Re: Spine and Poller run times (is this good or bad?)

On 23.08.2013 11:33, James Bensley wrote:
> Hi all,
> 
> I am wondering if there is something wrong with my Cacti installation
> or if this is normal:
> 
>>From the logs, when the poller runs (I am using spine, I have compiled
> the latest version) everything looks good to me;
> 
> 4554:08/21/2013 04:20:04 PM - SPINE: Poller[0] Time: 1.0897 s,
> Threads: 2, Hosts: 3
> 5391:08/21/2013 04:20:05 PM - SPINE: Poller[0] Time: 1.2657 s,
> Threads: 2, Hosts: 6
> 5579:08/21/2013 04:20:05 PM - SPINE: Poller[0] Time: 1.3529 s,
> Threads: 2, Hosts: 7
> 5610:08/21/2013 04:20:05 PM - SPINE: Poller[0] Time: 1.4433 s,
> Threads: 2, Hosts: 8
> 7586:08/21/2013 04:20:06 PM - SPINE: Poller[0] Time: 3.1098 s,
> Threads: 2, Hosts: 29
> 8435:08/21/2013 04:20:08 PM - SPINE: Poller[0] Time: 4.6443 s,
> Threads: 2, Hosts: 35
> 9483:08/21/2013 04:20:12 PM - SPINE: Poller[0] Time: 9.0247 s,
> Threads: 2, Hosts: 29
> 10015:08/21/2013 04:20:15 PM - SPINE: Poller[0] Time: 11.6772 s,
> Threads: 2, Hosts: 51
> 10017:08/21/2013 04:20:23 PM - SYSTEM STATS: Time:321.6814
> Method:spine Processes:10 Threads:2 Hosts:188 HostsPerProcess:19
> DataSources:9617 RRDsProcessed:4378
> 10022:08/21/2013 04:20:33 PM - SPINE: Poller[0] Time: 29.3343 s,
> Threads: 2, Hosts: 28
> 
> So that's 188 hosts polled and 4.3k~ RRDs were updated in about 30
> seconds. That seems very good to me in fact!
That's not unusual.

> 
> However looking here: /utilities.php?action=view_tech I can see that
> the run time is very long;
> Last Run Statistics    Time:388.4284 Method:spine......
Hmm, that does not match the other stats. It should take the last runtime from
SYSTEM STATS.

> 
> Like most people I am polling every 5 minutes. Watching the 5 minute
> polls running on the CLI via htop and top, I can see that my spine
> processes start up, the CPU on the server shoots up to 100% for about
> 30 seconds, then all the spine processes exit, and the server drops to
> %1~ usage and all is quiet. Yet I can always see that the poller.php
> is still running until a little while into the next 5 minute poll
> event. Why is the PHP script still running, after spine has finished?
> Is the server still busy "processing" the data spine received i.e.
> putting it into RRDs, updating MySQL etc? There is disk I/O after
> spine has exited.
The real _polling_ does not last very long. But updating rrd files depends on
the IO speed of your system. As many rrds are updated, many seeks are required.
There's an article on using fadvise to at least suppress read ahead caching
which does not make sense with cacti.
A good move is to put mysql db and rrd files on different file systems/drives.
If update time is still too long, using BOOST plugin will help (we currently do
2,000,000 updates for 100,000 data sources in about 3 minutes using BOOST).

R.

> 
> I am running Cacti 0.8.7g but tonight I plan to upgrade to 0.8.8b.
> 
> Many thanks,
> James,

------------------------------------------------------------------------------
Introducing Performance Central, a new site from SourceForge and 
AppDynamics. Performance Central is your source for news, insights, 
analysis and resources for efficient Application Performance Management. 
Visit us today!
http://pubads.g.doubleclick.net/gampad/clk?id=48897511&iu=/4140/ostg.clktrk