Cacti-0.8.3, Solaris 9 (Works like a champ!)
However...
Is there a way to determine the order in which Cacti polls devices? Or, even better, is there a way to tell Cacti to skip devices it can't reach after n-number of seconds?
I encountered an interesting situation this evening: I have 2 graph trees defined, one for a Nortel Passport switch and one for a Cisco 6513. The Passport is listed first in the Graph Trees section and I assume that this results in it being probed first.
During this evening, the Passport was down for a 30 minute span and, afterward, I found a big gap in ALL the graphs within Cacti, not only the graphs for the networks beyond the Passport. I looked at the cron logs and the rrd.log and, sure enough, I found that there is a 30 minute gap in the rrd.log. The cron log shows that it kicks off every 5 minutes during this span but never completes the process.
The only thing I can gather is that Cacti attempted to scan devices on the Passport and eventually timed out before even proceeding to scan the devices on the 6513.
It may be worth mentioning that the server running Cacti is directly attached to the 6513 and reaches the Passport through the 6513.
What are my options?
Sincerest thanks for all your assistance.
Regards,
-Samson
Polling Sequence and Priorities
Moderators: Developers, Moderators
I just noticed the same thing tonight with some of my polling hosts (windows boxes)... if one of those is down, polling is severely slowed... I'm using cactid so perhaps it could be modified to hand off polling tasks from the queue to whatever thread is available...
I'm assuming the way it works now is to hand out a list of polling tasks to each thread... so if one task in a thread gets stuck, the remaining tasks in that thread are screwed... I could be way off though...
I'm assuming the way it works now is to hand out a list of polling tasks to each thread... so if one task in a thread gets stuck, the remaining tasks in that thread are screwed... I could be way off though...
If you use cmd.php for polling you may try to customize it a bit to skip non responsive hosts. I've put an idea of such change here:
http://www.raxnet.net/board/viewtopic.php?t=2073
For cactid similar solution can applied which checks if device is responding before polling all SNMP objects from it.
- bulek
http://www.raxnet.net/board/viewtopic.php?t=2073
For cactid similar solution can applied which checks if device is responding before polling all SNMP objects from it.
- bulek
Who is online
Users browsing this forum: macan and 4 guests