So the quick summary (IIRC), is that we had a transaction which had been flagged as "prepared", but didn't actually have an entry in the txn-queue for every document that was referenced by the transaction.
Note that some of the changes to mgo that should be in 2.3 will at least prevent this from growing inordinately large (as txns that would cause a queue to grow >1000 long should end up immediately flagged as aborted.)
So the quick summary (IIRC), is that we had a transaction which had been flagged as "prepared", but didn't actually have an entry in the txn-queue for every document that was referenced by the transaction.
Note that some of the changes to mgo that should be in 2.3 will at least prevent this from growing inordinately large (as txns that would cause a queue to grow >1000 long should end up immediately flagged as aborted.)