On Wed, 2009-08-12 at 08:56 -0400, Alan Conway wrote:
> On 08/12/2009 07:25 AM, Gordon Sim wrote:
> > The output from make check-long is now showing lots of warnings of the
> > following form:
> >
> > soak-1: 2009-08-12 06:40:56 warning Timer woken up 3ms late
> >
> > Is this a serious issue? If so what do we need to do to prevent it
> > happening? If not, is warning the appropriate logging level?
> >
> >
> 
> I've also been seeing a lot of those messages since sometime last week. As it 
> stands its not a useful message at any log level so I'd be inclined to remove 
> it 
> or else build in some fudge factor so it only reports if the Timer is 
> significantly late. Hard to guess what's a good value for "significant".

I think that these messages are less important than the overrun
messages, but useful in retrospect - to understand that a timing delay
isn't the result of anything happening in qpidd, but caused by the
machine being so busy (for whatever reason) that the kernel couldn't
wake the timer thread up in time.

If you are seeing lots and lots of these messages it almost certainly
means that the io threads (or I guess another process entirely) are
monopolising the cpus to the extent that the timing thread can't be
restarted, and this is perhaps important in itself.

Currently the notion of significant is 1ms for active Timer tasks and
500ms for cancelled timer tasks. The significance of a delay will
strongly depend on what time dependent tasks are going on so it's
difficult to know currently the heartbeat related tasks are the most
time sensitive and they fail after delays of about 0.5 * the heartbeat
interval.

As it currently stands we could probably increase the limit to 50ms and
not miss really significant delays.

Thoughts?

Andrew




> 


---------------------------------------------------------------------
Apache Qpid - AMQP Messaging Implementation
Project:      http://qpid.apache.org
Use/Interact: mailto:[email protected]

Reply via email to