Re: PROBLEM: Epoll frequently fails to notify connects at connectbursts

[Davide Libenzi]

On Tue, 25 Aug 2009, Davide Libenzi wrote:

> On Wed, 26 Aug 2009, Will Brown wrote:
> 
> > PLEASE CC: ME ON ANY REPLIES, thank you.
> > 
> > *** 1. Epoll frequently fails to notify connects at connect bursts
> > 
> > *** 2. Sometimes, when several connects come in quickly, epoll
> > establishes tcp connections (according to 'netstat') but then fails to
> > notify the application by returning from epoll_wait(). In those cases
> > a subsequent connection or an EPOLLIN or EPOLLOUT event can cause a
> > return from epoll_wait(), which then causes the previously ignored
> > connection to be reported. But then the more recent event is not
> > notified, the whole sequence lags and everything gets out of whack.
> > The problems can easily be reproduced on my Ubuntu machine which runs
> > a kernel version 2.6.28 using supplied test program. I also have a
> > Ubuntu hosted in VMWare running a 2.6.27 version kernel, which appears
> > to NOT have the same problem.
> 
> Multiple quasi-simultaneous events for the same fd are merged. This is why 
> on POLLIN on accepting sockets you want to do something like:
> 
> 	while ((afd = accept(sfd, ...)) != -1)
> 		add_client(afd);

Of course, with  sfd  in non-blocking mode. Alternatively, you drop EPOLLET 
for the  sfd  socket.


- Davide


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/