Boost 1.7 and cacti 0.8.7b issues

[TheWitness]

There are a few fixes in Cacti 0.8.7c to restart RRDtool if it crashes, so the boost server should be a bit more reliable in those cases moving forward. I do like the inittab compliant implementation that I saw the other day as well.

TheWitness

[khufure]

We have Boost installed with RRD 1.2.27. I haven't found any bad datasources or scripts. Would you be willing to either post or send me a personal message with your apache and mysql tuning settings? Also, if you have boost installed, do/did you get the following errors?:

10/16/2008 08:33:33 AM - BOOST SERVER: Poller[0] WARNING: Broken connection detected '127.0.0.1'
10/16/2008 08:33:32 AM - CMDPHP: Poller[0] ERROR: SQL Assoc Failed!, Error:'2006', SQL:"SELECT * FROM poller_output_boost_processes"

As Witness already mentioned, this is boost server dying. You have to make provisions if the process dies. I ran a dirty script out of cron per minute, but the inittab or demontools suggestions are better.

I wouldn't bother with any apache tuning, except for php.ini being not 8 megs . MySQL tunes for us:

Code: Select all

innodb_buffer_pool_size                         = 1G
query_cache_size                        = 32M
query_cache_limit               = 2M
tmp_table_size                  = 1536M
max_allowed_packet              = 16M
max_connections                         = 500
max_heap_table_size             = 1536M
myisam_sort_buffer_size = 16M
key_buffer                                      = 128M
table_cache                                     = 2048
sort_buffer_size                        = 32M

Our DBAs tuned MySQL in general and the cacti instance is not modified much. The tmp entries are for the boost memory db and most likely quite a bit too large. So, maybe this helps but this is not specifically tuned for Cacti just a large MySQL site.

Related links:
http://forums.cacti.net/viewtopic.php?t=11610
http://forums.cacti.net/viewtopic.php?t=19338

[deu439356]

1) In the php.ini file, are you referring to the max memory size or the max post size?

2) I kepting repeating a 'ps -ef | grep boost' while viewing graphs and the server wasn't dying. We don't have it in inittab, so it shouldn't respawn if it was dying and the PID never changed.

[monachus]

I'm having problems with Boost. I was running on a single host and have been having problems with the number of hosts/services we're polling, so I'd like to move to a distributed poller model. I understand Boost is the way to go, but it doesn't seem to be working as expected. I've had the on-demand updating in place for a while, and today I attempted to turn on the Boost server.

I'm running Cacti v0.8.7b-r2. First, I checked out the SVN version of Boost and found that the tab disappeared entirely from my configuration. I reverted to 1.7 and the configuration options returned. I cloned my cacti host, so the poller and the server are the same. I removed the cron entry from the server and fired up the Boost server. Things looked good. I activated the cron job on the poller host and let it run for a while. It doesn't work.

Occasionally the RRDs update, but most of the time the boost server shows errors like:

Code: Select all

11/20/2008 12:54:53 PM - BOOST_SERVER: Poller[0] ERROR: illegal attempt to update using time 1227193097 when last update time is 1227193097 (minimum one second step)

I've been all over the forums and google and I understand the RRD structure and how it knows where to insert the next update. I am 100% sure that I have only one poller process running. I understand that the errors can appear if the time is different on the poller and the server, and that's not the case either. What I don't understand is why the timestamps are identical, as if there are two updates being attempted when there should only be one.

I've turned the poller off and will have to live with the big gap in my graphs, but I'd like to know if anyone has advice on where to turn to get this running.

Additionally, I'm curious about how running multiple poller hosts works. Is there logic built into Cacti where it knows when it has handed off a list of devices to be polled to one node and won't hand them off to another node? How does this work?

Thanks in advance for the assistance!

Regards,

Adrian Goins - President/CEO
Arces Network, LLC
Arkes Sp. z o.o.

[TheWitness]

Boost 2.0 is official. Please Check it Out Here:

http://forums.cacti.net/viewtopic.php?p=150337#150337

Regards,

TheWitness