Re: "job.split" file not found

[Harsh J <ha...@cloudera.com>]

Hello Yuval,

This _could_ be a CDH specific issue, so I'm moving it to the CDH
users list (cdh-user@cloudera.org). [bcc'ing mapreduce-user@]

Some questions inline:

On Tue, Mar 22, 2011 at 11:15 PM, Yuval Oren <yu...@ooyala.com> wrote:
> Hello all,
> I'm having trouble getting a large mapping job to complete. Several of
> thousands of mappers are failing with this error:

Could you give us some details on how large your job was (how many
mappers, did all fail, if not all fail then when do the maps start to
fail)?

> java.io.FileNotFoundException: File does not exist:
> /data/hadoop/cache/mapred/mapred/staging/yuval/.staging/job_201011120027_144772/job.split

The job.split file is created on the DFS's staging root, and it ought
to be preserved there until the job completes, usually. Are you
running any other operation that could've affected the staging
directory's contents (under /data/hadoop/ on your DFS) while your job
was executing?

> This is on Cloudera's CDH3 release. Any ideas?

Which version of CDH3 are you using? Is it B4?

-- 
Harsh J