Context of when and why the failure occurred:

The CMAP spec requires interrupting in-use connections on a monitor timeout because otherwise a dead connection’s response time can block for the OS’s TCP retransmission timeout, and only a socket timeout and interruptInUseConnections can abort.

Stack trace:

N/A

Detailed steps to reproduce the problem?

1. SDAM dedication connection is connection B; Cursor blocks on getMore for connection A
2. The network breaks all existing connection but new connections still work, e.g. LB4 failover
3. Heartbeat on B times out, closes
4. Monitor creates connection C, retry concludes healthy, no clear / interupt
5. Nothing interrupts A and it waits minutes for the TCP transmission / socket timeout

Definition of done: what must be done to consider the task complete?

Clear with interruptInUseConnections=true on the first monitor timeout (resolve the GODRIVER-2577 delay) without reintroducing the FaaS regression (see comments).

Unskip these tests:

TestUnifiedSpec/server-discovery-and-monitoring/tests/unified/interruptInUse-pool-clear.json/Connection_pool_clear_uses_interruptInUseConnections=true_after_monitor_timeout
TestUnifiedSpec/server-discovery-and-monitoring/tests/unified/interruptInUse-pool-clear.json/Error_returned_from_connection_pool_clear_with_interruptInUseConnections=true_is_retryable
TestUnifiedSpec/server-discovery-and-monitoring/tests/unified/interruptInUse-pool-clear.json/Error_returned_from_connection_pool_clear_with_interruptInUseConnections=true_is_retryable_for_write

The exact Go version used, with patch level:

N/A

The exact version of the Go driver used:

N/A

Describe how MongoDB is set up. Local vs Hosted, version, topology, load balanced, etc.

N/A

The operating system and version (e.g. Windows 7, OSX 10.8, ...)

N/A

Security Vulnerabilities

N/A