statsd-influxdb-backend

StatsD InfluxDB backend

A naive InfluxDB backend for StatsD.

It can ship events to InfluxDB using two different strategies which can be used at the same time.

Regular Flush Strategy

StatsD will flush aggregated metrics with a configured interval. This is the regular StatsD mode of operation.

Proxy Strategy

This will map every incoming StatsD packet to an InfluxDB event. It's useful if you want to store the raw events in InfluxDB without any rollups.

CAVEATS

This is pretty young and I do not have much experience with InfluxDB yet. Especially the event buffering and the event mapping might be problematic and inefficient.

InfluxDB is also pretty young and there might be breaking changes until it reaches 1.0.

Please be careful!

Installation

$ cd /path/to/statsd
$ npm install statsd-influxdb-backend

Configuration

You can configure the following settings in your StatsD config file.

{
  graphitePort: 2003,
  graphiteHost: "graphite.example.com",
  port: 8125,
  backends: [ "./backends/graphite", "statsd-influxdb-backend" ],
 
  influxdb: {
    host: '127.0.0.1',   // InfluxDB host. (default 127.0.0.1)
    port: 8086,          // InfluxDB port. (default 8086)
    version: 0.8,        // InfluxDB port. (default 0.8)
    ssl: false,          // InfluxDB is hosted over SSL. (default false)
    database: 'dbname',  // InfluxDB database instance. (required)
    username: 'user',    // InfluxDB database username.
    password: 'pass',    // InfluxDB database password.
    flush: {
      enable: true       // Enable regular flush strategy. (default true)
    },
    proxy: {
      enable: false,       // Enable the proxy strategy. (default false)
      suffix: 'raw',       // Metric name suffix. (default 'raw')
      flushInterval: 1000  // Flush interval for the internal buffer.
                           // (default 1000)
    },
    includeStatsdMetrics: false, // Send internal statsd metrics to InfluxDB. (default false)
    includeInfluxdbMetrics: false // Send internal backend metrics to InfluxDB. (default false)
                                  // Requires includeStatsdMetrics to be enabled.
  }
}

Activation

Add the statsd-influxdb-backend to the list of StatsD backends in the config file and restart the StatsD process.

{
  backends: ['./backends/graphite', 'statsd-influxdb-backend']
}

Unsupported Metric Types

Proxy Strategy

• Counter with sampling.
• Signed gauges. (i.e. bytes:+4|g)
• Sets

InfluxDB Event Mapping

StatsD packets are currently mapped to the following InfluxDB events. This is a first try and I'm open to suggestions to improve this.

Set

StatsD package client_version:1.1|c, client_version:1.2|c as Influx event:

[
  {
    name: 'visior',
    columns: ['value', 'time'],
    points:  [['1.1', 1384798553000], ['1.2', 1384798553001]]
  }
]

If you are using Grafana to visualize a Set, then using this query or something similar

SELECT version, count(version) FROM client_version GROUP BY version, time(1m)

Also, to count for the size of unique value, another InfluxDB event is also pushed

[
  {
    name: 'visitor_count',
    columns: ['value', 'time'],
    points:  [set.length, 1384798553001]
  }
]

Counter

StatsD packet requests:1|c as InfluxDB event:

Flush Strategy

[
  {
    name: 'requests.counter',
    columns: ['value', 'time'],
    points: [[802, 1384798553000]]
  }
]

Proxy Strategy

[
  {
    name: 'requests.counter.raw',
    columns: ['value', 'time'],
    points: [[1, 1384472029572]]
  }
]

Timing

StatsD packet response_time:170|ms as InfluxDB event:

Flush Strategy

[
  {
    name: 'response_time.timer.mean_90',
    columns: ['value', 'time'],
    points: [[445.25761772853184, 1384798553000]]
  },
  {
    name: 'response_time.timer.upper_90',
    columns: ['value', 'time'],
    points: [[905, 1384798553000]]
  },
  {
    name: 'response_time.timer.sum_90',
    columns: ['value', 'time'],
    points: [[321476, 1384798553000]]
  },
  {
    name: 'response_time.timer.std',
    columns: ['value', 'time'],
    points: [[294.4171159604542, 1384798553000]]
  },
  {
    name: 'response_time.timer.upper',
    columns: ['value', 'time'],
    points: [[998, 1384798553000]]
  },
  {
    name: 'response_time.timer.lower',
    columns: ['value', 'time'],
    points: [[2, 1384798553000]]
  },
  {
    name: 'response_time.timer.count',
    columns: ['value', 'time'],
    points: [[802, 1384798553000]]
  },
  {
    name: 'response_time.timer.count_ps',
    columns: ['value', 'time'],
    points: [[80.2, 1384798553000]]
  },
  {
    name: 'response_time.timer.sum',
    columns: ['value', 'time'],
    points: [[397501, 1384798553000]]
  },
  {
    name: 'response_time.timer.mean',
    columns: ['value', 'time'],
    points: [[495.6371571072319, 1384798553000]]
  },
  {
    name: 'response_time.timer.median',
    columns: ['value', 'time'],
    points: [[483, 1384798553000]]
  }
]

Proxy Strategy

[
  {
    name: 'response_time.timer.raw',
    columns: ['value', 'time'],
    points: [[170, 1384472029572]]
  }
]

Gauges

StatsD packet bytes:123|g as InfluxDB event:

Flush Strategy

[
  {
    name: 'bytes.gauge',
    columns: ['value', 'time'],
    points: [[123, 1384798553000]]
  }
]

Proxy Strategy

[
  {
    name: 'bytes.gauge.raw',
    columns: ['value', 'time'],
    points: [['gauge', 123, 1384472029572]]
  }
]

Proxy Strategy Notes

Event Buffering

To avoid one HTTP request per StatsD packet, the InfluxDB backend buffers the incoming events and flushes the buffer on a regular basis. The current default is 1000ms. Use the influxdb.proxy.flushInterval to change the interval.

This might become a problem with lots of incoming events.

The payload of a HTTP request might look like this:

[
  {
    name: 'requests.counter.raw',
    columns: ['value', 'time'],
    points: [
      [1, 1384472029572],
      [1, 1384472029573],
      [1, 1384472029580]
    ]
  },
  {
    name: 'response_time.timer.raw',
    columns: ['value', 'time'],
    points: [
      [170, 1384472029570],
      [189, 1384472029572],
      [234, 1384472029578],
      [135, 1384472029585]
    ]
  },
  {
    name: 'bytes.gauge.raw',
    columns: ['value', 'time'],
    points: [
      [123, 1384472029572],
      [123, 1384472029580]
    ]
  }
]

Backend Metrics

The following internal metrics are calculated for each flush:

• statsd.influxdbStats.flush_time - Time taken to process a complete flush in ms. Excluding the asynchronous HTTP Post.
• statsd.influxdbStats.http_response_time - Response time in ms of the InfluxDB HTTP endpoint when POSTing data.
• statsd.influxdbStats.payload_size - The size in bytes of the JSON payload.
• statsd.influxdbStats.num_stats - The number of metrics sent to InfluxDB in the last flush.

These are added to the set of internal statsd metrics. If both influxdb.includeStatsdMetrics and influxdb.includeInfluxdbMetrics are enabled, then these will be sent to InfluxDB when using the flush strategy.

The internal metrics can also can be viewed using the stats command on the StatsD TCP Admin Interface

Contributing

All contributions are welcome: ideas, patches, documentation, bug reports, complaints, and even something you drew up on a napkin.