void EventLoop::Impl::processTasks() { TaskQueue tq; std::unique_lock<LockType> ul(task_mutex_); task_queue_.swap(tq); while (auto node = tq.front_node()) { tq.pop_front(); auto &task_slot = node->element(); task_slot.state = TaskSlot::State::RUNNING; ul.unlock(); {// execute the task LockGuard g(task_run_mutex_); if (task_slot.state != TaskSlot::State::INACTIVE) { task_slot(); task_slot.state = TaskSlot::State::INACTIVE; } } ul.lock(); if (task_slot.token) { task_slot.token->removeTaskNode(node); } } }
// pack the jobs into multiple packages - 2000 jobs per package // and insert it into the ready queue of server that requested to steal tasks //int Worker::migrateTasks(int num_tasks, ZHTClient &clientRet, int index){ void* migrateTasks(void *args) { Worker *worker = (Worker*) args; int index; while (ON) { //while(migrateq.size() > 0) { while (migratev.any()) { pthread_mutex_lock(&mq_lock); if (migratev.any()) { //int *index = (int*)args; //index = migrateq.front(); index = migratev.pop(); //migrateq.pop(); //cout << "1 worker = " << worker->selfIndex << " to index = " << index << " size = " << rqueue.size() << endl; } else { //cout << "migratev count = " << migratev.count() << endl; pthread_mutex_unlock(&mq_lock); continue; } if (index < 0 || index >= worker->num_nodes) { //cout << "bad index: worker = " << worker->selfIndex << " to index = " << index << endl; pthread_mutex_unlock(&mq_lock); continue; } pthread_mutex_unlock(&mq_lock); //cout << "2 worker = " << worker->selfIndex << " to index = " << index << " size = " << rqueue.size() << endl; pthread_mutex_lock(&m_lock); pthread_mutex_lock(&lock); int32_t num_tasks = rqueue.size() / 2; if (num_tasks < 1) { pthread_mutex_unlock(&lock); pthread_mutex_unlock(&m_lock); continue;; } try { //cout << "going to send " << num_tasks << " tasks" << endl; mqueue.assign(rqueue.end() - num_tasks, rqueue.end()); rqueue.erase(rqueue.end() - num_tasks, rqueue.end()); //cout << "rqueue size = " << rqueue.size() << " mqueue size = " << mqueue.size() << endl; } catch (...) { cout << "migrateTasks: cannot allocate memory while copying tasks to migrate queue" << endl; pthread_exit(NULL); } pthread_mutex_unlock(&lock); map<uint32_t, NodeList> update_map = worker->get_map(mqueue); int update_ret = worker->zht_update(update_map, "nodehistory", index); /*if(index == worker->selfIndex) { cout << "ALERT: MIGRATING TO ITSELF" << endl; }*/ int num_packages = 0; long total_submitted = 0; num_tasks = mqueue.size(); while (total_submitted != num_tasks) { Package package; string alltasks; package.set_virtualpath(worker->ip); package.set_operation(22); num_packages++; int num_tasks_this_package = max_tasks_per_package; int num_tasks_left = num_tasks - total_submitted; if (num_tasks_left < max_tasks_per_package) { num_tasks_this_package = num_tasks_left; } for (int j = 0; j < num_tasks_this_package; j++) { //TaskQueue_item* qi = migrate_queue->remove_element(); if (mqueue.size() < 1) { if (j > 0) { total_submitted = total_submitted + j; package.set_realfullpath(alltasks); string str = package.SerializeAsString(); pthread_mutex_lock(&msg_lock); int32_t ret = worker->svrclient.svrtosvr(str, str.length(), index); pthread_mutex_unlock(&msg_lock); } //pthread_mutex_unlock (&m_lock); //return total_submitted; //pthread_exit(NULL); total_submitted = num_tasks; break; } try { alltasks.append(mqueue.front()->task_id); alltasks.append("\'\""); // Task ID /*stringstream num_moves_ss; num_moves_ss << (mqueue.front()->num_moves + 1); alltasks.append(num_moves_ss.str()); alltasks.append("\'\""); // Number of moves*/ if (LOGGING) { migrate_fp << " taskid = " << mqueue.front()->task_id; //migrate_fp << " num moves = " << (mqueue.front()->num_moves + 1); } delete mqueue.front(); mqueue.pop_front(); } catch (...) { cout << "migrateTasks: Exception occurred while processing mqueue" << endl; } } if (total_submitted == num_tasks) { break; } total_submitted = total_submitted + num_tasks_this_package; package.set_realfullpath(alltasks); string str = package.SerializeAsString(); //cout << "r1: " << total_submitted << " tasks" << endl; pthread_mutex_lock(&msg_lock); int32_t ret = worker->svrclient.svrtosvr(str, str.length(), index); //cout << "r1 sent" << endl; pthread_mutex_unlock(&msg_lock); } pthread_mutex_unlock(&m_lock); //cout << "matrix_server: No. of tasks sent = " << total_submitted << endl; } } }
