[collectd] How to store MySQL statistics?

[Jan Huijsmans]

On Tue, Dec 20, 2005 at 05:14:49PM +0100, Niki Waibel wrote:
> > You'll get a lot of files this way, but hey, unix should be able to cope
> > with them.
> well, if someone have > 20000 users, > 1000 groups it could
> be some sort of a problem. not for the filesystem, but for the
> user...
> 
> also imagine you have about 20 mountpoints in addition ...

Ahhhh.. that's a lot... Especially when there are still sysadmins
around who aren't able to delete files in a dairectory when rm *
results in a too long argument list.

> > To differentiate between uiser or group id I would use the type
> > field,
> 
> yes, that could be done. sthg like
>         quota-u100-niki-_.rrd
>         quota-g500-users-_.rrd
> but for me it is more logically sorted in the fs
> if the actual name would come first...
> but
>         quota-uniki-100-_.rrd
>         quota-gusers-500-_.rrd
> is not very nice.

As sysadmin I would prefer the first option (id - name), as sysadmins
usually ignore names. (and u100-niki is easier to interpret when
looking at the files, human nature)

> also you could get rid of either the name (niki) or the id (100).
> but i thought that i leave it there. on some sites it could be that 
> users names are deleated and overwritten... so it is good imho to
> have both fealds there.

I would suspect those site to leave a uid unused for a couple of days
to prevent new users having access to the old user's data.

> > but I'm still missing the use of the name field. The name of the
> > group/user can be found via systemcalls
> 
> yes.
> 
> > or in plain files
> 
> yes mean /etc/passwd, /etc/groups ???
> that is not a good way to get that information.
>         a) no caching
>         b) what about dns, nis, nis+, ldap environments?
> i think that the system calls have to be used.

Yes, you're right.

> > while processing
> > the rrddata instead of during data gathering. (or at least, it looks like it
> > would simplify the gathering process a bit)
> 
> would it? in which function would you recommend to place the
> id -> name
> translation?

None. For the data gathering there is no need for the mapping.
(or is quota unable to report by just id)

> are you (Jan Huijsmans) interested in the quota-plugin as well?
> also for large-scale companies?

Me, nah... I'm just interested in the programming side and the
arguments used for the decissions made.

I'm currently working at a pretty large company (for the netherlands),
but I've just started to use collectd myself and for my own company.
I'm thinking of using quotas on the systems of my company, but I currently
have less then 50 users. I hope to scale up someday, but to be honest,
I'm not seeing it happen...

---

Jan Huijsmans              bofh at koffie.nu

... cannot activate /dev/brain, no response from main coffee server