Processor Load not displaying

[jojo]

I have this issue, i'm running zabbix on a centos box and it's been performing quite well. We're testing it out, and we have a few hundred machines to monitor.

Now one thing i noticed today is that the graphs stopped displaying processor load info all of a sudden. By default it displays the last hours worth of data, so i changed it to display the last two hours, still nothing. I tried looking at 4 hours worth of data on the graph and BOOM! all the data was actually there!

The php memory usage is set to 32MB. anyone else experienced this problem?

[dantheman]

Which version of zabbix are you running?

[jojo]

sorry, forgot to put that in, it's zabbix server version 1.1.1

[pdwalker]

I've seen that problem a number of times with my processor graphs. It comes and goes. No idea why it happens with just the two hour data and not the 4 or 1 hour data.

If I can make it consistently reproducible, I'll send the details.

- Paul

[jojo]

OK over the last few days, i've relocated our zabbix server from a VMware server to an actual box running CentOS release 4.3 (Final) with 1GB ram with dual PIII 700 CPU's.

we're now running zabbix server version 1.1.2 and it's having that problem.

These are the packages that are installed on the machine:

rpm -qa | grep httpd:

httpd-suexec-2.0.52-22.ent.centos4
httpd-2.0.52-22.ent.centos4

rpm -qa | grep php:

php-pear-4.3.9-3.9
php-4.3.9-3.9
php-mysql-4.3.9-3.9
php-gd-4.3.9-3.9

rpm -qa | grep mysql:

mysql-server-4.1.12-3.RHEL4.1
mysql-4.1.12-3.RHEL4.1
php-mysql-4.3.9-3.9
mysqlclient10-3.23.58-4.RHEL4.1
mysql-devel-4.1.12-3.RHEL4.1

We did a database dump from the old database and imported it all into the new server and everything is up and running. Both servers were using mysql databases.

The majority of our users that are monitoring processor load have to change the period to display 3 hours worth of data or more.

My php.ini file is set to use 64M which worked with the previous server:
memory_limit = 64M ; Maximum amount of memory a script may consume (8MB)

i've tried upping this number to as far as 512 and i've even tried it with different version'd frontends to see if that makes a difference.

Here is the item that i'm monitoring across the 20 odd servers:
Description: Processor load5
Type: ZABBIX agent
Key: system.cpu.load[,avg5]
Type of information: Numeric (Float)
Use multiplier: Do not use
Update interval (in sec): 10
Keep history (in days): 7
Keep trends (in days) 365
Status: Monitored
Store value: As is

each item on the graph is configured as follows:
Parameter: blahblahblah: Processor Load5
Type: Simple
Function: All
Draw style: Line
Y axis side: Right
Sort order: 0

I can get the data to display if i change the update interval to 30 seconds but thats not useful for any of us who need to monitor the load averages. So everyone is changing their period from 1 hour to custom (3 hour) incriments.

Has anyone experienced this? or found a fix for it?

[pdwalker]

What are you trying to gain from 10 second monitoring that 30 second monitoring cannot do?

As of this time, monitoring the data too quickly causes the empty graphs. Perhaps you can query you db to see if the values are getting stored in the db? Otheriwse it might be a graph drawing problem.

- Paul

[jojo]

Yeh i managed to fix it up by downgrading back to 1.1.1 and everything works.

I'll recompile 1.1.2 again and double check i'm using the correct bin/libs/frontend because it works fine with 1.1.1

Values were getting stored in the DB we could see that but no graphs unless you changed the period to something higher than 1 hour. 10 second monitoring is fine for us but anything longer isn't going to work for us. The servers we're monitoring are medical related so we have to keep an eye on them.

[oliverm]

Weird, i'm seeing the exact same issue on 1.4 with graphs being blank when showing 1 or 2 hours.

Olly

[eli.stair]

Changing 'update interval' causes old data to be "missing" from graphs

I've just found a way to trigger this bug, or one /very/ similar. Running zabbix 1.4.2, mysql5.0.21. If I change the update interval on a set of values that are being graphed (that graph is also in a screen) from 300sec to 60sec updates, viewing any graphs containing that data lose most of the data samples in the graph. ALL the new data collected from the update interval change shows up, but only about 10% of the old data points appear in any graphs.

It looks like the query generating the graphs is only matching data at certain timeperiod boundaries, samples taken that align with the NEW update interval. Other samples, though they are present, are silently ignored.

This is a pretty big problem, since I just made this change on my primary mysql master server for my zabbix install. Switching the metrics' update interval back to 300 from 60 fixed the display problem with old data, and it displays the "temporarily 60-sec" data just fine along with the historical 5-minute data.

Historical data needs to be present and show up in graphs despite changes in the currently-defined polling interval! It looks at first glance this is being caused by data-range selection queries in one of the chart*php files, but haven't tracked down the order and dependencies of those yet to isolate it.

/eli