Namenode reporting empty files for originally non-empty files

[Christian Kunz <ck...@yahoo-inc.com>]

On a cluster running hadoop-0.17.2:

We have a handful of files that originally were not empty but now reported
with 0 size.
I checked the corresponding blocks on the DataNodes and they are indeed
non-empty.

I restarted the namenode, with no success.

With the non-empty blocks I can re-create these files, but because *this is
a big issue* some hadoop developers might want to get more information
before I replace the files in dfs.

But it would have to be within the next couple of hours.

In addition I will file a jira.

-Christian