I have run into this problem before, too. What I found that was a record was deleted by someone on the subscriber side. Transactional replication goes from publisher to subscriber, so when the subscriber's record got deleted, there was no way for the publisher to know that. So it all gets jammed up because its trrying to send a replication command for a record that no longer exists.

There is surely a better resolution for this, but I was in a hurry to get things running again. I ran the snapshot agent and had the subscription re-initialized so the day could continue. It was painful, but it got done.

After that, I ended up switching that table to snapshot replication (as it was before) because if that happened once, it was probably going to happen again. Snapshot doesn't care what you do on the subscriber - its gonna wipe it out and repopulate it anyway.