Wrapper around Connection that provides automatic retry and failover for MongoDB replica set operations.

This class wraps a Connection and automatically retries failed operations on different replica set members. It detects retriable errors (network failures, "not master" errors, etc.) and seamlessly fails over to another suitable server.

Usage example:

ReplicaSet rs(config);
ReplicaSetConnection::Ptr conn = new ReplicaSetConnection(rs, ReadPreference::Primary);

OpMsgMessage request("mydb", "mycollection");
request.setCommandName(OpMsgMessage::CMD_FIND);
request.body().add("filter", filterDoc);

OpMsgMessage response;
conn->sendRequest(request, response);  // Automatic retry on failure

THREAD SAFETY: This class is NOT thread-safe, just like Connection. Each thread must have its own ReplicaSetConnection instance, or use connection pooling with external synchronization.

For multi-threaded applications, use ReplicaSetPoolableConnectionFactory with Poco::ObjectPool.

Creates a ReplicaSetConnection for the given replica set and read preference. The connection is established lazily on first use.

Destroys the ReplicaSetConnection.

Returns the address of the currently connected server. Throws Poco::NullPointerException if not connected.

Returns a reference to the underlying Connection. Throws Poco::NullPointerException if not connected.

Returns true if currently connected to a server.

Returns true if the currently connected server still matches the read preference. Returns false if not connected or if the server no longer satisfies the read preference. This is useful for connection pool validation to detect when a server role has changed (e.g., primary became secondary).

Reads a response for a previously sent request.

Forces reconnection by selecting a new server from the replica set. Useful if you detect an error and want to explicitly retry.

Sends a request and reads the response. Automatically retries on retriable errors with failover.

Throws Poco::IOException if all retry attempts fail.

Sends a one-way request (fire-and-forget). Sets MSG_MORE_TO_COME flag and acknowledged=false.

Note: One-way requests are not retried on failure.