[mpich-discuss] hydra proxy questions

Pavan Balaji balaji at mcs.anl.gov
Thu Mar 1 16:49:42 CST 2012


If you kill mpiexec, the proxies should clean up the application processes.

  -- Pavan

On 03/01/2012 04:47 PM, Steve Krueger wrote:
> It's not really a problem. I just wanted to know what the expected behavior is.
>
> What should happen if you kill the mpirun process? Are there any MPI calls
> that talk to the hydra proxy process?
>
>
>
>> -----Original Message-----
>> From: Pavan Balaji [mailto:balaji at mcs.anl.gov]
>> Sent: Monday, February 27, 2012 11:07 PM
>> To: mpich-discuss at mcs.anl.gov
>> Cc: Steve Krueger
>> Subject: Re: [mpich-discuss] hydra proxy questions
>>
>>
>> No, it's not expected.  Can you try out the latest version of MPICH2 to
>> see if this problem still exists?
>>
>>    -- Pavan
>>
>> On 02/27/2012 08:39 AM, Steve Krueger wrote:
>>> I am using MPICH2 1.3.2p1 with hydra launcher and sock device.
>>>
>>> I've noticed that when I start a job via mpirun, the follow processes
>>>
>>> are started on my behalf.
>>>
>>> 1) mpirun
>>>
>>> 2) ssh to all the machines
>>>
>>> 3) hydra_pmi_proxy on all the machine
>>>
>>> 4) my exe on all the machines
>>>
>>> If I kill the mpirun process, 1,2 and 3 are all terminated, but my exe (4)
>>>
>>> continues to function normally.
>>>
>>> My questions:
>>>
>>> 1) Is this expected behavior?
>>>
>>> 2) Should killing the mpirun process kill all the ranks?
>>>
>>> 3) Is there any operation that the ranks could do that require the proxy
>>> process?
>>>
>>> I'm asking, as I've noticed strange hangs in my exe potentially related
>>> to creating new
>>>
>>> MPI_Comm's, but only after I kill the mpirun process.
>>>
>>> sk
>>>
>>>
>>>
>>> _______________________________________________
>>> mpich-discuss mailing list     mpich-discuss at mcs.anl.gov
>>> To manage subscription options or unsubscribe:
>>> https://lists.mcs.anl.gov/mailman/listinfo/mpich-discuss
>>
>> --
>> Pavan Balaji
>> http://www.mcs.anl.gov/~balaji
>
>

-- 
Pavan Balaji
http://www.mcs.anl.gov/~balaji


More information about the mpich-discuss mailing list