can not remove inode when cluster state error

[mirrorll]

with a 3 node and 6 osd cluster.(node1: [osd.1, osd.2]. node2:[osd.3, osd.4], node3:[osd.5, osd.6] ).

1. if node3 down, then some pg become degrade, in this time can not remove inode.
2. after osd_out_time pg become active, can remove, and remove it.
3. then in some times node3 up, join to cluster again. but osd.5 and osd.6 has some object of the already remove inode, with peering they write to other osd.