Re: [Condor-users] Condor doesn't recognize that tasks have finished

[John Wheez <john@xxxxxxxxxx>]

ah now i am trying to set the lease duration from within my submit 
file...i allow users to estiamte how long the calculations should 
take..then i add 10% on to that amount...hopefully that should minimize 
any cpus from being stuck in an endless lease....


John Wheez wrote:

> I am fiding that if I have a large amount of jobs which take a while 
> to compute that Condor will ecentually stop realizing that tasks have 
> finished. Once this happens all my cpus become unavailable because 
> each cpu thinks it is processing a job. If i go to the actual CPU, the 
> usage is at 0% idle..and the result of the program was saved to the 
> network. This happens on dual CPU and single CPU machines too.
>
> If I place teh tasks on HOLD , Condor starts to work again but 
> eventually sometimes teh CPU's again become available for the same 
> resons.
>
> Here is the error o see in the StarterLog of  a machine which still 
> thought the task was running:
>
> 5/15 23:00:50 About to exec C:\WINDOWS\System32\cmd.exe /Q /C task01.bat
> 5/15 23:00:50 Create_Process succeeded, pid=2588
> 5/15 23:01:16 Process exited, pid=2588, status=0
> 5/15 23:01:39 getpeername failed so connect must have failed
> 5/15 23:02:04 Connect failed for 30 seconds; returning FALSE
> 5/15 23:02:04 FileTransfer: Unable to connect to server 
> <192.168.0.3:9611>    <<<-----Windows XP STARTD Machine
> 5/15 23:02:04 JIC::allJobsDone() failed, waiting for job lease to 
> expire or for a reconnect attempt  << LEase never expires
>
> Condor Collector/Negotiator: Condor 6.7.6 on Linux 7.2 / Dec Alpha
>
> Condor StartD Machines: Condor 6.7.6 on Windows XP
> _______________________________________________
> Condor-users mailing list
> Condor-users@xxxxxxxxxxx
> https://lists.cs.wisc.edu/mailman/listinfo/condor-users