Re: [Condor-users] incomplete udp for command 0 / command 2

[Todd Tannenbaum <tannenba@xxxxxxxxxxx>]

Daniel Forrest wrote:
> Rob,
>
> > After monitoring condor udp traffic for a while, I've found an
> > interesting problem.. sometimes clients will start misbehaving, and
> > send only part of the data needed for update_master_ad /
> > update_startd_ad commands.
> >
> > Has anyone seen this? Any ideas on what's causing it?
>
> Yes, and my first thought is, is this Windows?
>
> <snip>
>
> > Both clients are windows XP, condor version 6.8.6. I've noticed this
> > behavior on both, sometimes one, sometimes the other. The problem is
> > gone after a condor_restart, but will eventually re-occur. The
> > client logfiles don't show anything interesting.
> >
> > Any ideas on how to debug / fix this would be welcome.
>
> This is a problem with UDP under Windows, it considers a packet "sent"
> when the sendto() call is made, not when the packet has actually hit
> the wire.  So if sendto() is called too rapidly (e.g. when collector
> update packets are split) you can lose the previous UDP packet if it
> hasn't really been sent yet.
>
> What we did was add "D_NETWORK" to the MASTER_DEBUG and STARTD_DEBUG
> flags in the config file.  The added delay of logging the UDP packets
> seems to be enough to keep this from happening.
>
> You can alternatively use "UPDATE_COLLECTOR_WITH_TCP = True" and avoid
> UDP entirely.


Starting with Condor v6.8.7, we made some changes to how Condor handles 
datagrams.  This fixed problems we observed on some Unix platforms that 
occasionally caused UDP datagrams to never be sent if the datagrams are 
very large (e.g. a machine ad with many custom attributes) or if lots of 
datagrams are sent in rapid succession (e.g. a machine with lots of slots).

It is quite possible these same changes will improve life on Windows, 
although we never confirmed it.



--
Todd Tannenbaum                       University of Wisconsin-Madison
Condor Project Research               Department of Computer Sciences
tannenba@xxxxxxxxxxx                  1210 W. Dayton St. Rm #4257
Phone: (608) 263-7132                 Madison, WI 53706-1685