[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [HTCondor-users] howto set ulimit -l unlimited for openmpiscript or other condor jobs on debian 9 for htcondor 8.8?



Hi Harald,

Am 29.03.19 um 21:14 schrieb Harald van Pee:
> Hello Tim,
> 
> thank you for your good explanation in principle this works:
> condor.service - Condor Distributed High-Throughput-Computing
>    Loaded: loaded (/lib/systemd/system/condor.service; enabled; vendor preset: 
> enabled)
>   Drop-In: /etc/systemd/system/condor.service.d
>            ââcondor.conf
> BUT:
> LimitMEMLOCK=unlimited

I did not confirm by testing, but the documentation in the man page
( https://manpages.debian.org/stretch/systemd/systemd.exec.5.en.html , see the section with LimitMEMLOCK)
states that "infinity" should be used instead of the formerly known "unlimited". 
As usual, systemd tools use their own names for things... 

> 
> results still in 
> max locked memory       (kbytes, -l) 64
> 
> therefore I tried numbers and found out 
> LimitMEMLOCK=32768
> results in 
> max locked memory       (kbytes, -l) 32

That matches the documentation (see table 1). The default unit of LimitMEMLOCK is in bytes. 
One nice advantage of systemd  (there are some advantages indeed!) 
is that you should also be able to write something like:
LimitMEMLOCK=4M:16M
to get a soft limit of 4M and a hard limit of 16M (per process in the tree). 

All the best, have a nice weekend and let us know how it goes,
Oliver

> 
> LimitMEMLOCK=99999999
> results in 
> max locked memory       (kbytes, -l) 97656
> 
> LimitMEMLOCK=9999999999
> results in 
> max locked memory       (kbytes, -l) 9765624
> 
> obviously something goes wrong but at least it seems high numbers can be used 
> as a work around.
> 
> Now the question is is the problem in htcondor or systemd or the debian 
> kernel? Or do I still something wrong?
> 
> Best
> Harald
> 
> 
> On Friday, March 29, 2019 3:49:21 PM CET Tim Theisen wrote:
>> Hello Harald,
>>
>> The /etc/init.d/condor file is not used. Distributing it was a mistake
>> on our part.
>>
>> So, we should modify the systemd configuration.
>>
>> When updating the condor service, one should not modify the system
>> installed service file. Instead, create a file with overrides to the
>> distributed configuration. Here are the steps:
>>
>> 1. mkdir /etc/systemd/system/condor.service.d 2. Put the following 2 lines
>> into
>  /etc/systemd/system/condor.service.d/condor.conf
>>
>> [Service]
>> LimitMEMLOCK=unlimited
>>
>> 3. Force a systemd reload: systemctl daemon-reload
>>
>> 4. Finally, restart HTCondor: systemctl restart condor
>>
>> ...Tim
>>
>> On 3/28/19 3:56 PM, Harald van Pee wrote:
>>
>>> Dear htcondor experts,
>>>
>>>
>>>
>>> we running htcondor 8.8.1 on debian 9 with vanilla universe jobs without 
>>> problems and now want to start openmpi jobs in parallel universe, but
>>> to do so 
>>> we need to set max locked memory to a high value
>>> I have set
>>> ulimit -l unlimited
>>> in 
>>> /etc/init.d/condor
>>> and
>>> the openmpiscript itself.
>>> And added to
>>> /lib/systemd/system/condor.service
>>> [Service]
>>> ...
>>> LimitMEMLOCK=unlimited
>>>
>>>
>>>
>>>
>>> then I have down a
>>> systemctl daemon-reload
>>> systemctl restart condor
>>>
>>>
>>>
>>> on all condor hosts.
>>>
>>>
>>>
>>> But after starting openmpiscript 
>>> ulimit -l 
>>> inside of the script shows
>>> max locked memory       (kbytes, -l) 64
>>>
>>>
>>>
>>> And as expected the job does not run properly.
>>>
>>>
>>>
>>> What I have done wrong? How I have to set the
>>> max locked memory
>>> limit for condor jobs/scripts?
>>>
>>>
>>>
>>> Best regards
>>> Harald
>>> _______________________________________________
>>> HTCondor-users mailing list
>>> To unsubscribe, send a message to htcondor-users-request@xxxxxxxxxxx with
>>> a
>  subject: Unsubscribe
>>> You can also unsubscribe by visiting
>>> https://lists.cs.wisc.edu/mailman/listinfo/htcondor-users
>>>
>>>
>>>
>>> The archives can be found at:
>>> https://lists.cs.wisc.edu/archive/htcondor-users/
>>
>>
>> -- 
>> Tim Theisen
>> Release Manager
>> HTCondor & Open Science Grid
>> Center for High Throughput Computing
>> Department of Computer Sciences
>> University of Wisconsin - Madison
>> 4261 Computer Sciences and Statistics
>> 1210 W Dayton St
>> Madison, WI 53706-1685
>> +1 608 265 5736
>>
>>
> 
> 

Attachment: smime.p7s
Description: S/MIME Cryptographic Signature