'poller_output' Question

[CPF]

Hi,

I've recently started using the very excellent 1-minute poller support in Cacti v0.8.7b

This is outstanding and is working really well.

I'm now if the process of changing some of my Weathermaps over to using the 'poller_output' support.

The first one that I've changed is working great, but I'm seeing these errors in the log:

[Howie]

Does it carry on doing that after a few poller cycles?

I can see that you might get it when you've just switched...

[CPF]

Hi Howie,

[Howie]

OK - I'll have a play. I'm using 1 minute polling and weathermap on one of our installations too.

[CPF]

Cool,

Thanks for that. Let me know if there's any more info that you need, or anything that you'd like me to try.

Cheers,

Graham.

[Howie]

Figured it out. What is going on is that it tries to match the filename you give it against the ones in the cactidb. However, Cacti stores them as <rrd_path>/blah.rrd, not the full path, so I have to get rrd_path, and do a replace with the string <rrdpath> so that I can find it in the database.

If you use a format for the filename that doesn't exactly match rrdpath, then my replace fails, and so weathermap can't find which poller_output entry to use. The reason it still works is that it falls back to using regular rrdtool if it can't use poller_output.

So the solution is to change the path you use for the rrd file. My guess is that replacing the / with \ will fix it.

[CPF]

Hi Howie,

Thanks for the pointers, but no luck. I've tried the following permutations:

[CPF]

Hi Howie,

Just been doing some more thinking about this one. Where does weathermap look for the poller data?

I'll be the first to admit that I don't really know the inner working of the polling process. What I have noticed that the table 'poller_output' in my Cacti database is almost always empty. Is this right?

[Howie]

No, that's what should happen.

The way it works is (roughly, I'm no Cacti internals expert): Cacti poller collects data, inserts it into poller_output, and then a different part of the process reads that data out and updates your rrd files with it, deleting poller_output rows as it goes.

What the plugin architecture allows us to do is intercept various internal parts of Cacti and tinker with them before Cacti gets a chance to do it's normal thing. One of those 'hooks' is called poller_output, and sits in the path between the poller, and the poller_output table (or possibly poller_output and the rrd updater). Plugins like Thold and Weathermap can watch the data go past and pick it out to save somewhere else (weathermap_data, thold_data) before the poller removes it.

This saves weathermap from running 40 copies of rrdtool to fetch the data for your map, so it's quicker. If you use Boost, then the rrd files aren't even updated every poller cycle, so you *have* to use poller_output to get current data.

[Howie]

More constructively, try these changes to WeatherMapDataSource_rrd.php:

Around line 18 (just after the commented-out block):

[CPF]

Hi Howie,

Thanks again for all your help with this. I've made the changes that you suggested, but without success.

Here's some more information...

Tracking down one of the specific items that I'm interested in gives me this:

[Howie]

[CPF]

Hi Howie,

I did make both the changes, but there's no <path_rra> in the log file. I'm going to PM both the log and the modified 'WeatherMapDataSource_rrd.php' to make sure that it's not me!

Thanks again for the help.

Graham.

[CPF]

Hi Howie,

Thanks for the file that you PM'd me. This does now report the 'rra_path' line, as you suggested:

[Howie]

Could you have a look through your DEBUG cacti.log for lines that say 'poller_output: Got one! '?

This is output by the other side of the process, where weathermap looks at the poller data for 'interesting' things. It sounds like the TARGET part is correctly inserting lines in weathermap_data now.