Using Spine & 'Balance Process Load' causes Spine to max CPU

[mrquad450]

Running Cacti 1.02 and Spine 1.02 on Linux debian 3.16.0-4-686-pae #1 SMP Debian 3.16.39-1 (2016-12-30) i686 GNU/Linux - virtual on vmware - has been for well over a year.

This was an existing, working Cacti server upgraded from 0.8.8b to v1.02 over the past couple of days.

I enabled Spine and enabled the 'Balance Process Load' on the Poller settings page.

This is a very small deployment, only:
2017-02-14 13:20:03 - SYSTEM STATS: Time:2.3878 Method:spine Processes:2 Threads:8 Hosts:18 HostsPerProcess:9 DataSources:205 RRDsProcessed:109

After enabling, was getting tons of these logs:
2017-02-14 13:12:00 - SPINE: Poller[1] ERROR: Spine Timed Out While Processing Devices Internal

'top' showed two Spine processes - combined were using 200% CPU usage - maxed out the virtual limit of 7k mhz use.

I'm willing to spin up another instance and set it up in the broken state again if needed to gather strace or similar. Zero RRD's were being processed - the spine poller simply hung and timed out before being killed off.

I've added a CPU chart showing the drastic difference on the CPU - the latter third of the graph is after disabling the 'Balance Process Load'.

Thanks,
John

[JMoMo]

I have seen similar problems with Spine while testing Cacti 1.0.x. I had to give up trying to use Spine and went back to the php poller just to keep testing.

Performance in Cacti 1.0 seems to be generally much worse than the old 0.8.8x, based on my experience of the last two weeks. I don't think the problem is in Spine specifically, but it's causing the poller to fail.

The 1.0 release seems to have been a low-quality rush job.

[maxvel]

Hello guys. I've been using your product since 2007. But the last update(1.1) brought a lot of problems. The system worked day after I updated from 1.0.6 to 1.1. There were such errors in the logs
Poller[1] ERROR: Spine Timed Out While Processing Devices Internal. My cpu was loaded at 800%. My all graphics stopped drawing.
I have:
CentOS release 6.8 (Final)
Intel(R) Xeon(R) CPU E5420 @ 2.50GHz
Graphs 21,371

To solve the problem, I again returned to version 1.0.6

[Osiris]

Running Cacti 1.02 and Spine 1.02 on Linux debian 3.16.0-4-686-pae #1 SMP Debian 3.16.39-1 (2016-12-30) i686 GNU/Linux - virtual on vmware - has been for well over a year.

This was an existing, working Cacti server upgraded from 0.8.8b to v1.02 over the past couple of days.

I enabled Spine and enabled the 'Balance Process Load' on the Poller settings page.

This is a very small deployment, only:
2017-02-14 13:20:03 - SYSTEM STATS: Time:2.3878 Method:spine Processes:2 Threads:8 Hosts:18 HostsPerProcess:9 DataSources:205 RRDsProcessed:109

After enabling, was getting tons of these logs:
2017-02-14 13:12:00 - SPINE: Poller[1] ERROR: Spine Timed Out While Processing Devices Internal

'top' showed two Spine processes - combined were using 200% CPU usage - maxed out the virtual limit of 7k mhz use.

I'm willing to spin up another instance and set it up in the broken state again if needed to gather strace or similar. Zero RRD's were being processed - the spine poller simply hung and timed out before being killed off.

I've added a CPU chart showing the drastic difference on the CPU - the latter third of the graph is after disabling the 'Balance Process Load'.

Thanks,
John

That is very useful information. Can you please log that to GitHub?