This Docker image provides an easily configurable Elasticsearch node. Via port mappings, it is easy to create an arbitrarily sized cluster of nodes. As long as the versions match, you can mix-and-match "real" Elasticsearch nodes with container-ized ones.
To start an Elasticsearch data node that listens on the standard ports on your host's network interface:
docker run -d -p 9200:9200 -p 9300:9300 itzg/elasticsearch
You'll then be able to connect to the Elasticsearch HTTP interface to confirm it's alive:
{
"status" : 200,
"name" : "Charon",
"version" : {
"number" : "1.3.5",
"build_hash" : "4a50e7df768fddd572f48830ae9c35e4ded86ac1",
"build_timestamp" : "2014-11-05T15:21:28Z",
"build_snapshot" : false,
"lucene_version" : "4.9"
},
"tagline" : "You Know, for Search"
}
Where DOCKERHOST would be the actual hostname of your host running Docker.
To run a multi-node cluster (3-node in this example) on a single Docker machine use:
docker run -d --name es0 -p 9200:9200 es
docker run -d --name es1 --link es0 -e UNICAST_HOSTS=es0 es
docker run -d --name es2 --link es0 -e UNICAST_HOSTS=es0 es
and then check the cluster health, such as http://192.168.99.100:9200/_cluster/health?pretty
{
"cluster_name" : "elasticsearch",
"status" : "green",
"timed_out" : false,
"number_of_nodes" : 3,
"number_of_data_nodes" : 3,
"active_primary_shards" : 0,
"active_shards" : 0,
"relocating_shards" : 0,
"initializing_shards" : 0,
"unassigned_shards" : 0
}
9200- HTTP REST9300- Native transport
/data- location ofpath.data/conf- location ofpath.conf
The following configuration options are specified using docker run environment variables (-e) like
docker run ... -e NAME=VALUE ... itzg/elasticsearch
Since Docker's -e settings are baked into the container definition, this image provides an extra feature to change any of the settings below for an existing container. Either create/edit the file env in the /conf volume mapping or edit within the running container's context using:
docker exec -it CONTAINER_ID vi /conf/env
replacing CONTAINER_ID with the container's ID or name.
The contents of the /conf/env file are standard shell
NAME=VALUE
entries where NAME is one of the variables described below.
Configuration options not explicitly supported below can be specified via the OPTS environment variable. For example, by default OPTS is set with
OPTS=-Dnetwork.bind_host=_non_loopback_
NOTE: That option is a default since bind_host defaults to localhost as of 2.0, which isn't helpful for
port mapping out from the container.
If joining a pre-existing cluster, then you may need to specify a cluster name different than the default "elasticsearch":
-e CLUSTER=dockers
When joining a multi-physical-host cluster, multicast may not be supported on the physical network. In that case, your node can reference specific one or more hosts in the cluster via the Zen Unicast Hosts capability as a comma-separated list of HOST:PORT pairs:
-e UNICAST_HOSTS=HOST:PORT[,HOST:PORT]
such as
-e UNICAST_HOSTS=192.168.0.100:9300
You can install one or more plugins before startup by passing a comma-separated list of plugins.
-e PLUGINS=ID[,ID]
In this example, it will install the Marvel plugin
-e PLUGINS=elasticsearch/marvel/latest
Many more plugins are available here.
Since the container gives the Elasticsearch software an isolated perspective of its networking, it will most likely advertise its published address with a container-internal IP address. This can be overridden with a physical networking name and port using:
-e PUBLISH_AS=DOCKERHOST:9301
Author Note: I have yet to hit a case where this was actually necessary. Other than the cosmetic weirdness in the logs, Elasticsearch seems to be quite tolerant.
Rather than use the randomly assigned node name, you can indicate a specific one using:
-e NODE_NAME=Docker
If you refer to the Node section of the Elasticsearch reference guide, you'll find that there's three main types of nodes: master-eligible, data, and client.
In larger clusters it is important to dedicate a small number (>= 3) of master nodes. There are also cases where a large cluster may need dedicated gateway nodes that are neither master nor data nodes and purely operate as "smart routers" and have large amounts of CPU and memory to handle client requests and search-reduce.
To simplify all that, this image provides a TYPE variable to let you amongst these combinations. The choices are:
- (not set, the default) : the default node type which is both master-eligible and a data node
MASTER: master-eligible, but holds no data. It is good to have three or more of these in a large clusterDATA(orNON_MASTER) : holds data and serves search/index requests. Scale these out for elastic-y goodness.GATEWAY: only operates as a client node or a "smart router". These are the ones whose HTTP port 9200 will need to be exposed
A Docker Compose file will serve as a good example of these three node types:
version: '2'
services:
gateway:
image: itzg/elasticsearch
environment:
UNICAST_HOSTS: master
TYPE: GATEWAY
ports:
- "9200:9200"
master:
image: itzg/elasticsearch
environment:
UNICAST_HOSTS: gateway
TYPE: MASTER
MIN_MASTERS: 2
data:
image: itzg/elasticsearch
environment:
UNICAST_HOSTS: master,gateway
TYPE: DATA
In combination with the TYPE variable above, you will also want to configure the minimum master nodes to avoid split-brain during network outages.
The minimum, which can be calculated as (master_eligible_nodes / 2) + 1, can be set with the MIN_MASTERS variable.
Using the Docker Compose file above, a value of 2 is appropriate when scaling the cluster to 3 master nodes:
docker-compose scale master=3
When using Docker Swarm mode (starting with 1.12), the overlay and ingress network interfaces are assigned
multiple IP addresses. As a result, it creates confusion for the transport publish logic even when using
the special value _eth0_.
To resolve this, add
-e DISCOVER_TRANSPORT_IP=eth0
replacing eth0 with another interface within the container, if needed.
The same can be done for publish/binding of the http module by adding:
-e DISCOVERY_HTTP_IP=eth2