[ovs-dev] master: Unreasonably long 16518ms poll interval

Flavio Leitner fbl at redhat.com
Wed May 28 23:19:33 UTC 2014


On Wed, May 28, 2014 at 03:47:00PM -0700, Alex Wang wrote:
> Hey Flavio,
> 
> 
> Here is the detailed root cause:
> 
> For revalidator threads, there is one leader thread that calls poll_block()
> regularly.
> Other Non-leader threads do not call poll_block() in their main loop, and
> instead
> synchronizes with leader thread by waiting on the barrier.
> 
> However, non-leader revalidators could call poll_block() in
> nl_sock_recv__() due
> to block on reading flow dump from kernel.  This can happen fairly
> frequently
> when there are very few flows.
> 
> The execution of poll_block() will check the last_wakeup time, and
> calculate the
> poll interval (time_msec() - last_wakeup).  if the poll interval is greater
> than 1000 ms,
> a warning you reported is issued.
> 
> 
> 
> Sorry for the confusion, this should be fixed.  We'll discuss a solution
> and post a fix soon.

Ok, thanks a lot Alex for the detailed explanation.
I see no real issue besides that message, so no rush.

fbl




More information about the dev mailing list