The sharding balancer shutdown process eventually calls Balancer::waitForBalancerToStop in order to wait for the balancer thread to terminate. This method is called both by replication promotion to primary and by the shutdown sequence.

If a node being promoted to primary is shutdown at the same time, it is possible that two calls enter Balancer::waitForBalancerToStop concurrently which will both call std::thread::join, which is not allowed.