Very high load under spine and scripts not working (CMD OK)

[Canto]

Hello,

On monday, my cacti system (cacti version 0.8.7.e and cactid as poller) suddenly began to experience very high CPU load during polling cycles and most of the graphs using perl scripts for data collection began showing no data. I still don't know what is causing this cpu load (this linux system is dedicated to Cacti). I should also add that I had more "The POPEN timed out" messages in my cacti logs. Could it be a clue about a possible cause of the problem ?

As I couldn't find the root cause of my problem, I decided to give a try to new version of poller and cacti so I upgraded to cacti 0.8.7.g and Spine 0.8.7.g. This had not improved anything : still heavy load average during polling cycles and all my perl scripts return result as 0 :

Code: Select all

 SPINE: Poller[0] Host[717] TH[1] DS[4942] SCRIPT: /usr/bin/perl /var/www/cacti/scripts/riverbed/riverbed_traffic-0_0-generic.pl "SNMP_COMMUNITY" a.b.c.d , output: 0  

when running the same by hand, it is OK but much slower than normal because of the system load :

Code: Select all

time /usr/bin/perl /var/www/cacti/scripts/riverbed/riverbed_traffic-0_0-generic.pl  "SNMP_COMMUNITY" a.b.c.d
InLan0_0:3207129987 OutWan0_0:2216490431 OutLan0_0:297665518 InWan0_0:2332069408 
real    0m13.068s
user    0m0.487s
sys     0m0.020s

when usually the delay is around 1s ...

When switching to CMD.php as the poller, the load average is much lower than with SPINE and the scripts are correctly executed with normal results. Only drawback with CMD is that all the polling can't be done under the 300s timeframe (because of the large number of hosts/objects to poll)

Any idea ?

Thanks

[Canto]

Hi,

For your information, by applying the latest Spine 0.8.7.g patch the scripts results are now correctly returned !....
But unfortunately I am still suffering from very high load average without being able to identify the cause....