Boost 3.0 Released

[TheWitness]

All,

Boost 3.0 is released. This enhancement will halve the time it takes to run the boost process. In addition, we add more logging and some nice statistics (waiting on someone to create a graph template for this).

As always, you can obtain from the Doc's site. Here is the ChangeLog.

— 3.0 —

* feature: Deep performance tuning and code cleanup
* feature: Add logging to boost updates for performance tuning
* feature: Add hooks to support 95th Percentile and Bandwidth Summation

— 2.6 (Unreleased) —

* bug: Initial RRDfile update taks two boost cycles and not one
* bug: Cacti RRDtool fetch command bypasses boost causing issues
* bug: Don't optimize boost table if it's in MEMORY storage engine

[brylant]

Any chance for "upgrade guide"...?

[TheWitness]

Well, if already running Boost 2.x, the process is quite simple.

1) Download the tar.gz to /tmp
2) untar
3) verify permissions
4) cp -rp /tmp/boost /var/www/html/cacti/plugins
5) goto Console->Plugin Management
6) QED

TheWitness

[toe_cutter]

Awesome improvement!
From ~70s to ~55s

<removed because of brainfail>

[uhtred]

hi,
after installing boost plugin (3.0-1) i got the following failure in my cacti log:

02/26/2010 09:53:28 AM - CMDPHP: Poller[0] ERROR: SQL Row Failed!, Error:'145', SQL:"SHOW CREATE TABLE plugin_camm_syslog"
02/26/2010 09:53:28 AM - CMDPHP: Poller[0] ERROR: SQL Row Failed!, Error:'145', SQL:"SHOW CREATE TABLE plugin_camm_snmptt_stat"
02/26/2010 09:53:17 AM - CMDPHP: Poller[0] ERROR: SQL Row Failed!, Error:'145', SQL:"SHOW CREATE TABLE plugin_camm_syslog"
02/26/2010 09:53:17 AM - CMDPHP: Poller[0] ERROR: SQL Row Failed!, Error:'145', SQL:"SHOW CREATE TABLE plugin_camm_snmptt_stat"

02/26/2010 09:57:21 AM - CMDPHP: Poller[0] ERROR: SQL Row Failed!, Error:'145', SQL:"SHOW CREATE TABLE plugin_camm_syslog"
02/26/2010 09:57:21 AM - CMDPHP: Poller[0] ERROR: SQL Row Failed!, Error:'145', SQL:"SHOW CREATE TABLE plugin_camm_snmptt_stat"

02/26/2010 10:05:03 AM - CMDPHP: Poller[0] ERROR: SQL Row Failed!, Error:'145', SQL:"SHOW CREATE TABLE plugin_camm_syslog"
02/26/2010 10:05:03 AM - CMDPHP: Poller[0] ERROR: SQL Row Failed!, Error:'145', SQL:"SHOW CREATE TABLE plugin_camm_snmptt_stat"

after this 3 times its clean.
the camm plugin is installed, but not enabled. the tables above doesn't exists in my db.

[massimo]

Hi, I'm very interested in this plugin and porting it to another monitoring system (Zenoss). Maybe somebody can help me understand how its done in Cacti (i.e., architecture, dependencies, data-flow etc.). In the meantime I will investigate how this idea and concept fits into the Zenoss' system. Any help, directions... appreciated.

Thanks,
massimo

[KyleT]

Hi

I'm having issues when adding new devices using boost.

When not using boost, cacti creates the RRDs as follows:

Code: Select all

-rw-r--r-- 1 cacti  cacti   91068280 Mar 17 11:23 cactirealtime_load_1min_56174.rrd

However, when boost is enabled, I create a graph, then I go view that graph (causing on demand update), the rrd is created as follows:

Code: Select all

-rw-r--r-- 1 apache apache  60712448 Mar 16 15:22 cactirealtime_time_56215.rrd

Cacti user is apart of the Apache group and visa versa.

