Hi Mischa Salle,
Thanks very much!
I'll test if it works.
Xiaofei
在 2015/2/12 21:41, Mischa Salle 写道:
> Hi Xiaofei,
>
> There are a few things you can try on the Argus server, see also
> https://twiki.cern.ch/twiki/bin/view/EGEE/ArgusEMIDeployment#Known_Issues
> in particular disabling the caching of the PDP responses in the PEPd,
> perhaps also the increase of the default memory.
>
> If that still is not enough, it might be needed to increase the number
> of simultaneously handled requests, see
> https://twiki.cern.ch/twiki/bin/view/EGEE/AuthZPEPDConfig#Advanced_Configuration_Options
> the entry 'maximumRequests' in the [SERVICE] and [PDP] sections. Both
> default to 200. I have no experience with increasing those. You probably
> should keep the two values identical to each other.
>
> Good luck!
>
> Best wishes,
> Mischa Sallé
>
> On Thu, Feb 12, 2015 at 02:45:22AM +0800, Yan Xiaofei wrote:
>> Hi Paolo,
>>
>> There are still some error about timeout problem.
>> There are lots of pilot jobs by cms.
>> Is there something I can do for tunning the performance of argus server?
>>
>> Thanks very much.
>> Xiaofei.
>>
>> 在 2015/2/12 0:35, Yan Xiaofei 写道:
>>> Hi Paolo,
>>> Thank very much.
>>> I changed the configuration by your advise. And wait if it
>>> resolve the problem.
>>> All our CE and Argus are common installation, no load balancer for argus.
>>> We have 1100 job slots running now. Maybe more than 2000 in the future.
>>> Is there any other recommendation for tunning the argus performance?
>>> How many requests that one argue server can support at the same time?
>>>
>>> Best Regards.
>>> Xiaofei
>>>
>>> 在 2015/2/9 18:20, Paolo Andreetto 写道:
>>>> On 08/02/2015 06:30, Yan Xiaofei wrote:
>>>>> Hello
>>>>>
>>>>> I have a argus server.
>>>>> We have lots of jobs running recently. The cream ce always
>>>>> report socket timeout problem with argus server.
>>>>> Here is the error message from job submit:
>>>>>
>>>>> FATAL - EOF detected during communication. Probably service
>>>>> closed connection or SOCKET TIMEOUT occurred.
>>>>>
>>>>> The error message from cream ce was:
>>>>> 08 Feb 2015 12:25:11,478 WARN
>>>>> org.glite.ce.commonj.authz.argus.ArgusConfigHandler - Missing
>>>>> or wrong argument timeout; default value used
>>>>> 08 Feb 2015 12:25:11,478 WARN
>>>>> org.glite.ce.commonj.authz.argus.ArgusConfigHandler - Missing
>>>>> or wrong argument connection_per_host; default value used
>>>>> 08 Feb 2015 12:25:11,479 WARN
>>>>> org.glite.ce.commonj.authz.argus.ArgusConfigHandler - Missing
>>>>> or wrong argument max_connection; default value used
>>>>>
>>>>> I change argus pepd server parameter:
>>>>> connectionTimout=180
>>>>> maximumRequests=3000
>>>>> requestQueueSize=2000
>>>>>
>>>>> But I can find info fom argus-pepd log that the
>>>>> connectionTimout still use default value.
>>>>>
>>>>> And the cream ce still had lots of timeout problem.
>>>>>
>>>>> Xiaofei
>>>> Hi Yan
>>>>
>>>> You can try to change the parameter on the client side, i.e. the
>>>> CREAM service.
>>>> In the file /etc/glite-ce-cream/cream-config.xml, in the tag
>>>> argus-pep, add the following parameters (default values used in
>>>> the example):
>>>>
>>>> timeout="5000" (HTTP connection timeout in millis)
>>>> connection_per_host="5" (maximum number of connections per
>>>> host to keep alive)
>>>> max_connection="20" (maximum total number of connections in
>>>> the connections pool to keep alive)
>>>>
>>>> then you've got to restart the CREAM service.
>>>>
>>>> Is there anything particular in your installation, for example
>>>> are you using load balancers for Argus or very complex network
>>>> structures?
>>>> --
>>>> ----------------------
>>>> Ing. Paolo Andreetto
>>>> INFN Sezione di Padova
>>>> Via Marzolo, 8
>>>> 35131 Padova - Italy
>>>>
>>>> Tel: +39 049.967.7378
>>>> ----------------------
|