questions on the clients' shutdown strategy of mongo-c-driver

[asyouknow]

i have briefly read the code of mongo-c-dirver 1.1.0 and mongo-c-driver. i have found that in function mongoc-client_pool_push there is a strategydifferent.

in 1.1.0:

void

mongoc_client_pool_push (mongoc_client_pool_t *pool,

                         mongoc_client_t      *client)

{

   ENTRY;

   bson_return_if_fail(pool);

   bson_return_if_fail(client);

   /*  

    * TODO: Shutdown old client connections.

    */

   /*  

    * TODO: We should try to make a client healthy again if it

    *       is unhealthy since this is typically where a thread

    *       is done with its work.

    */

   mongoc_mutex_lock(&pool->mutex);

   _mongoc_queue_push_head(&pool->queue, client);

   mongoc_cond_signal(&pool->cond);

   mongoc_mutex_unlock(&pool->mutex);

   EXIT;

}

in mongoc-1.3.5

void

mongoc_client_pool_push (mongoc_client_pool_t *pool,

                         mongoc_client_t      *client)

{

   ENTRY;

   BSON_ASSERT (pool);

   BSON_ASSERT (client);

   mongoc_mutex_lock(&pool->mutex);

   if (pool->size > pool->min_pool_size) {

      mongoc_client_t *old_client;

      old_client = (mongoc_client_t *)_mongoc_queue_pop_head (&pool->queue);

      if (old_client) {

          mongoc_client_destroy (old_client);

          pool->size--;

      }   

   }   

   mongoc_mutex_unlock(&pool->mutex);

   mongoc_mutex_lock (&pool->mutex);

   _mongoc_queue_push_tail (&pool->queue, client);

   mongoc_cond_signal(&pool->cond);

   mongoc_mutex_unlock(&pool->mutex);

   EXIT;

}

in a multi-thread scene, when i push a client back it will destroy the oldest client(min_pool_size is 0 default). And it can not be estimated that when to new a client(think 2 threads, it may run simultaneously or one by one to do mongo op), which may cause some  efficiency problem.

I wander may this change is neccessary?

thanks!

[A. Jesse Jiryu Davis]

Yes, you pointed out a bug that we've fixed in the most recent release, 1.4.0:

https://jira.mongodb.org/browse/CDRIVER-1149

Starting with 1.4.0, the default behavior is to keep mongoc_client_t objects alive in the pool.

[asyouknow]

I have read the code of 1.4.0, it make the client pool LIFO and wonder why it close the client at the head? I think the client in tail is the oldest?

在 2016年8月23日星期二 UTC+8上午9:35:27，A. Jesse Jiryu Davis写道：

[A. Jesse Jiryu Davis]

Thanks, investigating:

https://jira.mongodb.org/browse/CDRIVER-1558

[A. Jesse Jiryu Davis]

libmongoc 1.4.2 has been released with the fix, thanks for reporting it.