We have noticed that the 'Reset' command works inconsistently for SQL replication. if we reset a replication configuration it may or may not work. It always removed the item from the collection in /docs/Raven/SqlReplication/Status, but it will sometimes come back immediately showing as being fully replicated.
We have noticed that it often happens that if we do more than one, the first one is more likely to succeed, but subsequent resets will almost always fail. I haven't had a chance to dig into the code on this, but my theory is that the thread running the replication is not checking for concurrency on the tracking document and is overwriting it when the batch is complete. so the busier replication is, the more likely it is to ignore a reset command.
The only workaround we have found is to delete the replication instance, wait at long enough for a replication batch or two to finish, and re-add the replication instance.
Also note, a 'nice to have' feature would be a way to reset multiple replication instances at once so they process as a batch.