example_config.yml

---
name: api77
bind: 0.0.0.0
port: 5555
# plugins_dir: /etc/herald/plugins
plugins_dir: ./herald/plugins
# Load the plugins using dynamic import from the above dir
# Plugin is identified by using `herald_plugin_name`
#   1. That should be instantiated by passing the whole plugin hash as arg
#   2. Has a method called `run` which gets scheduled by the main program for
#      the respective interval
#   3. Has a method called `respond` which returns the response that should be
#      sent to the haproxy
#   4. Has a method called `stop` which will do plugin clean up
#   4. Implements caching may be using a common interface available to all plugins
#
# Caching abstraction may be written a common module/class that also takes care of staleness
# part of things.
# Main program calls respond on instance of any plugin that maps to agent-send if received from
# haproxy
# agent-send would be sending string in the following form:
# "<client_name>/<plugin_name>\n"
plugins:
  # name is used for caching keys in the agent, all in memory only
  # and also used with the agent level name to form a string that can
  # be matched against the string (agent-send) received from haproxy
  # for example, "api77/unique_name" in this case
  - name: api_service
    herald_plugin_name: herald_file
    file_path: /tmp/state
    # herald_plugin_name: herald_http
    # url: http://localhost:9000/health-check
    # indicate to the plugin that the results parsed using
    # json
    is_json: yes
    # Controls the timeout for plugin graceful shutdown
    stop_timeout: 10
    # How often to poll (seconds), 0 means poll request inline
    # and caching is disabled when interval is 0
    # interval: 10
    interval: 5
    # Mark things stale if no updates since (seconds)
    staleness_interval: 10
    # Other options for response are up, down, maint, drain and so on
    # noop means don't respond at all
    # staleness_response: down
    staleness_response: noop
    # pattern rules
    # keys from the response can be used in the patterns_metric or
    # thresholds_metric. expression needs to be evaualted in safe context
    # for security. The response is available in a special dict called `r`.
    #
    # could also have predefined vars available for in the expression
    patterns_metric: "r['health']"
    # first match wins
    patterns:
      - ready: '.*healthy.*'
      - down: '.*unhealthy.*'
      - drain: '.*maxed.*'
    thresholds_metric: "r['outgoing-messsage-per-second']"
    thresholds:
      # respond with "up" if less than 7000
      # - up: "<7000"
      # respond with "drain" if this threshold is met, i.e. greater than 7000
      - drain: ">7000"
      # respond with "down" if this threshold is met, that is equal to 0
      - down: 0
      # or any state
      # Calculate weight percentage based on this number and respond with that
      - pct: 7000
      # respond with min_threshold_response % when percentage is less than or
      # equal to 0% - defaults to 0. This is to avoid a scenario where a cluster
      # goes above capacity and all nodes receive 0% (which means DRAIN)
        min_threshold_response: 1
    # If none of the rules match, use default_response
    # 'noop' is a keyword to not reply with any response
    default_response: noop