Elasticsearch (2.3.3) cluster on top of Kubernetes made easy.
Elasticsearch best-practices recommend to separate nodes in three roles:
Masternodes - intended for clustering management only, no data, no HTTP APIClientnodes - intended for client usage, no data, with HTTP APIDatanodes - intended for storing and indexing your data, no HTTP API
Given this, I'm hereby making possible for you to scale as needed. For instance, a good strong scenario could be 3 master, 2 client, 5 data nodes.
Attention: As of the moment, Kubernetes pod descriptors use an emptyDir for storing data in each data node container. This is meant to be for the sake of simplicity and should be adapted according to your storage needs.
- Kubernetes cluster (tested with v1.2.4 on top of Vagrant + CoreOS)
kubectlconfigured to access your cluster master API Server
Providing your own version of the images automatically built from this repository will not be supported. This is an optional step. You have been warned.
kubectl create -f service-account.yaml
kubectl create -f es-discovery-svc.yaml
kubectl create -f es-svc.yaml
kubectl create -f es-master-rc.yaml
Wait until es-master is provisioned, and
kubectl create -f es-client-rc.yaml
Wait until es-client is provisioned, and
kubectl create -f es-data-rc.yaml
Wait until es-data is provisioned.
Now, I leave up to you how to validate the cluster, but a first step is to wait for containers to be in the Running state and check Elasticsearch master logs:
$ kubectl get svc,rc,pods
NAME CLUSTER-IP EXTERNAL-IP PORT(S) AGE
elasticsearch 10.100.202.107 9200/TCP 10m
elasticsearch-discovery 10.100.99.125 <none> 9300/TCP 10m
kubernetes 10.100.0.1 <none> 443/TCP 21h
NAME DESIRED CURRENT AGE
es-client 1 1 5m
es-data 1 1 4m
es-master 1 1 7m
NAME READY STATUS RESTARTS AGE
es-client-5b1oc 1/1 Running 0 5m
es-data-0s6eg 1/1 Running 0 4m
es-master-tile7 1/1 Running 0 7m
$ kubectl logs -f es-master-tile7
(...)
[2016-06-08 15:07:22,363][INFO ][node ] [Stranger] version[2.3.3], pid[172], build[b9e4a6a/2016-04-21T16:03:47Z]
[2016-06-08 15:07:22,370][INFO ][node ] [Stranger] initializing ...
[2016-06-08 15:07:23,373][INFO ][plugins ] [Stranger] modules [reindex, lang-expression, lang-groovy], plugins [cloud-kubernetes], sites []
[2016-06-08 15:07:23,452][INFO ][env ] [Stranger] using [1] data paths, mounts [[/data (/dev/sda9)]], net usable_space [14.1gb], net total_space [15.5gb], spins? [possibly], types [ext4]
[2016-06-08 15:07:23,455][INFO ][env ] [Stranger] heap size [1015.6mb], compressed ordinary object pointers [true]
[2016-06-08 15:07:28,088][INFO ][node ] [Stranger] initialized
[2016-06-08 15:07:28,089][INFO ][node ] [Stranger] starting ...
[2016-06-08 15:07:28,233][INFO ][transport ] [Stranger] publish_address {10.244.101.2:9300}, bound_addresses {10.244.101.2:9300}
[2016-06-08 15:07:28,239][INFO ][discovery ] [Stranger] myesdb/6c9o-8CyStefdhCfmsGkIg
[2016-06-08 15:07:32,714][INFO ][cluster.service ] [Stranger] new_master {Stranger}{6c9o-8CyStefdhCfmsGkIg}{10.244.101.2}{10.244.101.2:9300}{data=false, master=true}, reason: zen-disco-join(elected_as_master, [0] joins received)
[2016-06-08 15:07:32,721][INFO ][node ] [Stranger] started
[2016-06-08 15:07:32,763][INFO ][gateway ] [Stranger] recovered [0] indices into cluster_state
[2016-06-08 15:09:16,291][INFO ][cluster.service ] [Stranger] added {{Tower}{Er6hmt5yTlO3N4HhpWl-Tg}{10.244.19.2}{10.244.19.2:9300}{data=false, master=false},}, reason: zen-disco-join(join from node[{Tower}{Er6hmt5yTlO3N4HhpWl-Tg}{10.244.19.2}{10.244.19.2:9300}{data=false, master=false}])
[2016-06-08 15:10:39,119][INFO ][cluster.service ] [Stranger] added {{Skywalker}{bY9BsKYmRhqcXpcqMDFKvw}{10.244.64.2}{10.244.64.2:9300}{master=false},}, reason: zen-disco-join(join from node[{Skywalker}{bY9BsKYmRhqcXpcqMDFKvw}{10.244.64.2}{10.244.64.2:9300}{master=false}])
As you can assert, the cluster is up and running. Easy, wasn't it?
