Hi,
On Thu, 2011-06-23 at 16:08 +0200, Massimo Sgaravatto - INFN Padova
wrote:
> Indeed I am curious to understand:
>
> a- why they had maxActive=500
This is the default value in Quattor templates.
> b- what was the value of max_connections in /etc/my.cnf
>
>
> If b < 2*a I am not surprised they saw this problem
The value for max_connections was not set in /etc/my.cnf. We have set it
to 650, and maxActive to 300.
But even after having changed these parameters, there were still error
messages in cream logs. We then reconfigured the CE with the Quattor
command ncm-ncd --co --all (this has the effect of restarting all the
services), and after that, we were back to green.
In the past, our CE was doing well with the old parameters, until we had
users submitting huge numbers of jobs per day (typically more than 10000
jobs a day). We tried to ban these users, but it didn't work (I will
post another thread on this subject.)
Maybe we need a second CE... Are there scaling rules fixing the number
of CEs needed by a site in function of the number of slots ?
Cheers,
Stéphane
>
>
> The other site I was referring to is CNAF, so not a small T2
>
>
> Cheers, Massimo
>
> On Thu, 23 Jun 2011, Maarten Litmaath wrote:
>
> > Ciao Massimo,
> >
> >> You had maxActive=500 ?
> >> The default provided with the standard conf file is 200
> >>
> >>
> >> In another site experiencing the same problem we suggested to set:
> >>
> >> [...]
> >
> > I suspect those measures only cure the symptoms of a bigger problem:
> > a small T2 should _not_ run into those limits in the first place!
> >
>
> \|||/
> -----------0oo----( o o )----oo0-------------------
> (_)
> INFN Sezione di Padova
> Via Marzolo, 8
> 35131 Padova - Italy E-mail: massimo.sgaravatto [at] pd.infn.it
> Tel: ++39 0499677360 Skype: massimo.sgaravatto
> Fax: ++39 0498275952
|