");
// sleep before starting
XBT_DEBUG("Let's sleep during %f", sleep_time);
MSG_process_sleep(sleep_time);
XBT_DEBUG("Hey! Let's join the system.");
join_success = join(&node);
}
if (join_success) {
XBT_DEBUG("Waiting ….");
while (MSG_get_clock() < init_time + deadline
// && MSG_get_clock() < node.last_change_date + 1000
&& MSG_get_clock() < max_simulation_time) {
if (node.comm_receive == NULL) {
task_received = NULL;
node.comm_receive = MSG_task_irecv(&task_received, node.mailbox);
// FIXME: do not make MSG_task_irecv() calls from several functions
}
if (!MSG_comm_test(node.comm_receive)) {
MSG_process_sleep(5);
} else {
// a transfer has occurred
msg_error_t status = MSG_comm_get_status(node.comm_receive);
if (status != MSG_OK) {
XBT_DEBUG("Failed to receive a task. Nevermind.");
MSG_comm_destroy(node.comm_receive);
node.comm_receive = NULL;
} else {
// the task was successfully received
MSG_comm_destroy(node.comm_receive);
node.comm_receive = NULL;
handle_task(&node, task_received);
}
}
}
//Cleanup the receiving communication.
if (node.comm_receive != NULL) {
if (MSG_comm_test(node.comm_receive) && MSG_comm_get_status(node.comm_receive) == MSG_OK) {
task_free(MSG_comm_get_task(node.comm_receive));
}
MSG_comm_destroy(node.comm_receive);
}
}
xbt_dynar_free(&node.pending_tasks);
return 1;
}
/* Master Process */
int master(int argc, char *argv[])
{
char * key;
struct HdmsgHost *hdmsg_host;
xbt_dict_cursor_t cursor = NULL;
int i = 0;
long remaining_inits = 0;
long remaining_mappers = 0;
long remaining_shufflers = 0;
long remaining_reducers = 0;
long expected_messages = 0;
msg_comm_t res_irecv;
msg_task_t task_com;
msg_task_t *tasks = xbt_new(msg_task_t, number_of_workers);
xbt_dynar_t comms = xbt_dynar_new(sizeof(msg_comm_t), NULL);
XBT_INFO("INITIALIZATION BEGIN");
// Initialize processes (mappers, shufflers, and reducers) on each host
xbt_dict_foreach(hosts, cursor, key, hdmsg_host)
{
if (hdmsg_host->is_worker)
{
MSG_process_create("Init", initializeProcs, NULL, hdmsg_host->host);
tasks[remaining_inits] = NULL;
res_irecv = MSG_task_irecv(&tasks[remaining_inits], "master");
xbt_dynar_push_as(comms, msg_comm_t, res_irecv);
remaining_inits++;
}
}
while (!xbt_dynar_is_empty(comms))
{
xbt_dynar_remove_at(comms, MSG_comm_waitany(comms), &res_irecv);
task_com = MSG_comm_get_task(res_irecv);
if (!strcmp(MSG_task_get_name(task_com), "init_exit"))
{
msg_host_t h = MSG_task_get_source(task_com);
MSG_task_destroy(task_com);
const char *host_name = MSG_host_get_name(h);
struct HdmsgHost *hdmsg_host = xbt_dict_get(hosts, host_name);
remaining_mappers += get_mapper_count(hdmsg_host);
remaining_shufflers += get_shuffler_count(hdmsg_host);
remaining_reducers += get_reducer_count(hdmsg_host);
remaining_inits--;
if (remaining_inits == 0)
{
XBT_INFO("INITIALIZATION COMPLETE");
// Add an extra message to account for the message sent when the shuffle phase begins
expected_messages = 1 + remaining_mappers + remaining_shufflers + remaining_reducers;
free(tasks);
tasks = xbt_new(msg_task_t, expected_messages);
for (i = 0; i < expected_messages; i++)
{
tasks[i] = NULL;
res_irecv = MSG_task_irecv(&tasks[i], "master");
xbt_dynar_push_as(comms, msg_comm_t, res_irecv);
}
XBT_INFO("MAP PHASE BEGIN");
// Activate Mappers
xbt_dict_foreach(hosts, cursor, key, hdmsg_host)
{
activate_mappers(hdmsg_host);
}
}
