If mongos attempts to read from a view, the shard will throw an exception that forces mongos to retry the command as an aggregation on the underlying collection. Currently, single replica set transactions abort on any error encountered while processing a command, so reading from a view in sharded transaction will abort the entire transaction. Instead, the shard should not abort on CommandOnShardedViewNotSupportedOnMongod and mongos should be able to update its transaction tracking state to account for this.