近期代码两次出core总结

近期修改intu后端对Searchhub、Suggestion异步连接时都出core。
其中sh只能怀疑是循环变量未初始化，for(int idx; i<max; i++)，目前线上已ok，应该就是这个问题

针对sugg目前怀疑是sugge和后端数据源变量共享导致的问题，问题代码如下: 
 (1)Sugg的状态变量仍采用ori_stat，而一旦suggestion处理完成，则再将其置回O_WAIT以便于后续处理sh或cache。这就触发了一个临界状态，即Sugg处理完成了，但超时方法也即将触发，这将导致刚刚borrow成功的sh连接被认作超时处理。
 (2)Sugg的连接变量仍采用hdle，当suggestion接收成功后，也会触发一个临界状态：
    process(worker);
    worker->hdle->pool->returnSket(worker->hdle);
    process(worker)会后续进行sh或cache连接borrow，borrow到的连接也会存至hdle，由于是异步的，return可能就return为刚刚borrow的成功的连接（被再次认为可用，出问题）
  目前suggestion异步定位是这两个问题。
 
 修改的话，即所有变量都不要共享为最佳原则。
 但即便如此问题2依然在超时或者错误重发时导致问题。即连接错误，重发后borrow到的连接也被迅速返回了。

 修改方案如下：
 int sugg_collector::process(Worker* worker) {
  unregisterWorker(worker->id);
  _INFO("[reqid=%08X] process worker, valid=%d", worker->id, worker->valid);
  if (worker->valid) {
    if (worker->hdle->rtype == CHUNK {
               ************此处省略部分代码****************
    }
    else if (worker->hdle->rtype == BODYLEN) {
               _INFO("go bodylen");
                worker->sugg_res = std::string(worker->hdle->recv_buf + worker->hdle->header.length(),worker->hdle->body_len);
               _INFO("sugg res = %s",worker->sugg_res.c_str());
          }
  }
  //return submitWorker(worker); //此处我们把submitWorker注释掉
 }
 
  并将fire代码修改为：
  if (should_process) {
   process(worker);
   worker->hdle->pool->returnSket(worker->hdle);
   submitWorker(worker);
  }
 而notifytimeout代码可直接修改为：
  if (is_to) {
   //worker->entity.is_res_to = 1;
   int drop_ret = receiver->dropSket(worker->hdle);
   _INFO("[reqid=%08X] sugg to triggered, drop_ret=%d", id, drop_ret);
   //process(worker); // 此处的process神马都不干，所以直接提交worker即可
   submitWorker(worker);
  }

后期注意事项：

1、加强对代码的理解，锁到底是锁到什么粒度要搞清楚；现在改代码总是照猫画虎老画出问题；
2、不同的处理流程不要共用变量，这会超出既有锁的控制范围，导致触发各种临界状态。
3、多注意积累，不要局限于简单的业务代码。
4、最后，希望是这两个问题导致，祈祷明天上线顺利。



*****************问题代码*********************

int sugg_collector::fire(unsigned int worker_id, int type, void* arg) {
    	_INFO("[reqid=%08X] sugg arrive", worker_id);
	Worker* worker = NULL;
	 worker = findWorker(worker_id);
 if (worker != NULL) {
  bool should_process = false;
  worker->lock();
  if (worker->ori_stat == O_WAIT) {
   timer->unregister_timer(worker->ori_timer_id, worker->id);
   should_process = true;
   if (type < 0) {
    _INFO("[reqid=%08X] sugg err", worker_id);
    worker->ori_stat = O_GET_FAIL;
   }
   else {
    worker->ori_stat = O_GET_SUCC;
   }
  }
  worker->unlock();
  if (worker->ori_stat == O_GET_FAIL) {
   worker->valid = false;
  }
  if (should_process) {
   process(worker);
   worker->hdle->pool->returnSket(worker->hdle);
  }
 }
 else {
  _INFO("[reqid=%08X] sugg al to", worker_id);
 }
    return 0;
}

int sugg_collector::notify_timeout(unsigned int id, const timeval & now,const unsigned int time_out_ms) {
    _INFO("[reqid=%08X] sugg to", id);
 timeval cur;
 gettimeofday(&cur, NULL);
 Worker* worker = NULL;
 worker = findWorker(id);
 if (worker != NULL) {
  _INFO("cost=%ld", cur.tv_sec * 1000 + cur.tv_usec / 1000 - worker->ori_timestamp.tv_sec * 1000 - worker->ori_timestamp.tv_usec / 1000);
  bool is_to = false;
  worker->lock();
  if (worker->ori_stat == O_WAIT) {
   //worker->reason = reason_recv_to;
   worker->valid = false;
   is_to = true;
   worker->ori_stat = O_TO;
           worker->entity.sugg_to++;
  }
  worker->unlock(); 
  if (is_to) {
   //worker->entity.is_res_to = 1;
   int drop_ret = receiver->dropSket(worker->hdle);
   _INFO("[reqid=%08X] sugg to triggered, drop_ret=%d", id, drop_ret);
   process(worker);
  }
 }
 else {
  _INFO("[reqid=%08X] sugg arr", id);
 }
 return 0;
}