[HOWTO] Cacti's setup for really BIG environments

[BorisL]

So i done all yours advice that you explain in this topic except for the Boost plugin.

You have no Boost plugin enabled though: RRDsProcessed is zero when it is enabled.
But even then, these advices are not likely to fix your periodic cpu consuption problem, I presume the cause of that problem is outside of cacti installation.

[tosage]

Yep i dont have this plugin enable in my installation.
Thanks for your reply, i go search what is the reason which create their leaks...

Ps : sorry for my English, is so poor :/

[BorisL]

My patchset for 0.8.7i is located here. Mind that this patch will need some more stuff in spine(use spine/branches/0.8.7 version from svn as base) and SQL alter (adding column polling_time, see here for details).

As an additional current issue (I hope this will be fixed as errata patch for 0.8.7i) currently one need to apply this patch too.

[c3226026]

My patchset for 0.8.7i is located here. Mind that this patch will need some more stuff in spine and SQL alter (adding column polling_time, see here for details).

As an additional current issue (I hope this will be fixed as errata patch for 0.8.7i) currently one need to apply this patch too.

Hello BorisL,

Thanks for these information, I've try your 0.8.7i patch, but I've this error :

Hunk #2 FAILED at 210 1 out 7 hunk FAILED : but in my poller.php file line 210 is empy

It's normal ?

Thanks for your return

[BorisL]

Thanks for these information, I've try your 0.8.7i patch, but I've this error :
Hunk #2 FAILED at 210 1 out 7 hunk FAILED : but in my poller.php file line 210 is empy
It's normal ?

This is not normal, I think you are trying to apply patch against non-PIA 0.8.7i distribution. Basically you need PIA in order to have BOOST plug-in up and running. Is my suggestion correct?

[c3226026]

No I use 0.8.7i-pia version, but I have always take some file on svn repository, so I think it's for that.

I've test on the "original" 0.8.7i package and all are fine (sorry )

but for your other "patch" http://bugs.cacti.net/view.php?id=2150, I don't understand what I need to do and on what file

Does I need to apply this path http://bugs.cacti.net/view.php?id=2070 on spine 0.8.7i ?

Thanks in advanced for your answer

[BorisL]

but for your other "patch" http://bugs.cacti.net/view.php?id=2150, I don't understand what I need to done and on what file

lib/sort.php. Remove '\' chars in explodes inside usort_data_query_index() function.

Does I need to apply this path http://bugs.cacti.net/view.php?id=2070 on spine 0.8.7i ?

Huh, this is the correct time to fetch files from svn, not "original" version This is a patch against spine/branches/0.8.7

[c3226026]

Thanks for your answer, for spine 0.8.7i, I haven't see new file on svn, spine 0.8.7i is tag 0.8.7i 6972 (it's not the original version ?)

Thanks in advanced for your last confirmation.

[BorisL]

Thanks for your answer, for spine 0.8.7i, I haven't see new file on svn, spine 0.8.7i is tag 0.8.7i 6972 (it's not the original version ?)

As you can see in the patch, it is built against revision 6836. You may try to patch rev 6972, I expect patch to be applied cleanly.

[c3226026]

Thanks I will try it

[BorisL]

Additionally, one may try to use new php-snmp library. It can be fetched from PHP trunk, I made best efforts to make it backwards compatible downto 5.2 branch. Using patched 0.8.7i and new php-snmp library will result in significantly increased speed of SNMP reindex operations (for example, new library can make use of GET query with multiple OIDs in single query, not one-by-one)

[c3226026]

I've check svn 6972 is same than original 0.8.7i version, no diff.

I'will try your patch on it.

For php-snmp, I just need to install it, no change to do after on cacti ?


[Edit]
For spine patch, I've tested, but I cannot apply it :

patch -p1 -N < ../data_rate-17.11.2011.patch
can't find file to patch at input line 5
Perhaps you used the wrong -p or --strip option?
The text leading up to this was:
--------------------------
|Index: poller.h
|===================================================================
|--- poller.h (revision 6836)
|+++ poller.h (working copy)
--------------------------
File to patch:



It's probably not same file than 0.8.7h, you have applyed it on you 0.8.7i spine version ?
[/Edit]

[BorisL]

For php-snmp, I just need to install it, no change to do after on cacti ?

You will need to restart HTTP server in order to allow php web-module load new SNMP module. Poller will hook it in automatically.

patch -p1 -N < ../data_rate-17.11.2011.patch

There should be no -p option here. There is nothing to strip in path in patch.

[tosage]

Hello all,

Anybody know this packet :

To speed-up the polling process at large deployments you could apply some performance tweaks. This includes a php opcode cacher:

aptitude install php5-xcache

and replacing the current poller process cron-job with multiple parallel instances:

*/5 * * * * root cd /opt/observium/ && ./poller.php -i 8 -n 0 >> /dev/null 2>&1
*/5 * * * * root cd /opt/observium/ && ./poller.php -i 8 -n 1 >> /dev/null 2>&1
*/5 * * * * root cd /opt/observium/ && ./poller.php -i 8 -n 2 >> /dev/null 2>&1
*/5 * * * * root cd /opt/observium/ && ./poller.php -i 8 -n 3 >> /dev/null 2>&1
*/5 * * * * root cd /opt/observium/ && ./poller.php -i 8 -n 4 >> /dev/null 2>&1
*/5 * * * * root cd /opt/observium/ && ./poller.php -i 8 -n 5 >> /dev/null 2>&1
*/5 * * * * root cd /opt/observium/ && ./poller.php -i 8 -n 6 >> /dev/null 2>&1
*/5 * * * * root cd /opt/observium/ && ./poller.php -i 8 -n 7 >> /dev/null 2>&1

I view this tweak for observium, maybe useful/useless in cacti ????

Thanks for comments

[BorisL]

I view this tweak for observium, maybe useful/useless in cacti ????

As far as I can see there are two tweaks in quotation.
First is to use xcache, php bytecode cacher. It will be userful when you have a lot of queries to your server.
Second seems to be specific to observium.

So basically these tweaks will not help cacti (unless you are suffering from php eating up your CPU compiling php code, I haven't seen such scenarios so far) to speed up.