This is obviously a very hazardous situation you are in (if your data is
important), so I'd suggest moving carefully. Make as many backups of as
many things you can.
The usual mechanism that Hadoop uses when upgrading is to rename
directories of the old format and keep them around until the admin
finalizes the upgrade. Here is the relevant method :
. You will probably have to dig into code and see what operations were
performed, where the failure occurred, and figure out how best to fix it.
In fact for your particular upgrade there were quite substantial changes
not just in the Namenode formats, but also the layouts on the datanodes
(which you may have to work on once you recover your namenode) .
https://issues.apache.org/jira/browse/HDFS-6482 . I'm guessing operations
will take a long time and may not work. Why do you need to rollback? We are
on 2.7.2 and its working fine for us.
On Wed, Oct 12, 2016 at 11:01 PM, Dinesh Kumar Prabakaran <
> Hi Guys,
> Did rolling upgrade from hadoop 2.5.2 to hadoop 2.7.2 and *did not
> finalize* the upgrade. Now I wished to rollback to 2.5.2 version based on
> Starting Name node 1 as active with *-rollingUpgrade rollback*, it gets
> shuts down with the following exception,
> Unexpected version of storage directory ..\Metadata\data\dfs\namenode.
> Reported: -63. Expecting = -57.*
> There is already a task regarding this but the status is *Open*.
> Please let me know are there any *work around *to rollback HDFS from
> rolling upgrade without any issues.
> Dinesh Kumar P