dreverri/appstats

Appstats collects raw event data and provides an API for reading and summarizing that data. Inspired by Cube but written in Erlang using LevelDB.

HTTP API

A very simple HTTP API has been provided to explore collected data.

Return a list of running sessions:

GET /sessions

["226635918053873426086712796756526707779"]

Return the first and last known time points for a running session:

GET /session/:session_id/timespan

{
  "start": 1368115003903,
  "stop": 1368139163500
}

Return the names of events collected during a particular period of time:

GET /session/:session_id/names?start=1367995259000&stop=1367995260000

[
  "riak.vnode_get.1164634117248063262943561351070788031288321245184",
  "riak.vnode_get.0",
  "riak.vnode_get.890602560248518965780370444936484965102833893376",
  "riak.vnode_get.1096126227998177188652763624537212264741949407232",
  "riak.vnode_get.1255977969581244695331291653115555720016817029120",
  "riak.vnode_get.1301649895747835411525156804137939564381064921088"
]

Return summary of data for all events during a period of time at a specified step interval:

GET /session/:session_id/data?start=1368139003903&stop=1368139063904&step=60000

[
  {
    "timestamp": 1368139003903,
    "data": {
      "riak.vnode_get.0": {
        "count": 115,
        "min": 36,
        "mean": 69.10434782608695,
        "max": 262
      },
      "riak.vnode_get.1164634117248063262943561351070788031288321245184": {
        "count": 144,
        "min": 33,
        "mean": 65.31944444444444,
        "max": 160
      },
      "riak.vnode_get.1324485858831130769622089379649131486563188867072": {
        "count": 140,
        "min": 35,
        "mean": 69.33571428571429,
        "max": 458
      }
    }
  },
  {
    "timestamp": 1368139063903,
    "data": {
      "riak.vnode_get.1164634117248063262943561351070788031288321245184": {
        "count": 148,
        "min": 31,
        "mean": 65.4391891891892,
        "max": 339
      },
      "riak.vnode_get.0": {
        "count": 150,
        "min": 26,
        "mean": 63.82,
        "max": 162
      },
      "riak.vnode_get.890602560248518965780370444936484965102833893376": {
        "count": 142,
        "min": 24,
        "mean": 66.90845070422536,
        "max": 233
      }
    }
  }
]

Return summary data for a particular event during a period of time at a specified step interval:

GET /session/:session_id/data?name=riak.vnode_get.0&start=1368139003903&stop=1368139063904&step=60000

[
  {
    "timestamp": 1368139003903,
    "data": {
      "riak.vnode_get.0": {
        "count": 115,
        "min": 36,
        "mean": 69.10434782608695,
        "max": 262
      }
    }
  },
  {
    "timestamp": 1368139063903,
    "data": {
      "riak.vnode_get.0": {
        "count": 150,
        "min": 26,
        "mean": 63.82,
        "max": 162
      }
    }
  }
]

Examples

StatsD Server

There is a basic statsd server implemented in the statsd_server directory. This application will listen on port 8125 for StatsD messages and store them in an Appstats session.

To start the server:

cd examples/statsd_server
make start

The collected data can be viewed in a browser at http://localhost:8080

Riak Worker

This example is a single riakc client worker repeatedly getting a random key from Riak. The example assumes there is a Riak instance listening for PB connections on "127.0.0.1:8087".

To start the example:

cd examples/riakc_example
make start

Data collected from the example can be viewed in a browser at http://localhost:8080

TODO

Better Visualizations

The current visualization uses Cubism to update the visuals in real time. The real time updates are neat but not necessary. Better, non-real time, visualizations should be developed (e.g. a heatmap of a metrics histogram).

It may be useful to look into using Vega

Queries

Support queries other than exact name match. Possible queries:

"riak.vnode_get.*" - group all vnode_get stats into a single metric per timestamp
"riak.vnode_get.*|riak.vnode_put.*" - return 2 stats per timestamp, each stat would group vnode_gets and vnode_puts

Retrieving names is slow

Retrieving names is slow.

More data over HTTP

Appstats computes much more information than is currently exposed over HTTP. This information should be exposed (e.g. percentiles, histograms).

dreverri / appstats