[collectd] collectd plugin-specific intervals

Bostjan Skufca bostjan at a2o.si
Thu Sep 30 02:58:38 CEST 2010

By my experience separate collectd daemons are possible. I am
compiling it from scratch and all it needs besides the installed files
(binaries etc. ) is config file, pid file and potentially unix socket
file. Now if you create 2 separate configs which both point to
/var/lib/collectd (example) but each handle it's own rrd files I
believe it should not be a problem.

Managing two separate daemons with the same name by startup scripts is
another "happy hacking hour" but it is not impossible, actually far
from it:)


PS: You can specify distinguishing config files and pid files on the
command line.

On 30 September 2010 02:46, Amos Shapira <amos.shapira at gmail.com> wrote:
> +1 from me too.
> I'd even ask for a different collection interval per instance, e.g.
> maybe a "df" change on /home or /boot is less important to check on
> every second than a "df" change on /var/log.
> While we are on this topic - until this gets implemented - would it be
> possible to work around this by having separate collectd servers for
> different intervals?
> --Amos
> On 30 September 2010 08:53, Bostjan Skufca <bostjan at a2o.si> wrote:
>> +1 for this feature
>> It could really help with IO "hell" also.
>> Nicolas: maybe you should currently apply some of the generic plugins
>> (exec, libperl, java) and collect the actual data only every 360th
>> time (if you use 10s interval and want to collect this specific item
>> only once every hour).
>> b.
>> On 29 September 2010 09:24, Nicolas Michael <email at nmichael.de> wrote:
>>> Hi,
>>> I recently came across collect and was impressed by its functionality and
>>> modular design. Great work!
>>> There is just one piece missing, which actually would be very important for me
>>> to make use of it: A possibility to specifiy intervals on a per-plugin basis.
>>> There are some counters which I need quite often (let's say, most OS statistics
>>> as cpu, network etc.), and some statistics which are rather expensive to collect
>>> and only change infrequently anyway (retrieved through some Oracle DB queries).
>>> For the first set of data I would need at least 10 second intervals, while the
>>> DB stuff should not be queried more often than once an hour. Those two intervals
>>> are so wide apart that it's impossible to find any useful "compromise" here.
>>> Browsing through your archives I've found some discussions from February this
>>> year about the exact same topic. Do you have any plans to implement such a
>>> feature in the near future (somehow generic for all plugins)? For my current
>>> purpose it might be sufficient to have such intervals for the oracle (and maybe
>>> Exec, GenericJMX and Java) plugins, but of course a generic solution would be
>>> even better.
>>> Thanks for your help,
>>> and again my respect for this great tool!
>>> Nick.
>>> _______________________________________________
>>> collectd mailing list
>>> collectd at verplant.org
>>> http://mailman.verplant.org/listinfo/collectd
>> _______________________________________________
>> collectd mailing list
>> collectd at verplant.org
>> http://mailman.verplant.org/listinfo/collectd

More information about the collectd mailing list