Currently we request to reset all the cursors before the step up. One of the reasons for this (I don't know if there are others) is that on a follower we open the stable table as a checkpointed btree but for leader we should always open it as a live btree to be able to serve writes to the stable table.

Currently we don't have a safe way to reopen a table in all the possible cases (see 
__clayered_can_advance_stable). One example could be when the cursor is positioned:
 

// start as a follower
cursor.search() // positioned on the stable
// step-up
cursor.update() // __clayered_can_advance_stable returns false since we are positioned on the cursor, and we try to write to a checkpointed btree 

It might be a part of our job that should be done to rework layered cursors and prepare them for async step-up/step-down