Firefly is a web application aimed at powerful, flexible time series graphing for web developers.
Firefly provides an interface for reading time series data from any number of server-side data stores and produces corresponding real-time graphs in a web browser. Multiple metrics can combine into a single graph, and graphs combine together in grids to form a Dashboard - which you can use to get a high-level view of your services.
The current release ships with a DataSource that reads RRD files - specifically, those produced by Ganglia. However it's simple to expand to other sources, from an in-house relational database to a cross-network API.
- Line, stacked, and area graphs
- Configure graphs with arbitrary numbers of data sources
- Configure grids of graphs -- great for creating dashboards of related information
- View graphs from multiple datacenters in the same dashboard
- Show historical overlays along with your real-time data
- Log-scale Y-axis
- Support for isolated and embedded graphs
- Native API support for annotations
Firefly is written in Python and requires Python 2.6 or greater. In addition you will need libcurl-dev, librrd-dev, and librtmp-dev after which you can simply run
pip install -e ..
Firefly's configuration is formatted entirely as YAML. YAML is pretty easy to pick up, but you'll still want to be familiar with the YAML Spec if you are not already.
The core of Firefly is the abstract concept of a DataSource. A DataSource just needs to expose a certain interface (methods to list available metrics, and get the data for a particular metric between particular timestamps) - beyond that, the system is datastore-agnostic and should be widely expandable. We've included a DataSource for parsing Ganglia RRD files out of the box. But Firefly is not just a front-end for Ganglia, and in time we'll be looking to release more of the metrics gathering tools we use behind Firefly.
For Ganglia support, the rrdcached integration wiki page might be of use, as Firefly will work best if it speaks to
rrdcached to protect your hard disks from getting hammered.
To get up and running immediately, make sure you update your submodules:
git submodule update --init. Then simply run
python -m firefly.main --testing -c firefly.yaml.example from your Firefly checkout. The configuration option is now required due to the secret key being stored in it.
For more help:
python -m firefly.main --help or take a peek in
Note: Some configuration options can only be specified in your YAML configuration file.
Firefly is divided into two parts: a data server and a UI server. How you deploy these parts of Firefly will depend on how your existing deployments are structured.
The UI server is a thin wrapper that mostly serves static content, so you'll likely only run one of them. The data server will be run wherever you need a local process to access stored metrics. For instance, if you have Ganglia's RRD files stored on a single server in each of two datacenters, you'd want to run data servers on those same machines. The UI server can be configured to know about both data servers, and the whole system will work from a unified interface. Additionally, the UI server can be behind HTTP Basic Auth, and will use a mildly secure short-lived token to talk to the data servers - a necessity, since browsers don't elegantly support Basic Auth dialogs when doing CORS AJAX requests.
In your data server configuration files, you will want to specify
omit_ui_server: true. This will instruct Firefly to only start a data server and not a UI server. You will also want to be sure that the RRD socket and storage locations are set appropriately for any data sources you configure, including Ganglia. You may also wish to change the port of the data server if you are running it behind a reverse proxy.
You will need to specify each data server you set up in the
data_servers section of the
ui_server configuration section. Data servers are specified by the URL they can be reached at in the
name attribute and a description of the data server's environment in the
If you are hosting Firefly on the same machine as other web services or running behind a reverse proxy, you might want to set
port to your desired values. By default Firefly runs with a url prefix of
/firefly/ when not in
Simply start each Firefly instance you have in your various environments with the appropriate configuration files:
python -m firefly.main -c <configuration file>
First, give yourself a base YAML configuration file.
cp firefly.yaml.example firefly.yaml. Firefly reads from firefly.yaml first and then overrides the values specified in this file with any command line options specified. This is allows you to flip various switches during development which you can then set later in your production config (configuration files specified with
-c will always override other command line options).
There are a few configuration options you'll want to fill in before you can start graphing. Firefly is divided into two parts: a data server and a ui server, each of which has its own set of configuration options. Below are the various options you'll want to set to get started from a fresh checkout.
If you have an accessible Ganglia instance running, you should set the location of the Ganglia RRD socket and storage in the
rrdcached_storage settings of
data_source_config: data_sources.ganglia_rrd.GangliaRRD: rrdcached_socket: "/path/to/your/ganglia/rrd/unix/domain/socket.sock" rrdcached_storage: "/path/to/your/ganglia/rrd/storage"
If you do not have Ganglia running, comment out this data source in the
data_sources: # - data_sources.ganglia_rrd.GangliaRRD - data_sources.stat_monster_rrd.StatMonsterRRD
If you want to run with any additional custom data sources, add them to the
data_sources section and provide the kwargs they will be passed in the
If you have Vagrant installed running
vagrant up will get you started. When the virtual machine is up you can point your browser to
http://localhost:8889. See Vagrantfile for the details of the setup.
python -m firefly.main -c <configuration file> --testing
This starts both a data server and a UI server in the same web server running on the local machine. The UI server is accessible on
localhost:8889 by default. Note that you do not need to have any data servers configured in the UI server for the UI server to know about the local data server running alongside it!
Firefly will also give you a test data source to use, which will produce a constant sine wave across all time periods.
Firefly supports annotating graphs with various events that you might be interested in. To add annotations simply send an HTTP
POST request to the
/add_annotation endpoint on each of your data servers that the annotation applies to. This endpoint expects four arguments in the
token- A token obtained from the UI server (
type- Specifies the type of annotation. Should be a single word with alphanumeric characters only. This is the text displayed on the graphs next to the annotation marker.
description- A description of the event this annotation is for.
time- The time the annotation occurred, as a floating point number of seconds since the epoch.
You can control where the data server and UI server put their SQLite database files with the
db_file configuration variable, which can be set for both