max_dirty_mb is an ages old kludge we had to ensure faster flushing that avoided various timeouts, limits on memory consumption and limits on data lost should the ost run out of space.

All of those now have proper mechanisms: we have grants to ensure nothing runs out of space, we have per-node dirty and clean pages limits and we have lock prolongation for active IO.

As such I believe we no longer need the max_dirty_mb