jobcache directories data retention

[Or Sher <or...@gmail.com>]

Hi all,
Our hadoop nodes suffer from high utilization of inodes, which probably
eventually brings us to blacklisted job trackers.
We found that a lot of the inodes are used under the jobcache library as
directories (most empty, some are not) of what it seems as a long finished
jobs.

It does looks like there is some kind of a retention job which removes all
unnecessary folders but I'm don't know where is it scheduled and I can I
configure it to run in shorter intervals or keep less data.

We're using CDH 4.3

Can someone shed some light here?

Thanks a lot.

-- 
Or Sher