concurrency_metrics and concurrency_replication_metrics are both long running tasks (on average 3.5 and 4.5 hours respectively). We should split them into dynamically generated tasks and set reasonable timeouts.