| Newsgroups |
gmane.comp.sysutils.pcp |
| Message-ID |
<[email protected]> |
Main problem is not collection, though, but visualization.
On Fri, Nov 25, 2016 at 7:52 PM Frank Ch. Eigler <[email protected]> wrote:
>
> nathans wrote:
>
> > [...]
> >> I would like to ask if where can I find some documentations for
> >> implementing into large scale setup like 3 thousand cluster nodes and a
> >> thousand of servers.
> >> [...]
>
> > Usually folk divide up the responsibilities among multiple monitoring
> hosts
> > (sometimes duplicating logging for redundancy) - a good system seems to
> be
> > having a dedicated node in each rack for monitoring the rest of the
> machines
> > in that same rack. Archives can then be pulled into a common storage
> system
> > (large) via nightly cron job or something along those lines.
>
> This advice is consistent with the observation that a single pcp
> collector installation cannot scale to thousands of servers, but to
> dozens (comfortably) or hundreds (under duress). Without a federation
> or remote-archive-access machinery, periodic copying of archive files
> or relaying data to a non-pcp data repository is necessary.
>
> - FChE
>
>
>
>