Postgres replication slot can get into a weird state during rebalancing

Hello, we have a fairly large K8s Connect cluster with 100s of Debezium 10.0 Beta 3 PG Connectors running against PG 9.6 servers using decoderbufs. During rebalancing very rarely we will get a "broken" slot with this error:

ERROR: subtransaction logged without previous top-level txn record

The only fix is to remove the connector, drop the slot, and then remake everything. I am not sure if this error is related to anything Debezium is doing, but I am posting here to get any insight because the error is so rare and there is very little information on the web. Has anyone else seen this?