Mailing List Archive

x x wrote:
>
> I use Redhat 6.2, following the high availability to send primary server and
> backup server for telnet and www services.
> However, when I use "ab" coming from apache to make a benchmark test, with
> command:
> ab -n 100000 -c 200 "URL".
> The heartbeat DIES. (I use ps to check the process heartbeat processes; all
> are gone). At the same time, all redirect rules is cleared because heartbeat
> dies.
> I would like to know how to solve this problem or any good suggestion.

I suspect that if you've run it under heavy load, that the system has
failed to hear it's own heart beat. When it does that, it shuts itself
down, because that basically indicates a heartbeat failure. This
problem has been observed before. Since you said that all processes
died, and it dropped the alias, I suspect that this is indeed the
problem, since it is unlikely that both of these would happen otherwise.

The fix to this problem is to lock heartbeat into memory, and raise its
priority, so it will always run. The code to do this is now in version
0.4.7b - a beta release. If you want to test it, that would be GREAT!

Holger Kiehl also had the same problem. I haven't announced this
release, because I forgot to ;-)

-- Alan Robertson
alanr@suse.com

> The fix to this problem is to lock heartbeat into memory, and raise its
> priority, so it will always run. The code to do this is now in version
> 0.4.7b - a beta release. If you want to test it, that would be GREAT!

The Linux soft real time makes it more likely, not a guarantee. Actually
with 2.2 its not too bad but with 2.3.x we are currently seeing some
600mS delay cases.

Alan Cox wrote:
>
> > The fix to this problem is to lock heartbeat into memory, and raise its
> > priority, so it will always run. The code to do this is now in version
> > 0.4.7b - a beta release. If you want to test it, that would be GREAT!
>
> The Linux soft real time makes it more likely, not a guarantee. Actually
> with 2.2 its not too bad but with 2.3.x we are currently seeing some
> 600mS delay cases.

Thanks Alan!

Understood. The minimum dead time configurable by heartbeat is about 3
sec. I think the default is 5 sec. Even 600ms should allow this to
function smoothly under most circumstances

If you're going to load your machine that heavily, you might not be able
to run the dead time down to 3 sec.

-- Alan Robertson
alanr@suse.com