Cacti users sometimes complain about NaN's in their graphs. Unfortunately, there are several reasons for this result. The following is a step-by-step procedure recommended for debugging.
Please have a look at your cacti log file. Usually, you'll find it at <path_cacti>/log/cacti.log. Else see Settings, Paths. Check for this kind of error:
SPINE: Host[...] DS[....] WARNING: SNMP timeout detected [500 ms], ignoring host '........'
For “reasonable” timeouts, this may be related to a snmpbulkwalk issue. To change this, see Settings, Poller and lower the value for The Maximum SNMP OID's Per SNMP Get Request. Start at a value of 1 and increase it again, if the poller starts working. Some agent's don't have the horsepower to deliver that many OID's at a time. Therefore, we can reduce the number for those older/underpowered devices.
For scripts, run them as cactiuser from cli to check basic functionality. E.g. for a perl script named your-perl-script.pl with parameters “p1 p2” under *nix this would look like:
su - cactiuser /full/path/to/perl your-perl-script.pl p1 p2 ... (check output)
For snmp, snmpget the _exact_ OID you're asking for, using same community string and snmp version as defined within cacti. For an OID of .22.214.171.124.4.something, community string of very-secret and version 2 for target host target-host this would look like
snmpget -c very-secret -v 2c target-host .126.96.36.199.4.something .... (check output)
First make sure that crontab always shows poller.php. This program will either call cmd.php, the PHP based poller _or_ spine, the fast alternative, written in C. Define the poller you're using at Settings, Poller. Spine has to be implemented seperately, it does not come with cacti by default.
Now, clear ./log/cacti.log (or rename it to get a fresh start)Then, change Settings, Poller Logging Level to DEBUG for _one_ polling cycle. You may rename this log as well to avoid more stuff added to it with subsequent polling cycles.
Now, find the host/data source in question. The Host[<id>] is given numerically, the <id> being a specific number for that host. Find this <id> from the Devices menue when editing the host: The url contains a string like
Check, whether the output is as expected. If not, check your script (e.g. /full/path/to/perl). If ok, proceed to next step
This procedure may be replaced by running the poller manually for the failing host only. To do so, you need the <id>, again. If you're using cmd.php, set the DEBUG logging level as defined above and run
php -q cmd.php <id> <id>
If you're using spine, you may override logging level when calling the poller:
./spine --verbosity=5 <id> <id>
All output is printed to STDOUT in both cases. This procdure allows for repeated tests without waiting for the next polling interval. And there's no need to manually search for the failing host between hundreds of lines of output.
The goal of bulkwalks is to reduce SNMP traffic overhead. It works by cramming several SNMP requests/responses into a single IP packet. This feature is not available with SNMP version 1. Some SNMP enabled devices do not like snmpbulkwalks.
Cacti supports this feature with SNMP enabled devices automatically when version 2 or version 3 has been selected. The field “max OIDs” for each hosts governs, how many packets are crammed together. Side note: In case too many SNMP packets will be crammed together, IP fragmentation takes care of splitting those into chunks manageable by the IP layer. You will see such an effect when e.g. your manual
snmpwalk -c <community string> -v 2c <target> <OID>
produces a result but cacti poller output shows NaN.
Now you have two different means to tackle such an issue:
- reduce “max OIDs” to 1: Cacti now will disable all Cacti-internal mechanisms to use snmpbulkwalk
- select SNMP version 1: as SNMP V1 does not support snmpbulkwalks, all Cacti-internal and Cacti-external bulkwalk mechanisms are disabled
Cacti-internal bulkwalk mechanisms: Spine checks the “max OIDs”. In case they are set to a value higher than 1, we will use snmpbulkwalk-like code. Else, we use standard snmpwalks.
Cacti-external bulkwalk mechanisms: It has been found, that php-snmp automatically uses snmpbulkwalk, even when only snmpwalk has been requested. As of current, php-snmp will join 20 requests/response. We can't change this setting externally. So the ultimate answer to this is to use SNMP version 1. The drawback of using SNMP version 1 is that e.g. COUNTER64 is not available with this setting. As a result, e.g. a Verbose Query from within the browser may fail while spine still works. Yes, crazy.
In most cases, this step make be skipped. You may want to return to this step, if the next one fails (e.g. no rrdtool update to be found)
From debug log, please find the MySQL update statement for that host concerning table poller_output. On very rare occasions, this will fail. So please copy that sql statement and paste it to a mysql session started from cli. This may as well be done from some tool like phpMyAdmin. Check the sql return code.
Down in the same log, you should find some
rrdtool update <filename> --template ...
You should find exactly one update statement for each file.
RRD files should be created by the poller. If it does not create them, it will not fill them either. If it does, please check your Poller Cache from Utilities and search for your target. Does the query show up here?
