Boost Statistics

[mfarrenk]

Boost has been fabulous for me. But I don't understand all of the statistics it's showing me. I suspect I may need to do some tuning on MySQL as well but I don't know for sure. Can someone please explain each statistic that I'm seeing so that I can optimize the setup even more? Thank you.

RRDUpdates:1057044 TotalTime:2209.5298 range_local_data_id:47.52 rcaston_add:16.53 get_records:25.37 results_cycle:1634.13 rrd_path:84.28 rrd_template:94.16 rrd_lastupdate:135.91 rrd_field_names:13.17 rrdupdate:1139.54 delete:372.3

[TheWitness]

Boost has been fabulous for me. But I don't understand all of the statistics it's showing me. I suspect I may need to do some tuning on MySQL as well but I don't know for sure. Can someone please explain each statistic that I'm seeing so that I can optimize the setup even more? Thank you.

RRDUpdates:1057044 TotalTime:2209.5298 range_local_data_id:47.52 rcaston_add:16.53 get_records:25.37 results_cycle:1634.13 rrd_path:84.28 rrd_template:94.16 rrd_lastupdate:135.91 rrd_field_names:13.17 rrdupdate:1139.54 delete:372.3

Well, being that I wrote most of it, I guess I should explain. Here are the facts, from my fuzzy memory:

RRDUpdates - The total number of measured points inserted into your RRDfiles. That ones easy.

TotalTime - From start to finish, how much time it took to update all your RRDfiles.

Range Local Data ID - Well I did not write that, but it's basically the time it took for find the records to update.

RCaston Add - Well, that would be Rodney Caston from (cant tell you where). He helped me test it on a big system a couple of years ago. He is a temporary table that holds measured items to update.

Get Records - Obvious, get the local_data_ids from the poller_output_boost table. So, SQL is not the issue.

Results Cycle - The sum of time doing all the updates.

RRDpath - Validate the presence of the RRDfiles and locate them.

RRDTemplate - Formulate the template statement that describes the order the data will appear in the update statement.

RRDLastUpdate - Check the RRDfiles to see when they was last updated.

RRDFieldNames - For Responses with multiple values, the parsing of the Internal RRD Data Source names to associate with the Cacti Output Fields returned by the script.

RRDUpdate - Well, that's the RRDtool Update statements. The lion's share for you.

Delete - Remove Rows from the Poller Output Boost table.

So, for you, Solid State (FusionI/O) disks will take that ~1100 seconds and change it to 9 (Uhm, no kidding).

TheWitness

[mfarrenk]

Boost has been fabulous for me. But I don't understand all of the statistics it's showing me. I suspect I may need to do some tuning on MySQL as well but I don't know for sure. Can someone please explain each statistic that I'm seeing so that I can optimize the setup even more? Thank you.

RRDUpdates:1057044 TotalTime:2209.5298 range_local_data_id:47.52 rcaston_add:16.53 get_records:25.37 results_cycle:1634.13 rrd_path:84.28 rrd_template:94.16 rrd_lastupdate:135.91 rrd_field_names:13.17 rrdupdate:1139.54 delete:372.3

Well, being that I wrote most of it, I guess I should explain. Here are the facts, from my fuzzy memory:

Thank you very much for the prompt reply! I wish I could go solid state (one of my workstations is -- nice!!).

But it doesn't seem like it should take 6 minutes to delete all that data from the tables. Am I wrong in that one? I do see that the retrieval of the data from the tables was relatively prompt in relation to the rest of the stats, but do you have any suggestions how I can reduce that delete time? The load average on my server still shoots up to 15 - 20 (Linux box, 8 core, 3 GB RAM) when it's processing the RRD files. The difference is that it's only that high for 20 - 30 minutes instead of all the time, which is was before Boost.

Thank you,
Matt

[Hyperlord]

Hey,

something for doku/wiki?
I'll copy your explaination to my personal docs anyway.


brgds

[TheWitness]

Hey,

something for doku/wiki?
I'll copy your explaination to my personal docs anyway.


brgds

Can you add it there too?

TheWitness

[TheWitness]

Make sure you table is memory type and that it does not include hash indexes.

Post create tabe for poller_output_boost and poller_output.

