Is it possible to offload spine polling to additional server

[smaug]

Hello;

I would like to offload some of the polling activity on my current cacti host. I am using cacti 0.8.8b and spine 0.8.8c for polling.

The cron entry for the existing poller is:
* * * * * cacti /usr/bin/php /usr/share/cacti/poller.php > /dev/null 2>&1

Is it possible for me to replace this with cron entries on a cluster of machines used only for spine polling? The spine config file on all machines will reference the same mysql database. I am hoping that I can use the spine --first and --last parameters to divide the polling work evenly across all the machines in the cluster.

For example, say the range of host ids in the cacti database range from 1-100. I would like to split the work with cron entries as follows:

machine a:
* * * * * cacti /usr/bin/spine --first=1 --last=50 > /dev/null 2>&1
machine b:
* * * * * cacti /usr/bin/spine --first=51 --last=100 > /dev/null 2>&1

Does anyone have experience with this? If so, are any changes required to the configuration options on the Settings/Poller page?

Thanks

[BSOD2600]

In theory, yes you can run spine from a different server. Which type of performance issues you running into? Using Boost plugin yet?

[smaug]

Hi BSOD2600;


Thanks for the response. The type of performance issues are:
- primarily the cpu is getting pegged every time spine launches; and
- secondarily, there are I/O waits on the disk storing the rrd files.

For the short-term, I will not have the opportunity to upgrade this hardware. If I can get concurrent spine pollers on multiple machines, it will alleviate the problem in the short-term.

I am going to have the opportunity to move the rrd files to a NAS device which all the spine pollers can access. This is part of a larger plan to accommodate about double the number of devices monitored by cacti. If I can get a cluster of spine pollers working on the existing devices, I have a migration path to double the coverage by cacti.

I am not using Boost yet. I decided not to try Boost until I have a machine with more memory.

I'm still interested in opinions regarding whether or not I need to change the configuration options in the Settings/Poller page and whether or not the changing the cron entry to spine with --first and --last is all that I need to do.

Thanks

[BSOD2600]

Boost will certainly help with the I/O issue. Consider giving it a try even with the current memory level.

What are the spine settings currently configured?

There are a few on the forum who have documented their non-standard spine clustered configurations. It's not officially supported however.