If rrd files were created e.g. with root ownership, a poller running as cactiuser will not be able to update those files
cd /var/www/html/cacti/rra ls -l localhost* -rw-r--r-- 1 root root 463824 May 31 12:40 localhost_load_1min_5.rrd -rw-r--r-- 1 cactiuser cactiuser 155584 Jun 1 17:10 localhost_mem_buffers_3.rrd -rw-r--r-- 1 cactiuser cactiuser 155584 Jun 1 17:10 localhost_mem_swap_4.rrd -rw-r--r-- 1 cactiuser cactiuser 155584 Jun 1 17:10 localhost_proc_7.rrd -rw-r--r-- 1 cactiuser cactiuser 155584 Jun 1 17:10 localhost_users_6.rrd
Run the following command to cure this problem
chown cactiuser:cactiuser *.rrd
You're perhaps wondering about this step, if the former was ok. But due to data sources MINIMUM and MAXIMUM definitions, it is possible, that valid updates for rrd files are suppressed, because MINIMUM was not reached or MAXIMUM was exceeded.
Assuming, you've got some valid rrdtool update in step 3, perform a
rrdtool fetch <rrd file> AVERAGE
and look at the last 10-20 lines. If you find NaN's there, perform
rrdtool info <rrd file>
and check the ds[…].min and ds[…].max entries, e.g.
ds[loss].min = 0.0000000000e+00 ds[loss].max = 1.0000000000e+02
In this example, MINIMUM = 0 and MAXIMUM = 100. For a ds.[…].type=GAUGE verify, that e.g. the number returned by the script does not exceed ds[…].MAX (same holds for MINIMUM, respectively).
If you run into this, please do not only update the data source definition within the Data Template, but perform a
rrdtool tune <rrd file> --maximum <ds-name>:<new ds maximum>
for all existing rrd files belonging to that Data Template.
At this step, it is wise to check step and heartbeat of the rrd file as well. For standard 300 seconds polling intervals (step=300), it is wise to set minimal_heartbeat to 600 seconds. If a single update is missing and the next one occurs in less than 600 seconds from the last one, rrdtool will interpolate the missing update. Thus, gaps are “filled” automatically by interpolation. Be aware of the fact, that this is no “real” data! Again, this must be done in the Data Template itself and by using rrdtool tune for all existing rrd files of this type.
Last resort would be to check, that the correct data sources are used. Goto Graph Management and select your Graph. Enable DEBUG Mode to find the whole rrdtool graph statement. You should notice the DEF statements. They specify the rrd file and data source to be used. You may check, that all of them are as wanted.
Up to cacti 0.8.6j, table poller_output may increase beyond reasonable size.
This is commonly due to php.ini's memory settings of 8MB default. Change this to at least 64 MB.
To check this, please run following sql from mysql cli (or phpmyadmin or the like)
select count(*) from poller_output;
If the result is huge, you may get rid of those stuff by
truncate table poller_output;
From cacti 0.8.7 on, measures were taken on both issues (memory size, truncating poller_output).
Most rpm installations will setup the crontab entry now. If you've followed the installation instructions to the letter (which you should always do ), you may now have two poller running. That's not a good thing, though. Most rpm installations will setup cron in /etc/cron.d/cacti
Now, please check all your crontabs, especially /etc/crontab and crontabs of users root and cactiuser. Leave only one poller entry for all of them. Personally, I've chosen /etc/cron.d/cacti to avoid problems when updating rpm's. Mosten often, you won't remember this item when updating lots of rpm's, so I felt more secure to put it here. And I've made some slight modifications, see
shell> vi /etc/cron.d/cacti */5 * * * * cactiuser /usr/bin/php -q /var/www/html/cacti/poller.php > /var/local/log/poller.log 2>&1
This will produce a file /var/local/log/poller.log, which includes some additional informations from each poller's run, such as rrdtool errors. It occupies only some few bytes and will be overwritten each time.
If you're using the crontab of user “cactiuser” instead, this will look like
shell> crontab -e -u cactiuser */5 * * * * /usr/bin/php -q /var/www/html/cacti/poller.php > /var/local/log/poller.log 2>&1
Pay attention to custom scripts. It is required, that external commands called from there are in the $PATH of the cactiuser running the poller. It is therefor recommended to provide /full/path/to/external/command
User “criggie” reported an issue with running smartctl. It was complaining “you are not root” so a quick chmod +s on the script fixed that problem.
Secondly, the script was taking several seconds to run. So cacti was logging a “U” for unparseable in the debug output, and was recording NAN. So my fix there was to make the script run faster, and the age of my box made it difficult to accomplish.
The timeout setting is governed by “Settings → Poller → Script and Script Server Timeout Value”. In general, it is recommended to make scripts faster to avoid that the poller does not finish in time.