TheWitness

[mfarrenk]

Make sure you table is memory type and that it does not include hash indexes.

Post create tabe for poller_output_boost and poller_output.

TheWitness

I just ran the boost_sql_myisam.sql template, then tweaked the length of the output column. I used the myisam template because I was testing this on a development/semi-production machine and, not knowing for sure what the memory requirements would be, didn't want to risk taking up too much memory. The script was:

DROP TABLE IF EXISTS `poller_output_boost`;
CREATE TABLE `poller_output_boost` (
`local_data_id` mediumint(8) unsigned NOT NULL default '0',
`rrd_name` varchar(19) NOT NULL default '',
`time` datetime NOT NULL default '0000-00-00 00:00:00',
`output` varchar(512) NOT NULL,
PRIMARY KEY USING BTREE (`local_data_id`,`rrd_name`,`time`),
KEY `time_local_data_id` USING BTREE (`time`,`local_data_id`),
KEY `local_data_id` USING BTREE (`local_data_id`)
) ENGINE=MyISAM ROW_FORMAT=FIXED;

DROP TABLE IF EXISTS `poller_output_boost_processes`;
CREATE TABLE `poller_output_boost_processes` (
`sock_int_value` bigint(20) unsigned NOT NULL auto_increment,
`status` varchar(255) default NULL,
PRIMARY KEY (`sock_int_value`)
) ENGINE=MEMORY;


Doing some research on the Web and I'm finding information about changing tables from MyISAM to InnoDB, but it's not clear if it's possible to change a table from MyISAM to memory. Would I need to shut down Boost before I make such a change or can MySQL just do the conversion with the table live?

Thank you,
Matt

[TheWitness]

For this table, use memory.

TheWitness

[TheWitness]

Found a bug in MEMORY tables (yes MySQL bug). Don't use BTREE indexes for the primary key. I will be releasing Boost 4.1 shortly to address.

TheWitness

[mfarrenk]

Found a bug in MEMORY tables (yes MySQL bug). Don't use BTREE indexes for the primary key. I will be releasing Boost 4.1 shortly to address.

TheWitness

I'm currently running on 5.0.51a (Debian Lenny). Is this version affected or is it a 5.1 bug?

[TheWitness]

That version of MySQL has the bug. This is the same version on my home LinkStation Debian Lenny box that I first discovered this issue on.

TheWitness

[aleu]

That version of MySQL has the bug. This is the same version on my home LinkStation Debian Lenny box that I first discovered this issue on.

TheWitness, sorry if this is not the right place to ask some general question related to the boost plugin. My cacti installation 0.8.7e with spine 0.8.7e and boost v3.1 seems to be working great with a minor issue. When boost updates the RRD files (configured to do so every 4hrs or 750k records), every now and then my poller jumps to 300 seconds and I loose some data. It happens only during periods when boost updates the rrds, otherwise my poller time is around 60 seconds. I have configured it to process 500 data sources at once (I believe te dedefault setting is 2000).

Boost is configured with MyISAM.sql. I was wondering whether this is the cause (I am not running the boost_sql_memory.sql option because I was not able to tune it well and was getting frequent gaps in the graphs despite setting max_heap_table_size to something like 750MB.)
Could you please advise me on how to tweak my installation not to get occasional gaps when boost updates my rrds?

Also here: http://docs.cacti.net/plugin:boost you are saying that "It's also important that the poller_output table is converted to memory, eliminate any disk I/O related to poller updates." and then "alter table poller_output, modify column output varchar(50) not null default ””, engine=memory; " I do not think that my poller_output is configured with the memory switch at the end as I am seeing it being set by cacti.sql file w/o it (I do not know how to check this though). I am sure this is what is happening to me as my I/O system simply cannot keep up with the updates and polling at the same time. I will gladly provide you with any information that will help in fixing this.

Thanks,
AL

[TheWitness]

Boost 4.1 will correct this behavior.

YheWitness

[aleu]

Boost 4.1 will correct this behavior.

Will I be able to use boost 4.1 with Cacti 0.8.7e or willl it require upgrade to 0.8.7g?

[TheWitness]

I "think" it should work with 0.8.7e. However, I have been so wrapped up with the release of 0.8.7g, that I have lost track of things.

TheWitness