We have 4 consumers for the same topic, all are supposedly up to date (0 backlog).
But when restarting brokers, thus loading cursor state from zookeeper and bookkeeper, we're consistently getting one consumer reset a long time backwards.
Looking at zookeeper metadata that consumer is the only one to have an outdated markDeleteLedgerId and markDeleteEntryId, this happens for all 10 partitions of the consumer.
Not only has this consumer an outdated cursor, it seems to no be updating it as it consumes all the backlog generated when a broker is restarted and thus consumer is reset.