[sheepdog] [PATCH v2 09/10] sheep: add event_loop checker

Liu Yuan namei.unix at gmail.com
Fri Aug 9 06:42:06 CEST 2013


On Fri, Aug 09, 2013 at 01:38:26PM +0900, MORITA Kazutaka wrote:
> At Fri, 9 Aug 2013 12:26:22 +0800,
> Liu Yuan wrote:
> > 
> > On Fri, Aug 09, 2013 at 12:22:03PM +0900, MORITA Kazutaka wrote:
> > > From: MORITA Kazutaka <morita.kazutaka at lab.ntt.co.jp>
> > > 
> > > We shouldn't spend too much time in the main thread.  This checker
> > > prints a warning message when sheep uses more than 1 ms in the main
> > > thread.
> > 
> > I can just get
> > 
> > Aug 09 11:49:14 [main] event_handler_exit(45) timer_handler wastes too much time in event loop: 1506.979 us
> > 
> > notice, any chance that we include who is the culprit?
> 
> Enabling graph tracer and reproducing the problem is not enough?
> Perhaps, it'd be better to check all the function internal
> event_loop().

Ah, yes.

> I've also noticed that event_loop() can be a bottleneck when we
> requests a large data like 4 MB.  I think of focusing on the problem
> in the next development cycle.

I remember I have an old patch set that offload data rx/tx to workers, I think
we can revisit this patchset later.

Thanks
Yuan



More information about the sheepdog mailing list