Open MPI logo

Open MPI Development Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Development mailing list

Subject: Re: [OMPI devel] RFC: Update libevent
From: Josh Hursey (jjhursey_at_[hidden])
Date: 2012-05-09 09:57:55


Thanks!

On Wed, May 9, 2012 at 9:52 AM, Ralph Castain <rhc_at_[hidden]> wrote:
> Sure! See attached the original note to the libevent folks
>
>
> Hi folks
>
> I'm working with priorities in events using libevent 2.0.13. Since I'm not using the most current release, I thought I'd ask about a behavior we are seeing that might be a potential bug.
>
> I have attached a simple reproducer of the problem. In short, suppose I have setup 8 priorities for my event base. I then execute an event at priority 4. In that callback, I setup and activate two new events - first one at priority 4, and then another at priority 0. We then cycle back through the event loop.
>
> Given the prioritized event base, I would expect the event at priority 0 to be called back first. However, this isn't what happens. Instead, the first event to be defined (in this case, the one at priority 4) is called back first.
>
> Is this the expected behavior? Can someone point me to the error in this reproducer that might be causing the problem?
>
> FWIW: I found that if I adjust the initial event's priority to something different than 4 (i.e., the level of the next event to be defined), then I get the expected behavior. So it appears that the problem is that when we return to the loop, we trigger any pending event of the same priority as the one we were in, instead of looping around to check the highest priority first.
>
> Thanks
> Ralph
>
>
>
>
>
>
> On May 9, 2012, at 7:45 AM, Josh Hursey wrote:
>
>> Ralph,
>>
>
>> Can you point me to the small reproducer for the libevent bug that was
>> fixed? I'm just curious.
>>
>> Thanks,
>> Josh
>>
>> On Fri, May 4, 2012 at 4:09 PM, Ralph Castain <rhc_at_[hidden]> wrote:
>>> FYI: 2.0.19-stable was released yesterday. I have a Mercurial repo all set to go:
>>>
>>> https://bitbucket.org/rhc/ompi-libevent2019
>>>
>>> Please check it out - timeout is now set for May 11th.
>>>
>>> Thanks
>>> Ralph
>>>
>>> On May 1, 2012, at 8:38 AM, Ralph Castain wrote:
>>>
>>>> WHAT:  Update libevent to 2.0.19 release
>>>>
>>>> WHEN:  As soon as it is released, expected around May 11
>>>>
>>>> WHY:     The 2.0.19 release contains a critical fix to a bug I recently discovered in the libevent 2.0.x series
>>>>
>>>>
>>>> Details:
>>>> I discovered a bug in libevent over the last few days that causes it to unexpectedly "invert" event priorities. It is a slightly subtle bug, but we were able to provide a simple reproducer and so the libevent folks were able to quickly implement a fix.
>>>>
>>>> Stated simply, if you were in an event of a given priority and activated an event of higher priority, that new event would not get serviced if any event of the current priority were to become active prior to leaving the current event. In other words, libevent would service all active events of the current priority before even looking to see if a higher priority event was active.
>>>>
>>>> The patch adds the following logic to event_active:
>>>>
>>>>>   IF <I am in an event> AND
>>>>>       IF <ev->base> EQ <current-base> AND
>>>>>       IF <pri> LT <current-pri>  THEN
>>>>>           <rescan queues on next loop>
>>>>
>>>>
>>>> Thus, a rescan only occurs if a higher priority event becomes active during an event of lower priority. Unfortunately, ORTE relies on this behavior to handle errors - without the change, an error reported in a message from a daemon (for example) cannot be serviced until ALL messages that arrive during the processing of the message have been handled. In the case of a large cluster that is receiving a long list of messages, this prevents the error from being handled for quite some time.
>>>>
>>>>
>>>
>>>
>>> _______________________________________________
>>> devel mailing list
>>> devel_at_[hidden]
>>> http://www.open-mpi.org/mailman/listinfo.cgi/devel
>>
>>
>>
>> --
>> Joshua Hursey
>> Postdoctoral Research Associate
>> Oak Ridge National Laboratory
>> http://users.nccs.gov/~jjhursey
>>
>> _______________________________________________
>> devel mailing list
>> devel_at_[hidden]
>> http://www.open-mpi.org/mailman/listinfo.cgi/devel
>
>
> _______________________________________________
> devel mailing list
> devel_at_[hidden]
> http://www.open-mpi.org/mailman/listinfo.cgi/devel

-- 
Joshua Hursey
Postdoctoral Research Associate
Oak Ridge National Laboratory
http://users.nccs.gov/~jjhursey