Connection pools

mongocxx::client vs mongocxx::pool

A standalone mongocxx::client uses a single-threaded algorithm to monitor the state of the cluster it’s connected to. When connected to a replica set, the thread “stops the world” every 60 seconds to check the status of the cluster. A mongocxx::pool, on the other hand, uses a separate background thread for each server in the cluster, each of which checks the status of the server it monitors every 10 seconds. Because of the performance advantages of monitoring the cluster in the background rather than “stopping the world”, it’s highly recommended to use a mongocxx::pool rather than a set of standalone clients if your application has access to multiple threads, even if your application only uses one thread.

Connection pools and thread safety

A mongocxx::pool can be shared across multiple threads and used to create clients. However, each mongocxx::client can only be used in a single thread. See the thread safety documentation for details on how to use a mongocxx::client in a thread-safe manner.

Configuring a connection pool

The number of clients in a connection pool is determined by the URI parameters minPoolSize and maxPoolSize. The minPoolSize and maxPoolSize options set resource usage targets for when the driver is idle or fully-utilized. When fully utilized, up to maxPoolSize clients are available. When clients are returned to the pool, they are destroyed until the pool has shrunk again to the minPoolSize.

maxPoolSize The maximum number of clients created by a mongocxx::pool (both in the pool and checked out). The default value is 100. Once it is reached, mongocxx::pool::acquire blocks until another thread returns a client to the pool.
minPoolSize Sets a target size for the pool when idle. Once this many clients have been created, there will never be fewer than this many clients in the pool. If additional clients above minPoolSize are created, they will be destroyed when returned to the pool. The default value is “0”, which disables this feature. When disabled, clients are never destroyed.

Using a connection pool

To use a connection pool, first create a mongocxx::pool, passing the URI as an argument. Then, call mongocxx::pool::acquire to receive a client from the pool. The client will automatically be returned to the pool when it goes out of scope.

See the connection pool example for more details.