Hi Malcolm,
On Tue, Jun 11, 2013 at 12:01:25AM +0000, Malcolm Handley wrote:
> That's sound advice. We'll upgrade soon and see whether anything changes.
>
> Is the latest snapshot really what you recommend for running in production? I
> was nervous about using a dev release but we desperately needed support for
> the proxy protocol.
You're right to feel nervous :-)
That said, Lukas is right on this point, and the issue you're talking about
was fixed in dev18 :
commit cf181c9d404815f890da7cd2243a5528edd7b4f9
Author: Willy Tarreau <w...@1wt.eu>
Date: Fri Jan 18 15:22:41 2013 +0100
BUG/MINOR: epoll: use a fix maxevents argument in epoll_wait()
epoll_wait() takes a number of returned events, not the number of
fds to consider. We must not pass it the number of the smallest fd,
as it leads to value zero being used, which is invalid in epoll_wait().
The effect may sometimes be observed with peers sections trying to
connect and causing 2-seconds CPU loops upon a soft reload because
epoll_wait() immediately returns -1 EINVAL instead of waiting for the
timeout to happen.
This fix should be backported to 1.4 too (into ev_epoll and ev_sepoll).
The fix was backported into 1.4.23 as well, so you can use it with Cyril's
proxy protocol patch. BTW, I seem to have finished merging fixes, so I'll
soon issue a dev19 and 1.4.24.
Best regards,
Willy
>
>
> On Fri, Jun 07, 2013 at 03:16 PM, Lukas Tribus <luky...@hotmail.com> wrote:
> > Hi Malcolm,
> >
> >
> > > This works but we find that the new haproxy process uses a lot of cpu
> > > (100% of one core) for about 20 seconds after the restart. During this
> > > time it looks as if various queues fill up and haproxy logs fewer
> > > requests than normal. Once the cpu load drops we get a surge of
> > > requests (which cause a spike in connections to our db, and a raft of
> > > other problems if we do this during heavy traffic).
> >
> > I strongly suggest to update your code to a more recent release, there
> > have been a lot of bug fixes since dev15.
> >
> > Grab dev18 or better yet, the latest snapshot.
> >
> >
> > It doesn't make sense to start troubleshooting on those obsolete releases,
> > especially in the development branch.
> >
> >
> > Lukas
