This is an alpha release of FnordMetric. Expect bugs and vulnerabilities.

Edit this page on GitHub

Getting Started with FnordMetric Server

This guide will walk you through starting a FnordMetric Server instance, inserting metric data and querying the metric data with ChartSQL. If you do not have installed FnordMetric yet, read the Installation page first.

Fnordmetric Server is a standalone HTTP server application. It exposes a web UI and a HTTP API to run ChartSQL queries and collect timeseries data.

You can start FnordMetric Server with or without a "storage backend". If FnordMetric Server is started without a storage backend you can only use the web interface to execute ChartSQL queries against external data sources (like a MySQL database). If it is started with a storage backend, you can also use the HTTP (and optionally the statsd) API to collect timeseries data into the storage backend and subsequently query that timerseries data using ChartSQL.

FnordMetric Server currently supports three storage backends: inmemory, disk, and hbase.

Starting Fnordmetric Server

For the getting started guide we will use the disk backend which stores the metric data in a folder on the local hard disk. To start a FnordMetric server instance with a local disk storage backend on HTTP port 8080 run:

$ mkdir -p /tmp/fnordmetric-data
$ fnordmetric-server --http_port 8080 --statsd_port 8125 --storage_backend disk --datadir /tmp/fnordmetric-data

Collecting Timeseries Data

FnordMetric Server records timeseries data in "Metrics". A Metric is somewhat equivalent to a table in a regular SQL database. Each metric has a unique name and consists of a collection of data points called "samples" that are recorded over time (i.e. a timeseries).

A "sample" is a single datapoint. Each sample contains at least a timestamp and a numeric value. To keep the table analogy, each metric is a table that has two default columns value and time and each sample is a row in that table.

You can query metrics using ChartSQL like normal tables:

> select time, value from mymetric;

| time                  |  value  |
| 2014-11-08 20:30:12   |  0.913  |
| 2014-11-08 20:30:42   |  0.837  |
| 2014-11-08 20:31:13   |  0.638  |
| 2014-11-08 20:31:41   |  0.326  |
| ...                   |  ...    |

As an example, we will monitor the http response times of a fictional web application. We will create one "metric" that will be called http_response_times and will record the http response times (latencies) of our application. We are going to insert a sample into this metric for each HTTP request that our web application serves.

There are a number of client libraries that allow you to send samples to FnordMetric Server using the HTTP or statsd API. For now, lets cheat a bit and manually send samples from the command line. The simplest way to send samples from your command line is using the statsd API. If you started FnordMetric Server on port 8125 (see above), you can use the netcat utility to send a sample via UDP+statsd.

Let's insert the value 42 into the http_response_times metric. In our example this means we record the response time of a single HTTP request that took 42ms.

$ echo "http_response_times:42" | nc -u -w0 8125

Execute this command a few times with different values to insert multiple samples into the metric. FnordMetric Server will automatically create the metric if it doesn't exist yet.

Execute Queries from the Web Interface

You should now be able to navigate to the admin interface on http://localhost:8080/ in your browser and see our newly created metric. Click on the metric to bring up the interactive query editor. This should look something like this:

Click around a bit to make yourself familiar with the Web UI. The charts you see above are generated using ChartSQL. The queries are automatically generated by the interactive query editor. Here is an example query:

Display the 100 latest samples from http_response_times as a line chart with the sample time plotted on the X axis and the sample value plotted on the Y axis.:


SELECT time as x, value as y
    FROM http_response_times
    ORDER BY time DESC
    LIMIT 100;

Adding labels

To allow you to drill down into your metric data in arbitrary dimensions, each sample can optionally be labelled with one or more "labels". Each label is a key: value pair.

In our example, assume we run our web application on multiple hosts in different datacenters. It would be nice to label each sample with hostname=... and datacenter=... so that we can roll up the http response times by host, datacenter or a combinaton of both.

Let's insert a few more example samples into our http_response_times metric and attach these labels:

$ echo "http_response_times[hostname=machine82][datacenter=ams1]:18" | nc -u -w0 8125
$ echo "http_response_times[hostname=machine83][datacenter=ams1]:42" | nc -u -w0 8125
$ echo "http_response_times[hostname=machine84][datacenter=ams1]:23" | nc -u -w0 8125

When querying metrics with ChartSQL, the label keys act as table columns so you can filter and aggregate/group by label values. Our http_response_times table now has 4 columns:

> select time, value, hostname, datacenter from http_response_times;

| time                  | value  | hostname   | datacenter  |
| 2014-11-08 20:30:12   | 18     | machine82  | ams1        |
| 2014-11-08 20:30:12   | 42     | machine83  | ams1        |
| 2014-11-08 20:30:12   | 23     | machine84  | ams1        |
| ...                   | ...    | ...        | ...         |

You can execute this query from the interactive query editor to display the last hour of samples in the http_response_times metric rolled up by hostname. It will draw a line chart with the sample time plotted on the X axis and the sample value plotted on the Y axis and one series per hostname:


SELECT hostname as series, time as x, value as y
    FROM http_response_times
    WHERE time > -1hour;

The result should look something like this:

You now have a running FnordMetric Server, but there is a lot more you can do. These are good docs to read next:

This guide will walk you through building a simple HTML5 dashboard using ChartSQL and FnordMetric Server: