Currently only write commands and user management commands handle writeConcern on mongos. They forward the command to each shard and then aggregate the responses and errors back together. Every command that does writes should do this. Commands just need to make sure that writeConcern is sent with the command to each shard and if multiple writeConcernErrors are returned, they need to be combined. If multiple error codes are returned, the error code should be 64 (writeConcernError).
- is related to
SERVER-20224 Commands which write should take a write concern
- related to
SERVER-23785 Remove ShardRegistry::kWriteConcernError in favor of using an error_class