should I use spine???

kido · Post by **kido** » Thu Mar 26, 2009 1:01 pm

Hi guys,

I just ran :

# php poller.php

and here's the result in cacti.log:

Code: Select all

SYSTEM STATS: Time:289.9115 Method:cmd.php Processes:1 Threads:N/A Hosts:17 HostsPerProcess:17 DataSources:535 RRDsProcessed:414

Now, according to http://www.cacti.net/spine_info.php , I understand that I should use spine, cause I'm almost at the 300s threshold.
I checked the download page and the latest spine version available is 0.8.7c. I have 0.8.7d cacti installed.

Q1: am I correct when I say that I should install spine (Time:289.9115") ?
Q2: will 0.8.7c spine work with 0.8.7d cacti?

Thank you!

mcutting · Post by **mcutting** » Thu Mar 26, 2009 1:28 pm

Looking at your stats, I wouldn't go for Spine just yet. I would, however, increase the number of pollers, as I'm sure your system can do more than just one.

Try this first. Spine 0.8.7c will work with he "d" version of Cacti.

kido · Post by **kido** » Fri Mar 27, 2009 5:50 am

Many thanks for your suggestion! It worked - I noticed some improvements.
I increased number of pollers to 4 and also increased memory_limit in php.ini from 16 to 32M.
It was a great increase in performance, indeed.
But now I added a few more device - there are 29 now (17 before changes) and TIME is again close to limit:

Code: Select all

SYSTEM STATS: Time:284.6405 Method:cmd.php Processes:4 Threads:N/A Hosts:29 HostsPerProcess:8 DataSources:938 RRDsProcessed:715

I intend to increase again the number of parallel pollers.
My questions is: how many pollers can I create? Would a number of 10, for instance, be supported?

Thanks for your help provided so far!

Post by **TheWitness** » Fri Mar 27, 2009 7:03 am

You must have some very slow running scripts. In that case, I typically collect the data asynchronously and set it aside, then the poller simply grabs the local data. Much faster. What scripts are running so slow? What are they doing?

TheWitness

mcutting · Post by **mcutting** » Fri Mar 27, 2009 7:03 am

You might also want to change parameters such as the amount of OID's per Get Request for example.

Also, your poller time is incredibly high for such few devices. To give you an idea, I get

Code: Select all

Cacti[5620]: SYSTEM: STATS: Time:19.3240 Method:spine Processes:3 Threads:15 Hosts:158 HostsPerProcess:53 DataSources:1496 RRDsProcessed:0

Ok, granted I use SPINE and BOOST, but I think you get the idea. How many Script Servers are you running ?

You might also want to take a look at Gandalf's POLLPERF plugin to get some ideas of what is slowing things down.

kido · Post by **kido** » Fri Mar 27, 2009 8:17 am

just to give a quick possible (!) explanation: the machine that hosts cacti is off-site : somewhere in eastern Europe and the monitored machines are in US...there's a major network delay , more then 200 ms to each monitored device....I think that could cause such big TIME values...If I'm wrong, please correct me.....The plan is that once I have a nice working cacti installation I'll migrate it on-site - hopefully performance will be improved...this is just to give you a quick answer (don't know if the correct one).

I didn't install plugin arch, I'll take a look later if things become worse...I just have a few dozens hosts to add...and of course lots of config/tuning,cause the env is very heterogeneous: all sorts of Win, Solaris, HP-UX, Linux, AIX...

kido · Post by **kido** » Fri Mar 27, 2009 12:21 pm

Hi,

I installed PA and polperf plugin, defined time and cycles in Settings. In Poller Performance there's nothing displayed (should I wait until ALL cycles are done??).

Here are some repetitive errors that might be interesting (from cacti.log):

Code: Select all

03/27/2009 09:16:50 PM - CMDPHP: Poller[0] ERROR: SQL Assoc Failed!, Error:'1146', SQL:"SELECT name, file, function FROM plugin_hooks WHERE status = 1 AND hook = 'poller_on_demand'"
03/27/2009 09:16:50 PM - CMDPHP: Poller[0] ERROR: SQL Assoc Failed!, Error:'1146', SQL:"SELECT name, file, function FROM plugin_hooks WHERE status = 1 AND hook = 'poller_output'"

Thanks again!

mcutting · Post by **mcutting** » Fri Mar 27, 2009 12:25 pm

Your cacti log needs to be in DEBUG before POLLPERF will see anything. Let this happen for a couple of passes, then place your log back to LOW. Leaving it in DEBUG for a long period of time will kill the disk space on your server.

kido · Post by **kido** » Fri Mar 27, 2009 12:34 pm

I'm pretty sure the plugin set DEBUG log level automatically, cause it was LOW before proceeding, and now it's on DEBUG. I guess it will reset back to LOW after job is done (I'll watch it, anyway...).

Number of cycles is set to 10 (default), 6 already passed, but still nothing displayed in Poller Performance page

Those errors from previous post occur continuously....

kido · Post by **kido** » Fri Mar 27, 2009 1:09 pm

OK, time's up for polperf , but nothing is in the corresponding page...

(log level reset to LOW)

Any ideas?

Thank you!

mcutting · Post by **mcutting** » Fri Mar 27, 2009 1:22 pm

Did you import the pa.sql file ? Judging by those errors, you have some tables missing in your db.

kido · Post by **kido** » Fri Mar 27, 2009 1:24 pm

OK, I "think" I found something...
According to http://forums.cacti.net/about30590.html ,first I should've imported the pa.sql into mysql , which I haven't done so far...

I checked that file and it creates certain tables that are NOT present in my db (checked that, as well)...

I'll do that and come back with the results...

kido · Post by **kido** » Fri Mar 27, 2009 1:25 pm

lol
simultaneous posting

mcutting · Post by **mcutting** » Fri Mar 27, 2009 1:27 pm

let me know how you get on.

Ps - Windoze of nix ?

Post by **TheWitness** » Fri Mar 27, 2009 1:27 pm

With that latency, you should use spine and balance the maxoids with the timeout until things are right.

TheWitness

should I use spine???

should I use spine???

Who is online