Load feedback and check agent for Haproxy



Herald is a load feedback and check agent for Haproxy.

The agent is configured in Haproxy using the agent-check server option. Once enabled haproxy periodically (agent-inter) connects to the backend on the configured port (agent-port) and acts on the response.

The response can either indicate an action to perform, such as MAINT, UP, DRAIN, etc or change the weight of the respective backend with a percentage, say 75%. Check Haproxy documentation here for more details.

There are many use cases for using the agent, load feedback being an obvious one.


  • Flexible and easy to configure
  • Extensible using plugins
  • High performance


We recommend using pip to install in a virtualenv :

$ virtualenv herald
$ pip install haproxy-herald


In your haproxy backend add the agent-check configuration like so :

frontend myservice
        bind *:8004
        maxconn 20000
        option tcp-smart-accept
        default_backend myservers

backend myservers
        balance leastconn
        option tcp-smart-connect
        fullconn 20000
        server myserver01 myserver01:8081       maxconn 10000 weight 100 check agent-check agent-port 5555
        server myserver02 myserver02:8081       maxconn 10000 weight 100 check agent-check agent-port 5555
        server myserver03 myserver03:8081       maxconn 10000 weight 100 check agent-check agent-port 5555

This instructs haproxy to connect to agent on port 5555.

On the backend servers, run herald with the following configuration :

name: myservice
port: 5555
plugins_dir: /etc/herald/plugins
  - name: myservice_plugin
    herald_plugin_name: herald_http
    url: 'http://localhost:8081/health_check'
    is_json: yes
    stop_timeout: 60
    interval: 30
    staleness_interval: 120
    staleness_response: noop
    thresholds_metric: "r['msg-pre-second']"
      - pct: 7000
        min_threshold_response: 1
    default_response: noop

Herald reads from /etc/herald/config.yml by default. The plugins are bundled in herald.plugins, which can be symlinked in /etc/herald/plugins.

Check the example_config.yml file for detailed configuration options.

With this configuration, herald will poll the health check url every 30s. Note that the response is also cached to avoid hitting the health check url too often.


The herald_http and herald_file plugins are provided in herald/plugins directory. These should serve most use cases. The plugin code is simple and easy to follow; Writing additional plugins should be very easy.

The following features are provided by the plugin framework :

  • Check result cacheing
  • Json parsing and processing parsed data using any python dict expression
  • Arthmetic expressions on the result
  • Calculate weight percentage on the result
  • Regex pattern matching on the result


  • Unit and integration tests
  • Improved logging
  • Configuration reload on SIGHUP