Scaling each type of node to handle your cluster is as easy as:
kubectl scale --replicas 3 rc/es-master
kubectl scale --replicas 2 rc/es-client
kubectl scale --replicas 2 rc/es-data
Did it work?
$ kubectl get rc,pods
NAME DESIRED CURRENT AGE
es-client 2 2 7m
es-data 2 2 6m
es-master 3 3 9m
NAME READY STATUS RESTARTS AGE
es-client-5b1oc 1/1 Running 0 7m
es-client-t44y2 1/1 Running 0 29s
es-data-0s6eg 1/1 Running 0 6m
es-data-3i5kh 1/1 Running 0 15s
es-master-ctfjz 1/1 Running 0 1m
es-master-tile7 1/1 Running 0 9m
es-master-tk14l 1/1 Running 0 1m
Let's take another look at the logs of one of the Elasticsearch master nodes:
$ kubectl logs -f es-master-tile7
(...)
[2016-06-08 15:07:22,363][INFO ][node ] [Stranger] version[2.3.3], pid[172], build[b9e4a6a/2016-04-21T16:03:47Z]
[2016-06-08 15:07:22,370][INFO ][node ] [Stranger] initializing ...
[2016-06-08 15:07:23,373][INFO ][plugins ] [Stranger] modules [reindex, lang-expression, lang-groovy], plugins [cloud-kubernetes], sites []
[2016-06-08 15:07:23,452][INFO ][env ] [Stranger] using [1] data paths, mounts [[/data (/dev/sda9)]], net usable_space [14.1gb], net total_space [15.5gb], spins? [possibly], types [ext4]
[2016-06-08 15:07:23,455][INFO ][env ] [Stranger] heap size [1015.6mb], compressed ordinary object pointers [true]
[2016-06-08 15:07:28,088][INFO ][node ] [Stranger] initialized
[2016-06-08 15:07:28,089][INFO ][node ] [Stranger] starting ...
[2016-06-08 15:07:28,233][INFO ][transport ] [Stranger] publish_address {10.244.101.2:9300}, bound_addresses {10.244.101.2:9300}
[2016-06-08 15:07:28,239][INFO ][discovery ] [Stranger] myesdb/6c9o-8CyStefdhCfmsGkIg
[2016-06-08 15:07:32,714][INFO ][cluster.service ] [Stranger] new_master {Stranger}{6c9o-8CyStefdhCfmsGkIg}{10.244.101.2}{10.244.101.2:9300}{data=false, master=true}, reason: zen-disco-join(elected_as_master, [0] joins received)
[2016-06-08 15:07:32,721][INFO ][node ] [Stranger] started
[2016-06-08 15:07:32,763][INFO ][gateway ] [Stranger] recovered [0] indices into cluster_state
[2016-06-08 15:09:16,291][INFO ][cluster.service ] [Stranger] added {{Tower}{Er6hmt5yTlO3N4HhpWl-Tg}{10.244.19.2}{10.244.19.2:9300}{data=false, master=false},}, reason: zen-disco-join(join from node[{Tower}{Er6hmt5yTlO3N4HhpWl-Tg}{10.244.19.2}{10.244.19.2:9300}{data=false, master=false}])
[2016-06-08 15:10:39,119][INFO ][cluster.service ] [Stranger] added {{Skywalker}{bY9BsKYmRhqcXpcqMDFKvw}{10.244.64.2}{10.244.64.2:9300}{master=false},}, reason: zen-disco-join(join from node[{Skywalker}{bY9BsKYmRhqcXpcqMDFKvw}{10.244.64.2}{10.244.64.2:9300}{master=false}])
