Some parts of the test infrastructure (e.g. test hooks) assume that they can connect to a replica set's nodes and run read operations on their local data (e.g. to run consistency validations). For regular replica sets, this assumption is sound even in the absence of a majority or when replication is stopped.

However, currently a single-shard cluster with replica set endpoint and config shard can become unavailable for reads when a majority of nodes is down, or replication is stopped. This is because loading the latest filtering metadata depends on the ShardServerCatalogCacheLoader, which forces a refresh on the primary and wait for replication. This can result in operations failing with errors such as FailedToSatisfyReadPreference.

This ticket tracks action items TODO(SERVER-98707) in the code that are will be resolved once this problem is resolved.