Three Nodes and One Cluster: Demo of ElasticSearch's Distributed Features

.gitignore

.DS_Store

logs/
data/

*.html

01_create_index.rb

require 'tire'

Tire.configure { logger STDERR }

Tire.index('my_index') do
  delete
  create settings: { number_of_shards: 3, number_of_replicas: 0 }
end

02_import_documents.rb

STDOUT.sync = true

require 'tire'

Tire.configure { logger STDERR }

last_batch = 0
count      = (ENV['COUNT'] || 100_000).to_i
pids       = []

4.times do |process|
 pids << Process.fork do
   batch = count / 4
   STDOUT.puts "Generating #{batch} documents in process #{process+1}..."

   documents = (1..batch).map do |i|
     { id: "#{process}-#{i}-#{Time.now.to_i}-#{rand(10000)}", title: "Document" }
   end

   Tire.index('my_index') { import documents }

   STDOUT.puts "\r"+"Process #{process+1} stored \e[1m#{batch}\e[0m documents.\n"
 end
end

pids.each { |pid| Process.wait }

03_store_document.rb

require 'timeout'
require 'tire'

Tire.configure { logger STDERR }

Timeout::timeout(1) do
  Tire.index('my_index') do
    store( {title: "Hello Red Cluster!"} )
  end
end

04_change_replicas.rb

require 'tire'

r = (ARGV.pop || 1).to_i

puts %Q|curl -X PUT 'http://localhost:9200/my_index/_settings' -d '{"index" : { "number_of_replicas" : #{r} }}'|

Tire::Configuration.client.put "#{Tire::Configuration.url}/my_index/_settings",
                          { :index => { :number_of_replicas => r } }.to_json

__readme.markdown

Three Nodes and One Cluster: Demo of ElasticSearch's Distributed Features

The Setup

Terminal:

Tab for each node.

Browser:

1. Bigdesk
2. Paramedic

Cleanup:

curl -X DELETE localhost:9200

Scripts:

• 01_create_index.rb
• 02_import_documents.rb
• 03_store_document.rb
• 04_change_replicas.rb

The Script

[KAREL]  Start node 1
         Inspect cluster state in Paramedic

[KAREL]  Setup index (shards = 3, replicas=0)
         Inspect state and stats in Paramedic (CPU spike, shard distribution, split evenly across 3 nodes)

[LUKAS]  Start node 2
         Inspect cluster node stats in BigDesk

[KAREL]  Import 100,000 documents
         Paramedic, show CPU spike, HTTP connections increase, ...
         Count documents (90,000)
         
[SHAY]   Start node 3
         
[KAREL]  Inspect cluster state in Paramedic

[KAREL]  Import another 200,000 documents
         
[SHAY]   Kill node 3
         Paramedic, health is _red_
         Count documents (~ 60,000)
         (We are missing data, but the cluster is operational)
         
[KAREL]  Index new document, `{"title" : "Hello Red Cluster!"}`
         
[LUKAS]  Search for the newly added document, "hello"
         
[SHAY]   Start node 3
         Paramedic, health is _green_, data are back (300,001 documents)

         * How to add high availability? Add replicas (= copies of the index). *

[LUKAS]  Increase `index.number_of_replicas` to 1

         ... wait for relocation ...

[KAREL]  Paramedic, show shard distribution/allocation across shards
         (primaries and replicas)

[SHAY]   Kill node 3

[KAREL]  Paramedic, cluster state is _green_

[SHAY]   Launch node 3 again

[KAREL]  Show re-allocation of shards

         * How to force the shards from a node? Eg. to use it for other index, etc? With allocation excludes *

[LUKAS]  curl -XPUT localhost:9200/my_index/_settings -d '{ "index.routing.allocation.exclude._id" : "NODEID" }'

[KAREL]  Show nodes missing from node 3 in Paramedic

---

http://git.io/3n1c

elasticsearch.yml

cluster:
  name: berlin_buzzwords

node:
  name: "Node 1"

path:
  logs: ./logs
  data: ./data

elasticsearch.yml

cluster:
  name: berlin_buzzwords

node:
  name: "Node 2"

path:
  logs: ./logs
  data: ./data

elasticsearch.yml

cluster:
  name: berlin_buzzwords

node:
  name: "Node 3"

path:
  logs: ./logs
  data: ./data