How to recover the state from the previous checkpoint in HDFS when the custer is DOWN and the edits file was broken.

["yajun.don" <do...@gmail.com>]

Hi, NN was unable to process any request since it's cast more exceptions.
and then the NN can't be started, because fseditlog reads a unknown opcode
when NN is loading edits file.

Now, the dfs.name.dir directory below is:

[yuchao@d1950 ~]$ ll /home/yuchao/hadoop/name/current/
total 4332
-rw-rw-r-- 1 yuchao yuchao   14289 Sep 23 16:13 edits
-rw-rw-r-- 1 yuchao yuchao 1049088 Sep 23 17:08 edits.new
-rw-rw-r-- 1 yuchao yuchao 2179387 Sep 23 15:13 fsimage
-rw-rw-r-- 1 yuchao yuchao 2184070 Sep 23 21:36 fsimage.ckpt
-rw-rw-r-- 1 yuchao yuchao       8 Sep 23 15:13 fstime
-rw-rw-r-- 1 yuchao yuchao     100 Sep 23 15:13 VERSION

and the fs.checkpoint.dir directory is:

[yuchao@d1950 ~]$ ll fordim/tmp/dfs/namesecondary/current/
total 4288
-rw-rw-r-- 1 yuchao yuchao   14289 Sep 23 18:03 edits
-rw-rw-r-- 1 yuchao yuchao       4 Sep 23 18:03 edits.new
-rw-rw-r-- 1 yuchao yuchao 2179387 Sep 23 18:03 fsimage
-rw-rw-r-- 1 yuchao yuchao 2178424 Sep 23 18:03 fsimage.ckpt