Welcome! Log In Create A New Profile

Advanced

100% cpu usage after reload (USR2) with "bind [email protected]"

Posted by Dennis Jacobfeuerborn 
Dennis Jacobfeuerborn
100% cpu usage after reload (USR2) with "bind [email protected]"
March 05, 2018 09:10PM
Hi,
today I started experimenting with the HAProxy 1.8.4 release and ran
into a Problem when it comes to reloading the configuration (USR2).
I'm running the release tarball via RPM on a CentOS 7 System in
master-worker mode and every time I perform a "systemctl reload" the
worker process suddenly uses 100% cpu and can only be killed via "kill -9".
Once I do so the old processes go away and two new ones show up and
everything goes back to normal.

After reducing the config bit by bit I was able to isolate the config
that is causing this:

bind [email protected] accept-proxy

Once I replace the "[email protected]" bit with an ip:port combination
the problem goes away so it seems the "[email protected]" part is the culprit here.

Regards,
Dennis
Dennis Jacobfeuerborn
Re: 100% cpu usage after reload (USR2) with "bind [email protected]"
March 10, 2018 04:20PM
On 05.03.2018 21:06, Dennis Jacobfeuerborn wrote:
> Hi,
> today I started experimenting with the HAProxy 1.8.4 release and ran
> into a Problem when it comes to reloading the configuration (USR2).
> I'm running the release tarball via RPM on a CentOS 7 System in
> master-worker mode and every time I perform a "systemctl reload" the
> worker process suddenly uses 100% cpu and can only be killed via "kill -9".
> Once I do so the old processes go away and two new ones show up and
> everything goes back to normal.
>
> After reducing the config bit by bit I was able to isolate the config
> that is causing this:
>
> bind [email protected] accept-proxy
>
> Once I replace the "[email protected]" bit with an ip:port combination
> the problem goes away so it seems the "[email protected]" part is the culprit here.

BTW what are the implications from using sockets over abns? Is there a
possibility of TCP socket exhaustion between the two frontends?

Regards,
Dennis
Le 05/03/2018 à 21:06, Dennis Jacobfeuerborn a écrit :
> Hi,
> today I started experimenting with the HAProxy 1.8.4 release and ran
> into a Problem when it comes to reloading the configuration (USR2).
> I'm running the release tarball via RPM on a CentOS 7 System in
> master-worker mode and every time I perform a "systemctl reload" the
> worker process suddenly uses 100% cpu and can only be killed via "kill -9".
> Once I do so the old processes go away and two new ones show up and
> everything goes back to normal.
>
> After reducing the config bit by bit I was able to isolate the config
> that is causing this:
>
> bind [email protected] accept-proxy
>
> Once I replace the "[email protected]" bit with an ip:port combination
> the problem goes away so it seems the "[email protected]" part is the culprit here.
>
> Regards,
> Dennis
>

Hi Dennis,

A recent commit could fix your problem:

http://git.haproxy.org/?p=haproxy.git;a=commitdiff;h=ec9516a6

It was pushed in upstream and will be backported in 1.8. Could you check
if it works please ?

--
Christopher Faulet
Le 12/03/2018 à 09:48, Christopher Faulet a écrit :
> Le 05/03/2018 à 21:06, Dennis Jacobfeuerborn a écrit :
>> Hi,
>> today I started experimenting with the HAProxy 1.8.4 release and ran
>> into a Problem when it comes to reloading the configuration (USR2).
>> I'm running the release tarball via RPM on a CentOS 7 System in
>> master-worker mode and every time I perform a "systemctl reload" the
>> worker process suddenly uses 100% cpu and can only be killed via "kill -9".
>> Once I do so the old processes go away and two new ones show up and
>> everything goes back to normal.
>>
>> After reducing the config bit by bit I was able to isolate the config
>> that is causing this:
>>
>> bind [email protected] accept-proxy
>>
>> Once I replace the "[email protected]" bit with an ip:port combination
>> the problem goes away so it seems the "[email protected]" part is the culprit here.
>>
>> Regards,
>> Dennis
>>
>
> Hi Dennis,
>
> A recent commit could fix your problem:
>
> http://git.haproxy.org/?p=haproxy.git;a=commitdiff;h=ec9516a6
>
> It was pushed in upstream and will be backported in 1.8. Could you check
> if it works please ?
>

Hi,

I finally found some time to investigate this issue and found an obvious
bug. Here is the patch.

Willy, unless you see something wrong in my patch, I think you can
safely merge it right now.

Thanks,
--
Christopher Faulet
On Fri, Mar 16, 2018 at 10:44:22AM +0100, Christopher Faulet wrote:
> Willy, unless you see something wrong in my patch, I think you can safely
> merge it right now.

OK, merged now. Thanks!
Willy
Sorry, only registered users may post in this forum.

Click here to login