I believe mongoS currently uses "nearest" without a maxStalenessSeconds option as the readPreference when communicating with the config server replica set. In the case where the "nearest" node is a lagged/stuck secondary, config reads (with afterCluserTime) may time out waiting for read concern. Adding a maxStalenessSeconds in the read preference mongoS uses seems to be an option. But I am wondering if mongoS already has other mechanism to retry such operations on a more up-to-date node.