graphite-cluster

Dockerized components to run a graphite cluster. Composed of:

• graphite-node: A series of carbon relays and a single graphite api, meant to be run on a single host.
• graphite-relay: A carbon relay node. If running more than one carbon cache, the cluster will require a relay node.
• graphite-api (TBD): A graphite api node. If running multiple graphite-node instances, the cluster will require an api node.

Single Node, Single Carbon Cache

The simplest deployment. Good for testing, or if expected metrics load is small. In this scenario, only a single graphite-node is needed:

docker run -d -e CARBON_CONFIG={\"cache_count\":1} -v /opt/graphite/data:/opt/graphite/storage/whisper -p 80:80 -p 2013:2013 -p 2014:2014 ainsleyc/graphite-cluster-node

• You'll probably want to bind the docker whisper path to a local path on your host with the '-v' option so that the data won't be lost when the docker is shut down.
• The docker uses the default graphite/carbon ports, which you can bind to the host with the '-p' flag:
  • Carbon line port (2013)
  • Carbon pickle port (2014)
  • Graphite webapp (80)

You can confirm that the node is working by visiting the api page (http://host:80) and checking if carbon/agents data is appearing in the local host path bound with the '-v' option.

Single Node, Multiple Carbon Caches

For larger write loads you can use multiple carbon caches. A common setup is a single cache for each cpu on the host. To run the node:

docker run -d -e CARBON_CONFIG={\"cache_count\":4} -v /opt/graphite/data:/opt/graphite/storage/whisper -p 80:80 -p 2014:2014 -p 2024:2024 -p 2034:2034 -p 2044:2044 ainsleyc/graphite-cluster-node

In this setup a relay is needed, so to cache pickle ports need to be exposed with the '-p' option. The pickle ports for multiple caches follow the pattern: [2014, 2024, 2034, ...]

To run the relay:

docker run -d -e CARBON_CONFIG={\"cache_hosts\":[{\"host\":\"192.168.0.10\"\,\"ports\":[2014\,2024\,2034\,2044]}]} -p 2003:2003 -p 2004:2004 ainsleyc/graphite-cluster-relay

• It's important that the "cache_count" parameter matches between the node and relay.
• The loopback address (127.0.0.1) will not work in the "host" field.
• The ports provided should match the ports exposed in the cache node, order must match.
• The docker uses the default carbon ports, which should be exposed with the '-p' option:
  • Carbon line port (2003)
  • Carbon pickle port (2004)

You can confirm that the relay is working by checking if carbon/relays data is appearing in the local host path bound with the '-v' option.

Multiple Node, Multiple Carbon Caches

TBD