DCP backfill doesn't read expected data

During backup the dcp connection hangs. Below are the logs from the connection of one of the hung vbuckets.

Wed Feb 4 18:03:36.334143 CST 3: (boss_event_1) DCP (Producer) eq_dcpq:cbbackup-qeYHjBNEGfOVoQKb - (vb 99) stream created with start seqno 0 and end seqno 381517
Wed Feb 4 18:03:36.433578 CST 3: (boss_event_1) DCP (Producer) eq_dcpq:cbbackup-qeYHjBNEGfOVoQKb - (vb 99) Sending disk snapshot with start seqno 0 and end seqno 381517
Wed Feb 4 18:03:36.436566 CST 3: (boss_event_1) DCP (Producer) eq_dcpq:cbbackup-qeYHjBNEGfOVoQKb - (vb 99) Backfill complete, 0 items read from disk, last seqno read: 381507
Wed Feb 4 18:03:36.436577 CST 3: (boss_event_1) Backfill task (1 to 381517) finished for vb 99 disk seqno 381517 memory seqno 381517

As we can see the backfill task should have read up to sequence number 381517, but it only reads up to 381507. This is the root cause and this behavior is unexpected. Note that we do have a cursor in the checkpoint manager for the dcp connection, but it is not expecting to actually read any items.

vb_99:eq_dcpq:cbbackup-qeYHjBNEGfOVoQKb:cursor_checkpoint_id: 29498
vb_99:eq_dcpq:cbbackup-qeYHjBNEGfOVoQKb:cursor_seqno: 381518
vb_99:eq_dcpq:replication:ns_1@node01.domain.com->ns_1@node03.domain.com:boss_event_1:cursor_checkpoint_id: 29498
vb_99:eq_dcpq:replication:ns_1@node01.domain.com->ns_1@node03.domain.com:boss_event_1:cursor_seqno: 381517
vb_99:last_closed_checkpoint_id: 29497
vb_99:num_checkpoint_items: 1
vb_99:num_checkpoints: 1
vb_99:num_items_for_persistence: 0
vb_99:num_open_checkpoint_items: 0
vb_99:num_tap_cursors: 2
vb_99:open_checkpoint_id: 29498
vb_99:persisted_checkpoint_id: 29497
vb_99:state: active