Tailing the oplog in a sharded cluster can be problematic due to failovers and rollbacks, which may cause events to not be replicated correctly. Several solutions exist to mitigate this issue, including reloading data models, stopping replication and checking buffers, and tailing oplogs of all nodes in a replica set. Additionally, orphan documents that appear in a sharded cluster can be a problem, particularly with multi-update operations, where updates may be recorded in the oplog even if they shouldn't have been visible to the outside. Workarounds for this issue include turning off the balancer process and manually distributing chunks, as well as periodically running cleanupOrphaned. Ultimately, solving this problem requires changes to the MongoDB code base.