As you can see, I have two issues:
1. When a graph and data source is deleted in the web interface, the rrd does not get deleted (if rrd created by poller).
2. When boost reaches its max time and dumps data into rrds, the rrds created by apache user are not updated due to permissions.

Could anyone advise the best way to fix this? Is there any way to get Cacti and Apache to create the files with a mask of 664?

Cheers
Kyle

[TheWitness]

Hi

I'm having issues when adding new devices using boost.

When not using boost, cacti creates the RRDs as follows:

Code: Select all

-rw-r--r-- 1 cacti  cacti   91068280 Mar 17 11:23 cactirealtime_load_1min_56174.rrd

However, when boost is enabled, I create a graph, then I go view that graph (causing on demand update), the rrd is created as follows:

Code: Select all

-rw-r--r-- 1 apache apache  60712448 Mar 16 15:22 cactirealtime_time_56215.rrd

Cacti user is apart of the Apache group and visa versa.

As you can see, I have two issues:
1. When a graph and data source is deleted in the web interface, the rrd does not get deleted (if rrd created by poller).
2. When boost reaches its max time and dumps data into rrds, the rrds created by apache user are not updated due to permissions.

Could anyone advise the best way to fix this? Is there any way to get Cacti and Apache to create the files with a mask of 664?

Cheers
Kyle

It all about having the correct directory permissions on the rra directory. You need to make sure that your Cacti poller user is in the Apache group and make sure that the apache group owns the directory and that the group has write permissions to the folder. You also need to set the bit on the directory permissions to have all file inherit the permissions of the directory instead of using the UMASK for the user.

Pretty sure you have to use a chmod with a +s on it. Don't use the syntax often enough to quote it.

Maybe someone else with more experience can help out here.

TheWitness

[meralias]

use 'chmod 2775' on the rra directory, making sure that the rra directory has the right ownership. (similar to running chmod g+s)

[KyleT]

Thank you both!!!

I completely forgot about permission inheritance .

[aleu]

Boost 3.0 is released. This enhancement will halve the time it takes to run the boost process. In addition, we add more logging and some nice statistics (waiting on someone to create a graph template for this).

Hmm, when I select "Enable direct population of poller_output_boost table by spine" during rrd updates quite large fraction of my graphs shows gaps in the graphs (poller runtime exceeds 300 seconds). Otherwise, the poller runtime is on the order of 65 seconds. Any idea why and which boost parameters I can tweak to mitigate this?

Current setup:
spine
2 concurrent poller processes
Enabled balance process load
30 threads
2 PHP script servers
60 sec script server timeout
50 OIDs per SNMP GET Request

Boost:
runs every 4 hrs (or 500,000 DSs)
Maximum sources per pass set ot 1,000
Memory limit for Boost and Poller set to 1GBytes
Maximum RRD Update Script Runtime 1 hour
Disabled direct population of poller_output_boost table (to avoid gaps every 4 hrs)
I am not using Boost Server
Image caching is enabled.

Thanks

[TheWitness]

Turn of Balance Processes plus 60 ss timeout is bad.

TheWitness

[BorisL]

2 concurrent poller processes
Enabled balance process load
30 threads

Do you have your MySQL ready for 60 simultaneous INSERTs? Check 'show processlist' while polling with direct population enabled.
What engine do you use for poller_output_boost table?

[aleu]

Turn of Balance Processes plus 60 ss timeout is bad.
TheWitness

Are you saying that with boost Balance Processes should be disabled by default? If so, can you explain why?
Also, I have a couple of scrips which take around 45 sec to complete, and if I lower the SS timeout below 60 I do ocasionally get gaps in these graphs.

[aleu]

2 concurrent poller processes
Enabled balance process load
30 threads

Do you have your MySQL ready for 60 simultaneous INSERTs? Check 'show processlist' while polling with direct population enabled.
What engine do you use for poller_output_boost table?

MYISAM. When I tried to use "Memory" I got gaps across all the graphs at random times (despite setting max_heap_table_size=768M). I will execute the "show processlist" command in a couple of hours when I get home.