Poller runtime varying very much

Christian · Post by **Christian** » Thu Feb 14, 2008 6:18 am

Hi,

we just upgraded cacti from version 0.8.6h to 0.8.7b and everything seems to be running smoothly.

My only concern is the poller runtime which is varying very much for some reason.

Any ideas?

kyosanim · Post by **kyosanim** » Thu Feb 14, 2008 6:34 am

what poller do you use?
cmd.php or spine?

Christian · Post by **Christian** » Thu Feb 14, 2008 7:07 am

spine 0.8.7a

BSOD2600 · Post by **BSOD2600** » Thu Feb 14, 2008 2:22 pm

Look in your log file for details of whats going on. Change the logging level to medium. Are devices getting reindexed/recached during those spikes? Number of rrds changing?

Post by **gandalf** » Thu Feb 14, 2008 2:39 pm

Christian wrote:Hi,

we just upgraded cacti from version 0.8.6h to 0.8.7b and everything seems to be running smoothly.

My only concern is the poller runtime which is varying very much for some reason.

Any ideas?

How was it before? Some specs of the server would perhaps help (memory constraint?)
Reinhard

Christian · Post by **Christian** » Thu Feb 14, 2008 5:44 pm

peecee wrote:Aside from the actual times, how are you doing that graph? Is there some option I need to enable to have that created? I'd like to know my poller times as well. We're having issues with timeouts, and I am trying to figure out why. Our old system is running in parallel and is not having the problem.

I'm using ss_poller

BSOD2600 wrote:Look in your log file for details of whats going on. Change the logging level to medium. Are devices getting reindexed/recached during those spikes? Number of rrds changing?

I'll crawl through the log next monday when I'm at work again. But why should the number of rrds change? Recaching statistics say, that there's recaching occuring for 1 host or so.

gandalf wrote:How was it before? Some specs of the server would perhaps help (memory constraint?)
Reinhard

The poller runtime was constantly about 40s to 60s before I did the upgrade.

The server queries 200 hosts creating 6000 rrds. CPU is AMD Athlon(tm) 64 Processor 3200+. Memory is 1 GiB and the OS is ubuntu.

What's even more confusing me is that the server seems only to be really doing something the first 40s-60s of the polling run. During this time processor load is very high. After this time the server seems to be waiting for something and is nearly idle.

Installed plugins are Weathermap 0.941, Thold 0.3.9 and Settings 0.3, by the way. There are no weathermaps yet though.

That's all I can tell you right now. I appreciate your support and I'll give you more information next monday.

Christian

Post by **gandalf** » Fri Feb 15, 2008 10:06 am

BTB: You should set #processes to 1 or 2 (assuming there's only one core in that machine) and #threads = 10-15, not 60
Reinhard

Christian · Post by **Christian** » Mon Feb 18, 2008 8:07 am

Here are more accurate information:

The server queries 200 hosts creating 6000 rrds. CPU is AMD Athlon(tm) 64 Processor 3200+. Memory is 1 GiB and the OS is ubuntu.

I set threads to 15, but this didn't change anything.

I even couldn't detect any re-indexing going on.

During the polling process the amount of free memory is sometimes as low as 6 MiB. But I didn't notice the amount of free memory being lower at the peaks.

Additionally here are 2 lines of my log which I don't understand:

Code: Select all

02/18/2008 01:34:14 PM - SYSTEM STATS: Time:253.4111 Method:spine Processes:1 Threads:60 Hosts:207 HostsPerProcess:207 DataSources:14800 RRDsProcessed:5236
02/18/2008 01:30:28 PM - CACTID: Poller[0] Time: 26.8180 s, Threads: 60, Hosts: 207

Can somebody tell me what cacti is doing during the 4 minutes between the poller being finished and the system stats being displayed?

kyosanim · Post by **kyosanim** » Mon Feb 18, 2008 10:31 am

i think that "CACTID" in your log is not normal.
when using spine you should have only "spine" in the logfile...

correct me if i'am wrong

Christian · Post by **Christian** » Wed Feb 20, 2008 8:35 am

CACTID is spine, isn't it?

I just want to know what cacti is doing or not doing for about 3 minutes between these 2 messages.

Can a lack of physical memory be the cause and are these peaks and steps in polling times normal?

Post by **gandalf** » Wed Feb 20, 2008 11:52 am

Christian wrote:CACTID is spine, isn't it?

Incorrect. Cactid is the predecessor of spine. WIth cacti <= 086, use cactid. WIth cacti >= 087, use spine. Don't mix them up
Reinhard

Christian · Post by **Christian** » Thu Feb 21, 2008 3:24 am

Oh great. Now you've got me confused

So am I correct that I've got 2 pollers running simultaneously?
If so, what could possibly be the cause for this? I've only got one cron job for cacti.

Post by **gandalf** » Thu Feb 21, 2008 3:52 pm

Christian wrote:Oh great. Now you've got me confused

So am I correct that I've got 2 pollers running simultaneously?

Why? with 086, use cactid.
WIth 087, use spine.
You can only select one from Settings->Poller
Reinhard

Christian · Post by **Christian** » Fri Feb 22, 2008 3:06 am

Sorry, I just saw in DEBUG log that I forgot to change the path to the new spine version.

Sorry for the inconvenience.

The spikes stille aren't gone though, polling times are the same.

Cacti

Poller runtime varying very much

Poller runtime varying very much

Re: Poller runtime varying very much

Who is online