[Carpet] Carpet IO oddity
Wolfgang Kastaun
kastaun at tat.physik.uni-tuebingen.de
Mon May 1 15:54:42 CEST 2006
Erik Schnetter wrote:
> On May 1, 2006, at 07:57:07, Wolfgang Kastaun wrote:
>
>> The order was 0 1 1 2 2 3 3 ....
>
>
> That points to two processes writing at the same time.
>
>> I dont know why it did not abort, but it ran for hours as I can see from
>> the output of the time command at the end of my log. I am pretty sure I
>> did not start this run twice the same time, this would leave behind
>> output from our queing system, but it did not.
>
>
> The queuing systems used at AEI and LSU have the property that they
> sometimes do not kill jobs that should be killed. Without user fault,
> there can be several processes running. For example, it could be that
> your queuing system started a job, lost contact to the demon on that
> node, assumed that it had died before it started, and restarted the job
> on another node a few seconds later.
>
> I just have an idea how thorn Formaline can detect this case.
>
> -erik
>
Good news, no it seems to work. Your hypothesis seems plausible, it is
only funny that the files are so cleanly interwoven: the two processes
must have been running with the same speed in our heterogenous environment.
So until it happens again we can forget about it.
Nevertheless it would be nice if Carpet could detect if files (also .asc
files) it creates are already there and issue a warning, or even abort.
Wolfgang.
More information about the developers
mailing list