Jean-Francois Malouin wrote:

* Dustin J. Mitchell <dustin AT zmanda DOT com> [20090908 13:08]:

On Tue, Sep 8, 2009 at 12:03 PM, Jean-Francois
Malouin<malin AT bic.mni.mcgill DOT ca> wrote:

Hmmm, more than a week now and no replies.
So I'll attempt to fix it myself: just to be on the safe side,
any adverse effect to just bump up REP_TIMEOUT to, say 10hrs?

Sorry, I thought I replied, but what I meant to say was essentially:
yes, that sounds like something to fix :)

As long as the client will die if the TCP connection goes away, I
would prefer to get rid of the timeout altogether.  These "long
enough" timeouts are really only relevant for UDP, where the OS
doesn't notify us of a lost connection (since there is no connection).

I'm reviving this thread as I got hit again last night on a server
running 2.6.1p1 that I reinstalled with REP_TIMEOUT=(12*60*60) ie,
12hrs. I'm still getting 6hrs timeouts on some DLEs:

planner: [disk /raid/nih, all estimate timed out]

REP_TIMEOUT is a client timeout, It must be configured on the client.
You can also use "estimate server", less precise estimate but a lot faster.

Jean-Louis