Good day everyone,

I have been noticing a odd problem with ganglia that I hope someone can help 
with. I was unable to find anything in the mailing list.

The short of it is that ganglia will randomly report all the nodes as dead, yet 
it will still show that it is updating the heartbeats.
(link to imageshack for screenshot)
http://img231.imageshack.us/img231/2910/snapshot18si.jpg

This results in my graphs having gaps.
http://img231.imageshack.us/img231/3770/snapshot26sh.jpg

While random, this will happen several times a day. Sometimes the gap is so 
small that it will not show on the day view, sometimes it will be upwards of 
ten minutes. Occasionally I will get the error message (usually a few days in 
between that I notice) :

Warning: arsort() expects parameters 1 to be array, null given in 
/var/www/html/ganglia/cluster_view.php on line 95

which is this line in the code
switch ($sort)
{
        case "descending":
->              arsort($sorted_hosts);
                break;
        case "by hostname":
                ksort($sorted_hosts);
                break;
        default:
        case "ascending":
                asort($sorted_hosts);
                break;
}

Now for the part that really confuses me. I have a second system monitoring 
other computers; same ganglia version and gmond (3.0.3 ). I have set both of 
them up so that they will also monitor the other group. Group A has these gaps 
on its nodes, but will not have them when viewing group B. Group B monitor does 
not have the gaps at all.

If it was one group of systems, I would look at the systems. If one of the 
monitors was doing it to both groups, I would look at the monitor as the 
problem. But it is just one monitor with one group and I have no idea what's 
going on.

I have already tried boosting ganglia's priority, and making certain the two 
groups were on different ports just in case there was a cpu/network flood 
causing it. To me, if it is still able to update the heartbeat then it "sees" 
the other systems. Why then report them as dead? (which is why, I assume, the 
graphs have gaps)

I would appreciate any comments and suggestions.
Thanks everyone,
Chris Stackpole

Reply via email to