Skip to content

db-art/mysql-statsd

Repository files navigation


Deprication

This project is no longer supported by Spil Games and has been adopted by DB-Art.

The new repository can be found here: MySQL-StatsD @ DB-Art

mysql-statsd

Daemon that gathers statistics from MySQL and sends them to statsd.

Usage / Installation

Install mysql_statsd through pip(pip is a python package manager, please don't use sudo!):

pip install mysql_statsd

If all went well, you'll now have a new executable called mysql_statsd in your path.

Running mysql_statsd

$ mysql_statsd --config /etc/mysql-statsd.conf

Assuming you placed a config file in /etc/ named mysql-statsd.conf

See our example configuration or read below about how to configure

Running the above command will start mysql_statsd in deamon mode. If you wish to see it's output, then run the command with -f / --foreground

Usage

$ mysql_statsd --help
usage: mysql_statsd.py [-h] [-c FILE] [-d] [-f]

optional arguments:
  -h, --help            show this help message and exit
  -c FILE, --config FILE
                        Configuration file
  -d, --debug           Prints statsd metrics next to sending them
  --dry-run             Print the output that would be sent to statsd without
                        actually sending data somewhere
  -f, --foreground      Dont fork main program

At the moment there is also a deamon script for this package

You're more than welcome to help us improve it!

Platforms

We would love to support many other kinds of database servers, but currently we're supporting these:

  • MySQL 5.1
  • MySQL 5.5
  • Galera

Both MySQL versions supported with Percona flavour as well as vanilla.

Todo:

Support for the following platforms

  • Mysql 5.6
  • MariaDB

We're looking forward to your pull request for other platforms

Development installation

To install package, setup a python virtual environment

Install the requirements(once the virtual environment is active):

pip install -r requirements.txt

NOTE: MySQL-Python package needs mysql_config command to be in your path.

There are future plans to replace the mysql-python package with PyMySQL

After that you're able to run the script through

$ python mysql_statsd/mysql_statsd.py

Coding standards

We like to stick with the python standard way of working: PEP-8

Configuration

The configuration consists out of four sections:

  • daemon specific (log/pidfiles)
  • statsd (host, port, prefixes)
  • mysql (connecitons, queries, etc)
  • metrics (metrics to be stored including their type)

Daemon

The daemon section allows you to set the paths to your log and pic files

Statsd

The Statsd section allows you to configure the prefix and hostname of the metrics. In our example the prefix has been set to mysql and the hostname is included. This will log the status.com_select metric to: mysql.<hostname>.status.com_select

You can use any prefix that is necessary in your environment.

MySQL

The MySQL section allows you to configure the credentials of your mysql host (preferrably on localhost) and the queries + timings for the metrics. The queries and timings are configured through the stats_types configurable, so take for instance following example:

stats_types = status, innodb

This will execute both the query_status and query_innodb on the MySQL server. The frequency can then be controlled through the time (in milliseconds) set in the interval_status and interval_innodb. The complete configuration would be:

stats_types = status, innodb
query_status = SHOW GLOBAL STATUS
interval_status = 1000
query_innodb = SHOW ENGINE INNODB STATUS
interval_innodb = 10000

A special case is the query_commit: as the connection opened by mysql_statsd will be kept open and auto commit is turned off by default the status variables are not updated if your server is set to REPEATABLE_READ transaction isolation. Also most probably your history_list will skyrocket and your ibdata files will grow fast enough to drain all available diskspace. So when in doubt about your transaction isolation: do include the query_commit!

Now here is the interesting part of mysql_statsd: if you wish to keep track of your own application data inside your application database you could create your own custom query this way. So for example:

stats_types = myapp
query_myapp = SELECT some_metric_name, some_metric_value FROM myapp.metric_table WHERE metric_ts >= DATE_SUB(NOW(), interval 1 MINUTE)
interval_myapp = 60000

This will query your application database every 60 seconds, fetch all the metrics that have changed since then and send them through StatsD. Obviously you need to whitelist them via the metrics section below.

Metrics

The metrics section is basically a whitelisting of all metrics you wish to send to Graphite via StatsD. Currently there is no possibilty to whitelist all possible metrics, but there is a special case where we do allow wildcarding: for the bufferpool_* we whitelist all bufferpools with that specific metric. Don't worry if you haven't configured multiple bufferpools: the output will be omitted by InnoDB and also not parsed by the preprocessor.

Important to know about the metrics is that you will have to specify what type they are. By default Graphite stores all metric equaly but treats them differently per type:

  • Gauge (g for gauge)
  • Rate (r for raw, d for delta)
  • Timer (t for timer)

Gauges are sticky values (like the spedometer in your car). Rates are the number of units that need to be translated to units per second. Timers are the time it took to perform a certain task.

An ever increasing value like the com_select can be sent various ways. If you wish to retain the absolute value of the com_select it is advised to configure it as a gauge. However if you are going to use it as a rate (queries per second) it is no use storing it as a rate in the first place and then later on calculate the integral of the gauge to get the rate. It would be far more accurate to store it as a rate in the first place.

Keep in mind that sending the com_select value as a raw value is in this case a bad habit: StatsD will average out the collected metrics per second, so sending within a 10 second timeframe 10 times a value of 1,000,000 will average out to the expected 1,000,000. However as the processing of metrics also takes a bit of time the chance of missing one beat is relatively high and you end up sending only 9 times the value, hence averaging out to 900,000 once in a while.

The best way to configure the com_select to a rate is by defining it as a delta. The delta metric will remember the metric as it was during the previous run and will only send the difference of the two values.

Media:

Art gave a talk about this tool at Percona London 2013: http://www.percona.com/live/mysql-conference-2013/sessions/mysql-performance-monitoring-using-statsd-and-graphite

Contributors

spil-jasper

thijsdezoete

art-spilgames

bnkr