[2016-06-08 15:15:46,512][INFO ][cluster.service ] [Stranger] added {{Norman Osborn}{Mxptb3y3Qp6R6xq9eddEzg}{10.244.19.3}{10.244.19.3:9300}{data=false, master=true},}, reason: zen-disco-join(join from node[{Norman Osborn}{Mxptb3y3Qp6R6xq9eddEzg}{10.244.19.3}{10.244.19.3:9300}{data=false, master=true}])
[2016-06-08 15:15:47,184][INFO ][cluster.service ] [Stranger] added {{Deathlok}{OQEV0LTSQh2Q1k7WVNCUBw}{10.244.64.3}{10.244.64.3:9300}{data=false, master=true},}, reason: zen-disco-join(join from node[{Deathlok}{OQEV0LTSQh2Q1k7WVNCUBw}{10.244.64.3}{10.244.64.3:9300}{data=false, master=true}])
[2016-06-08 15:16:12,750][INFO ][cluster.service ] [Stranger] added {{Hitman}{KW69CqVrQwyFbB1mbndK5w}{10.244.101.3}{10.244.101.3:9300}{data=false, master=false},}, reason: zen-disco-join(join from node[{Hitman}{KW69CqVrQwyFbB1mbndK5w}{10.244.101.3}{10.244.101.3:9300}{data=false, master=false}])
[2016-06-08 15:16:34,865][INFO ][cluster.service ] [Stranger] added {{Trapster}{vXWqivMrRKefxWaf_FJW6g}{10.244.19.4}{10.244.19.4:9300}{master=false},}, reason: zen-disco-join(join from node[{Trapster}{vXWqivMrRKefxWaf_FJW6g}{10.244.19.4}{10.244.19.4:9300}{master=false}])
Don't forget that services in Kubernetes are only acessible from containers in the cluster. For different behavior you should configure the creation of an external load-balancer. While it's supported within this example service descriptor, its usage is out of scope of this document, for now.
$ kubectl get svc elasticsearch
NAME CLUSTER-IP EXTERNAL-IP PORT(S) AGE
elasticsearch 10.100.202.107 9200/TCP 13m
From any host on your cluster (that's running kube-proxy), run:
curl http://10.100.202.107:9200
You should see something similar to the following:
{
"name" : "Hitman",
"cluster_name" : "myesdb",
"version" : {
"number" : "2.3.3",
"build_hash" : "b9e4a6acad4008027e4038f6abed7f7dba346f94",
"build_timestamp" : "2016-06-02T16:03:47Z",
"build_snapshot" : false,
"lucene_version" : "5.5.0"
},
"tagline" : "You Know, for Search"
}Or if you want to see cluster information:
curl http://10.100.202.107:9200/_cluster/health?pretty
You should see something similar to the following:
{
"cluster_name" : "myesdb",
"status" : "green",
"timed_out" : false,
"number_of_nodes" : 7,
"number_of_data_nodes" : 2,
"active_primary_shards" : 0,
"active_shards" : 0,
"relocating_shards" : 0,
"initializing_shards" : 0,
"unassigned_shards" : 0,
"delayed_unassigned_shards" : 0,
"number_of_pending_tasks" : 0,
"number_of_in_flight_fetch" : 0,
"task_max_waiting_in_queue_millis" : 0,
"active_shards_percent_as_number" : 100.0
}Note: If you are running Elasticsearch in a non-default namespace then you need to create two service accounts for both the default namespace and the custom one; this is especially important if the ability to scale data nodes is a requirement!