big pik on all servers?

vmicovic · Post by **vmicovic** » Thu Oct 25, 2007 9:44 am

Hello,

I have couple servers on one cacti.
Problem is, something in some times cacti say that ALL servers in SAME time have pik about 80% load of server.

it is false right?
how to fix that bug?

i put in attachment, too see what i mean ..

tnx. For advice and sorry for bad English.

Post by **Linegod** » Thu Oct 25, 2007 11:30 am

I'm going to guess that your servers are doing a logrotate at midnight. Either that, or a backup.

I'd cross compare it with the CPU load, and see if you see a similar spike, then check logrotates and cronjobs to see what is running at that time.

But, my first assumption would be that this is not a bug or error.

Post by **rony** » Thu Oct 25, 2007 11:31 am

What graph template is that?

vmicovic · Post by **vmicovic** » Thu Oct 25, 2007 12:54 pm

hmm, nope...
this hapent random...
now is @ night..

but in early logs i get same peaks (not this big) @ day ..
i look first @ cron jobs or scripts which runings.. all crons and scripts run evry morning @ 4:20

i dont know why this happent ...

and another servers dont have save services ..

i belive that is, IF server where is cacti serv. setup, that he make some big load and he put on all servers ... but, maybe mistake ...

how to chek and resolve this?

Post by **TheWitness** » Thu Oct 25, 2007 9:15 pm

Simply establish the trend and then login to the box during that time and provide some feedback as to what you find. If you have something like MacTrack installed, it could be database maintenance causing an over subscription of collector processes.

TheWitness

Howie · Post by **Howie** » Fri Oct 26, 2007 2:38 am

rony wrote:What graph template is that?

Isn't it the default load average template where the 3 averages are in a STACK (i.e. useless)? ucd/net - Load Average

That graph is *really* showing a load of 10 during the peak.

Looks to me like cron.daily or something similar since it starts at midnight on all servers. Or a backup process?

vmicovic · Post by **vmicovic** » Fri Oct 26, 2007 9:25 am

nope, all options of crontabs or something like that is imposible ..
because this happent on same server which is on windows and on server who dont have any services ...

something is wrong with cacti.

Post by **TheWitness** » Fri Oct 26, 2007 9:29 am

Ok, so confirm the time that this happens and then if it is consistent as to when it happens, post the processes that are causing the high load. Ball is in your court.

TheWitness

Howie · Post by **Howie** » Fri Oct 26, 2007 9:31 am

Actually which graph template is it?

Is it "Unix - Load Average" or "ucd/net - Load Average"? Because the first one ALWAYS measures the load on the local (cacti) server. The output from those two templates look the same, so it would be an easy mistake to make.

Is it possible you have a graph with that template attached to each host? The reason they are all the same, is that they *are* all the same, in that case.

vmicovic · Post by **vmicovic** » Fri Oct 26, 2007 4:41 pm

u have right.
i think the now where is the problem.

before about 2hours, my server where is cacti, have load about 10.0 and in all another graph he make same peaks. (because of that serv. where is cacti)

i think that cacti on my server where is he, a little bugy ...

and this is "Unix - Load Average"

did anyone have problems like this?

and one more think.
when i create new host, i put for hostname "local linux machine" is this right or i must to put somethink else (like SNMP-enabled generic host)?

Post by **TheWitness** » Fri Oct 26, 2007 6:34 pm

The unix graph templates only graph the local host. If you apply them to other devices, you will still get graphs for the localhost. I missed this completely. Howie hit the nail on the head.

TheWitness

vmicovic · Post by **vmicovic** » Sat Oct 27, 2007 10:09 am

ok, now just pls. tell me which "host" to put when i adding new linux host?

none
generic snmp
ucd/net snmp host

?

tnx.

Howie · Post by **Howie** » Sat Oct 27, 2007 11:13 am

vmicovic wrote:ok, now just pls. tell me which "host" to put when i adding new linux host?

none
generic snmp
ucd/net snmp host

?

tnx.

ucd/net host, as long as your linux hosts have net-snmp installed.

vmicovic · Post by **vmicovic** » Sat Oct 27, 2007 4:32 pm

all right, i put new graph for same serv.
Now graph its totaly difrent than anothers ..

one more question
this 10 in graph, is 10.0 load or 1.0?

tnx.

Howie · Post by **Howie** » Sat Oct 27, 2007 5:34 pm

It's 4 load average (top of the yellow part) over 1 minute.

Be careful with this graph - it shows the 1min, 5min and 15min load averages as a stack - all on top of each other. That is a meaningless thing - they don't add up to anything useful.

To get useful numbers from the graph, you should edit the graph template to change the AREA and STACKs into LINE1s or LINE2s. The STACK looks cool, but it doesn't mean anything. As it is, all it shows you is when it is busy against when it is quiet.

Cacti

big pik on all servers?

big pik on all servers?

Who is online