Boost 2.3 Released

[TheWitness]

All,

I have a new version of Boost. There have been a few changes in this release. They include:

--- 2.2 ---
bug: If running the boost server, mysql can timeout causing 2006 errors
bug: Typo relative to Linux Kernel versions in relative to command length
bug: Logging did not work on all occasions due to an undefined variable error
bug: Improve the counting algorythm to remove some bogus RRDtool update errors
bug: Check the validity of Multivalue updates prior to allowing updates
bug: Correct documentation relative to the the inittab process so that it works
bug: Prevent more than one Boost Server from running at a time by validating Socket bind

--- 2.3 ---
feature: Better max memory reporting for older PHP installs
feature: Reduce the load on the database server when displaying boost stats
feature: Integrate better with tholds vrule display (image caching off)
feature: Upgrade Automatically when either viewing the console or the plugin management page.

Enjoy,

TheWitness

[tuux1598g]

Currently running a fairly large installation using a heavily modified version of Boost 2.1, however a problem that has come up with this (and other versions) is that the temporary table used when running poller_boost already exists - bad rand(). This causes major headaches when it occurs (albeit not too often) so a simple patch to check that table doesn't exist before using it might help a number of users, with very little overhead:

Code: Select all

--- setup.php.orig      2009-02-10 09:29:14.000000000 +0000
+++ setup.php   2009-02-10 09:29:16.000000000 +0000
@@ -840,7 +840,15 @@

        include_once($config["library_path"] . "/rrd.php");

-       $temp_table_suffix = rand(0,10000000000);
+        while (1) {     // add code to check temp table doesn't already exist (this causes major problems)
+                $temp_table_suffix = rand(0,10000000000);
+
+                $boost_table_search=db_fetch_assoc("SHOW TABLES LIKE 'rcaston_$temp_table_suffix'");
+                if ( count($boost_table_search) )
+                        cacti_log("NOTE: Temporary table already exists, generating a new random number.", FALSE, "BOOST");
+                else
+                        break;
+        }

        /* suppress warnings */
        error_reporting(E_ALL);

[TheWitness]

Currently running a fairly large installation using a heavily modified version of Boost 2.1, however a problem that has come up with this (and other versions) is that the temporary table used when running poller_boost already exists - bad rand(). This causes major headaches when it occurs (albeit not too often) so a simple patch to check that table doesn't exist before using it might help a number of users, with very little overhead:

Code: Select all

--- setup.php.orig      2009-02-10 09:29:14.000000000 +0000
+++ setup.php   2009-02-10 09:29:16.000000000 +0000
@@ -840,7 +840,15 @@

        include_once($config["library_path"] . "/rrd.php");

-       $temp_table_suffix = rand(0,10000000000);
+        while (1) {     // add code to check temp table doesn't already exist (this causes major problems)
+                $temp_table_suffix = rand(0,10000000000);
+
+                $boost_table_search=db_fetch_assoc("SHOW TABLES LIKE 'rcaston_$temp_table_suffix'");
+                if ( count($boost_table_search) )
+                        cacti_log("NOTE: Temporary table already exists, generating a new random number.", FALSE, "BOOST");
+                else
+                        break;
+        }

        /* suppress warnings */
        error_reporting(E_ALL);

This whole situation worries me as this should not happen. The only process that uses temp tables is the main poller process. So, you should probably send me your "modified" version so that I can assess what should be carried upstream and where the bug is.

TheWitness

[tuux1598g]

This whole situation worries me as this should not happen. The only process that uses temp tables is the main poller process. So, you should probably send me your "modified" version so that I can assess what should be carried upstream and where the bug is.

You are absolutely correct, my mistake - this was occurring in my setup due to the modified version of boost I was using, should have realised that this fix was made to my own development rather than direct on boost before posting, I do apologise.

The current boost that we are running splits RRD files across n multiple directories (therefore multiple physical disks) using a plugin for path management, and then executes n boost poller processes each writing the data only to its unique path in parallel - this is where we resulted in two temp tables having the same name The only time this might occur in a 'normal' setup would be where the previous boost poller process crashed out, or the database heap was exhausted on the previous run (in which case you have bigger problems)...

[knallfrosch]

Hi,
it took me several hours of nasty debugging (newbie to boost, no PHP god) until I realized that boost isn't fully compatible with the "Structured RRA Path" option from 0.8.7c.

That means that boost simply isn't able to create new graphs when the subdirectory the graph shall go into (.../host_id/) isn't already there.

I found the code for the "non boost" directory creation in line 281 - 299 of lib/rrd.php but simply didn't have the nerve to try to implant it into the boost_server.php myselve (it's been a long day).
Maybe someone else feels lucky and wants to give it a try?

I feel that compatibility with the structured path would be a great thing for boost to go on with to keep directory seek time down.

Best regards
Knallfrosch

[TheWitness]

That's pretty easy to do. Consider it done. Surprised I did not figure that one out. Bug reports at http://bugs.cacti.net are always appreciated.

TheWitness

[mrvu]

I have problem with structured RRD path too. Waiting for the updated version of Boost.

http://forums.cacti.net/viewtopic.php?t=31041

Hi,
it took me several hours of nasty debugging (newbie to boost, no PHP god) until I realized that boost isn't fully compatible with the "Structured RRA Path" option from 0.8.7c.

That means that boost simply isn't able to create new graphs when the subdirectory the graph shall go into (.../host_id/) isn't already there.

I found the code for the "non boost" directory creation in line 281 - 299 of lib/rrd.php but simply didn't have the nerve to try to implant it into the boost_server.php myselve (it's been a long day).
Maybe someone else feels lucky and wants to give it a try?

I feel that compatibility with the structured path would be a great thing for boost to go on with to keep directory seek time down.

Best regards
Knallfrosch

[TheWitness]

All,

Attached is a patch that should resolve the structured path issue.

TheWitness

[mrvu]

Tested with Cacti 0.8.7c and Boost 2.3.
Thank you!

[dononeil]

Is there an updated template floating around somewhere for plotting the boost stats?

[TheWitness]

I have attached a nice "engineering" tool for boost for those capable of making out the stuff that makes me tick...

TheWitness

[drewshin99]

sorry if i've missed something, but i'm running boost 2.3 on windows and i dont see the "on demand" rrd updating feature working. if i understand correctly, if i pull up a graph and boost is still holding the update for the graph, boost will then update the graph's rrd with the held data?

thanks!

[TheWitness]

Correct. All updates are cached in a memory table until there is demand to view the graph, or one of two things occurs, 1) your table fills with X records or more, a time period Y elapses. If there is demand for the graph, it updates the RRD and then creates the image.

TheWitness