// thread to monitor ready queue and execute tasks based on num of cores availability void* check_ready_queue(void* args) { Worker *worker = (Worker*) args; TaskQueue_Item *qi; while (ON) { while (rqueue.size() > 0) { pthread_mutex_lock(&lock); if (rqueue.size() > 0) { qi = rqueue.front(); rqueue.pop_front(); } else { pthread_mutex_unlock(&lock); continue; } pthread_mutex_unlock(&lock); pthread_mutex_lock(&mutex_idle); worker->num_idle_cores--; pthread_mutex_unlock(&mutex_idle); if (!work_exec_flag) { work_exec_flag = 1; FILE *fp = fopen(file_worker_start.c_str(), "a+"); if (fp != NULL) { //fputs("fopen example", fp); char fbuf[100]; memset(fbuf, 0, sizeof(fbuf)); sprintf(fbuf, "%s:%d Got jobs..Started excuting\n", worker->ip.c_str(), worker->selfIndex); fwrite(fbuf, sizeof(char), strlen(fbuf), fp); fflush(fp); fclose(fp); } /*worker_start << worker->ip << ":" << worker->selfIndex << " Got jobs..Started excuting" << endl;*/ } //cout << "task to lookup = " << qi->task_id << endl; string value = worker->zht_lookup(qi->task_id); Package recv_pkg; recv_pkg.ParseFromString(value); //cout << "check_ready_queue: task " << qi->task_id << " node history = " << recv_pkg.nodehistory() << endl; int num_vector_count, per_vector_count; vector<vector<string> > tokenize_string = tokenize( recv_pkg.realfullpath(), '\"', '\'', num_vector_count, per_vector_count); //cout << "worker " << worker->selfIndex<< " pertask processing done" << endl; /*cout << "task = " << qi->task_id << " notify list: "; for(int l = 0; l < tokenize_string.at(1).size(); l++) { cout << tokenize_string.at(1).at(l) << " "; } cout << endl;*/ stringstream duration_ss; try { duration_ss << tokenize_string.at(0).at(1); } catch (exception& e) { cout << "void* check_ready_queue: num_vector_count = " << num_vector_count << " per_vector_count = " << per_vector_count << endl; cout << "void* check_ready_queue: (tokenize_string.at(0).at(1)) " << " " << e.what() << endl; cout << "void* check_ready_queue: value = " << value << endl; exit(1); } double duration_tmp; //duration = 1000000; duration_ss >> duration_tmp; long duration = floor(duration_tmp * 1000 + 0.5); //duration = duration / duration - 1; string client_id; try { client_id = tokenize_string.at(0).at(2); } catch (exception& e) { cout << "void* check_ready_queue: num_vector_count = " << num_vector_count << " per_vector_count = " << per_vector_count << endl; cout << "void* check_ready_queue: (tokenize_string.at(0).at(2)) " << " " << e.what() << endl; exit(1); } uint64_t sub_time; try { stringstream sub_time_ss; sub_time_ss << tokenize_string.at(0).at(3); sub_time_ss >> sub_time; } catch (exception& e) { cout << "void* check_ready_queue: num_vector_count = " << num_vector_count << " per_vector_count = " << per_vector_count << endl; cout << "void* check_ready_queue: (tokenize_string.at(0).at(3)) " << " " << e.what() << endl; exit(1); } timespec task_start_time, task_end_time; clock_gettime(CLOCK_REALTIME, &task_start_time); //uint32_t exit_code = sleep(duration); //cout << "The duration of this task is:" << duration << endl; uint32_t exit_code = usleep(duration); clock_gettime(CLOCK_REALTIME, &task_end_time); // push completed task into complete queue pthread_mutex_lock(&c_lock); cqueue.push_back(make_pair(qi->task_id, tokenize_string.at(1))); pthread_mutex_unlock(&c_lock); // append completed task uint64_t st = (uint64_t) task_start_time.tv_sec * 1000000000 + (uint64_t) task_start_time.tv_nsec; uint64_t et = (uint64_t) task_end_time.tv_sec * 1000000000 + (uint64_t) task_end_time.tv_nsec; timespec diff = timediff(task_start_time, task_end_time); pthread_mutex_lock(&mutex_idle); worker->num_idle_cores++; task_comp_count++; pthread_mutex_unlock(&mutex_idle); if (LOGGING) { string fin_str; stringstream out; out << qi->task_id << "+" << client_id << " exitcode " << " node history = " << recv_pkg.nodehistory() << exit_code << " Interval " << diff.tv_sec << " S " << diff.tv_nsec << " NS" << " server " << worker->ip; fin_str = out.str(); pthread_mutex_lock(&mutex_finish); //fin_fp << fin_str << endl; pthread_mutex_unlock(&mutex_finish); } delete qi; } } }