Requirements

This extension uses InfluxDB Line Protocol over HTTP to send metrics.

Since Sensu already uses eventmachine, you just have to ensure that em-http-request gem is present inside Sensu's embedded Ruby :

• em-http-request ruby gem

By default, this extension uses in memory metrics caching : it makes InfluxDB writes batched. This cache is flushed every 500 items or 6 seconds (this values can be changed in conf).

Caveat

• SSL/TLS connection to InfluxDB is still broken in Sensu 0.20.6. We're waiting for a Ruby / EM upgrade in Sensu project :
  • sensu/sensu#1084

Usage

The configuration options are pretty straight forward.

Metrics are inserted into the database using the check's key name as measurement name. So if you're using the sensu-plugins-load-checks community plugin :

my-host-01.load_avg.one 0.02 1444824197
my-host-01.load_avg.five 0.04 1444824197
my-host-01.load_avg.fifteen 0.09 1444824197

In this example, you'll have 3 differents measurements in your database :

> show measurements
name: measurements
------------------
name
my-host-01.load_avg.fifteen
my-host-01.load_avg.five
my-host-01.load_avg.one

> select * from "my-host-01.load_avg.one";
name: my-host-01.load_avg.one
------------------
time                  host        value  duration
2015-10-14T13:53:22Z  my-host-01  0.34   0.399 
2015-10-14T13:53:32Z  my-host-01  0.29   0.419
2015-10-14T13:53:42Z  my-host-01  0.39   0.392
2015-10-14T13:53:52Z  my-host-01  0.41   0.398
[...]

Additionally a duration value will be present based on the time it took the check to run (this is gleaned from the sensu event data).

The name of the measurement is based on the value of strip_metric described below. The name of the key host is grabbed from sensu event client name.

Extension not a handler

Note that the first push of this was a handler that could be called via pipe. This is now an actual extension that's more performant since it's actually in the sensu-server runtime. Additionally it's now using batch submission to InfluxDB by writing all the points for a given series at once.

Just drop the ruby file in /etc/sensu/extensions and create a set to wrap this extension into a callable handler. In this example, we created a metrics handler wrapping a debug output and this Influx extension :

/etc/sensu/conf.d/handlers/metrics.json :

{
  "handlers": {
    "metrics": {
      "type": "set",
      "handlers": [ "debug", "influxdb"]
      }
    }
  }
}

Note : Since Sensu 0.17.1 you can also use extension name directly :

Check definitions can now specify a Sensu check extension to run,
"extension", instead of a command.

Handler config

/etc/sensu/conf.d/influxdb.json

{
  "influxdb": {
    "database": "stats",
    "host": "localhost",
    "port": "8086",
    "username": "stats",
    "password": "stats",
    "use_ssl": false,
    "strip_metric": "somevalue",
    "tags": {
      "region": "my-dc-01",
      "stage": "prod"
    }
  }
}

Config attributes

• host, port, username, password and database are pretty straight forward. If use_ssl is set to true, the connection to the influxdb server will be made using https instead of http.

• tags hash is also pretty straight forward. Just list here in a flat-hash design as many influxdb tags you wish to be added in your measures.

• strip_metric however might not be. This is used to "clean up" the data sent to influxdb. Normally everything sent to handlers is akin to the graphite/stats style:

  something.host.metrictype.foo.bar
or
  host.stats.something.foo.bar

Really the pattern is irrelevant. People have different tastes. Adding much of that data to the column name in InfluxDB is rather silly so strip_metric provides you with a chance to add a value that strips off everything up to (and including that value). This allows you to continue sending to graphite or statsd or whatever and still use this handler.

Using the examples above, if you set the strip_metric to host, then the column in InfluxDB would be called metrictype.foo.bar or stats.something.foo.bar. If you set the value to foo then the column would simply be called foo

Note that :

• strip_metric isn't required.
• you can cleanup an arbitrary string from your keyname or use host as special value to cleanup the sensu event client name from your key.

Other attributes

• time_precision : global checks time precision (default is 's')
• buffer_max_size : buffer size limit before flush - This is the amount of points in the InfluxDB batch - (default is 500)
• buffer_max_age : buffer maximum age - Flush will be forced after this amount of time - (default is 6 seconds)

Check options

In the check config, an optional influxdb section can be added, containing a database option and tags. If specified, this overrides the default database option in the handler config and adds (or override) influxdb tags.

This allows events to be written to different influxdb databases and modify key indexes on a check-by-check basis.

You can also specify the time precision of your check script in the check config with the time_precision attribute.

Example check config

/etc/sensu/conf.d/checks/metrics-load.json

{
  "checks": {
    "metrics-load": {
      "type": "metric",
      "command": "metrics-load.rb",
      "standalone": true,
      "handlers": [
        "metrics"
      ],
      "interval": 60,
      "time_precision": "s",
      "influxdb": {
        "database": "custom-db",
        "tags": {
           "stage": "prod",
           "region": "eu-west-1"
        }
      }
    }
  }
}

Result :

load_avg.one,stage:prod,region:eu-west-1,host:iprint-test-sa-01.photobox.com value=1.04,duration=0.402  1444816792147
load_avg.five,stage:prod,region:eu-west-1,host:iprint-test-sa-01.photobox.com value=0.86,duration=0.398 1444816792147
load_avg.fifteen,stage:prod,region:eu-west-1,host:iprint-test-sa-01.photobox.com value=0.84,duration=0.375 1444816792147

 * will be sent to -> http://my-influx09.company.com:8086/db/custom-db/series?time_precision=s&u=sensu&p=sensu