Slide 1

Slide 1 text

Wednesday, November 20, 13

Slide 2

Slide 2 text

Heavy Lifting ZUHAIB SIDDIQUE • @ZUHAIB • ENGINEER Wednesday, November 20, 13

Slide 3

Slide 3 text

Wednesday, November 20, 13

Slide 4

Slide 4 text

1. Persistent chat rooms Wednesday, November 20, 13

Slide 5

Slide 5 text

2. File sharing & Chat history Wednesday, November 20, 13

Slide 6

Slide 6 text

3. @mentions + notifications Wednesday, November 20, 13

Slide 7

Slide 7 text

4. Integrations and API Wednesday, November 20, 13

Slide 8

Slide 8 text

5. Runs on everything Wednesday, November 20, 13

Slide 9

Slide 9 text

5. Runs on everything Wednesday, November 20, 13

Slide 10

Slide 10 text

5. Runs on everything Wednesday, November 20, 13

Slide 11

Slide 11 text

Group chat and IM for teams. Wednesday, November 20, 13

Slide 12

Slide 12 text

1,000,000,000 messages sent 1,200,000,000 Wednesday, November 20, 13

Slide 13

Slide 13 text

{ "id": "b0952da8-faa1-48f7-b1df-a8b2fb24bdfe", "from": { "nick": "Zuhaib Siddique", "group_id": 1, "user_id": 1 }, "room": { "group_id": 1, "name": "testing", "id": 1 }, "stanza_type": "muc_message", "date": "2013-11-14T23:20:13Z 424967", "stanza_data": { "body": "ElasticSearch FTW" } } What is a HipChat Message? Wednesday, November 20, 13

Slide 14

Slide 14 text

In the beginning there was … Wednesday, November 20, 13

Slide 15

Slide 15 text

Then we added... https://github.com/rnewson/couchdb-lucene Wednesday, November 20, 13

Slide 16

Slide 16 text

CouchDB + CouchDB-Lucene •Started to tip over at 200M Docs •m2.4xlarge Wednesday, November 20, 13

Slide 17

Slide 17 text

CouchDB + CouchDB-Lucene m2.4xlarge m2.4xlarge Worked well for the first 500M Wednesday, November 20, 13

Slide 18

Slide 18 text

m2.4xlarge hs1.8xlarge Wednesday, November 20, 13

Slide 19

Slide 19 text

But it still wasn’t big enough. Wednesday, November 20, 13

Slide 20

Slide 20 text

Dont F*ck the Customer Wednesday, November 20, 13

Slide 21

Slide 21 text

First Attempt at Scaling Wednesday, November 20, 13

Slide 22

Slide 22 text

Wednesday, November 20, 13

Slide 23

Slide 23 text

Why ElasticSearch for HipChat Wednesday, November 20, 13

Slide 24

Slide 24 text

Why ElasticSearch for HipChat { "query_string" : { "default_field" : "content", "query" : "this AND that OR thus" } } Wednesday, November 20, 13

Slide 25

Slide 25 text

Why ElasticSearch for HipChat Wednesday, November 20, 13

Slide 26

Slide 26 text

Search Error Rate Wednesday, November 20, 13

Slide 27

Slide 27 text

Wednesday, November 20, 13

Slide 28

Slide 28 text

Don’t Forget about CouchDB AWS Max EBS size Wednesday, November 20, 13

Slide 29

Slide 29 text

Stop the CouchDB bleeding Wednesday, November 20, 13

Slide 30

Slide 30 text

CouchDB end of the road Wednesday, November 20, 13

Slide 31

Slide 31 text

Wednesday, November 20, 13

Slide 32

Slide 32 text

(allthethings) ElasticSearch { "_index": "muc-2013.11", "_type": "document", "_id": "b0952da8-faa1-48f7-b1df-a8b2fb24bdfe", "_version": 1, "exists": true, "_source": { "from": { "nick": "Zuhaib Siddique", "group_id": 1, "user_id": 1 }, "room": { "group_id": 1, "name": "testing", "id": 1 }, "stanza_type": "muc_message", "date": "2013-11-14T23:20:13Z 424967", "_id": "b0952da8-faa1-48f7-b1df-a8b2fb24bdfe", "stanza_data": { "body": "ElasticSearch FTW" } } } { "id": "b0952da8-faa1-48f7-b1df-a8b2fb24bdfe", "from": { "nick": "Zuhaib Siddique", "group_id": 1, "user_id": 1 }, "room": { "group_id": 1, "name": "testing", "id": 1 }, "stanza_type": "muc_message", "date": "2013-11-14T23:20:13Z 424967", "stanza_data": { "body": "ElasticSearch FTW" } } ElasticSearch CouchDB Wednesday, November 20, 13

Slide 33

Slide 33 text

(allthethings) ElasticSearch { "_index": "muc-2013.11", "_type": "document", "_id": "b0952da8-faa1-48f7-b1df-a8b2fb24bdfe", "_version": 1, "exists": true, "_source": { "from": { "nick": "Zuhaib Siddique", "group_id": 1, "user_id": 1 }, "room": { "group_id": 1, "name": "testing", "id": 1 }, "stanza_type": "muc_message", "date": "2013-11-14T23:20:13Z 424967", "_id": "b0952da8-faa1-48f7-b1df-a8b2fb24bdfe", "stanza_data": { "body": "ElasticSearch FTW" } } } { "id": "b0952da8-faa1-48f7-b1df-a8b2fb24bdfe", "from": { "nick": "Zuhaib Siddique", "group_id": 1, "user_id": 1 }, "room": { "group_id": 1, "name": "testing", "id": 1 }, "stanza_type": "muc_message", "date": "2013-11-14T23:20:13Z 424967", "stanza_data": { "body": "ElasticSearch FTW" } } ElasticSearch CouchDB Wednesday, November 20, 13

Slide 34

Slide 34 text

CouchDB vs ElasticSearch CouchDB Spike to 800ms! Avg 100ms with ES Wednesday, November 20, 13

Slide 35

Slide 35 text

master Running ElasticSearch clients HipChat Customer Thanks Philip O’Toole - Loggly Wednesday, November 20, 13

Slide 36

Slide 36 text

$ fab create_esindex:hostname=esindex-d4 Wednesday, November 20, 13

Slide 37

Slide 37 text

•Spin up EC2 Instance •Install Oracle Java 7 •Create Provision IOPs EBS •Install and configure ElasticSearch •??? •Profit! Behind the Scenes Wednesday, November 20, 13

Slide 38

Slide 38 text

# Elasticsearch nodes = partial_search(:node, "role:elasticsearch_server AND chef_environment:#{node.chef_environment}", :keys => { 'hostname' => [ 'hostname' ] }) if nodes.empty? Chef::Log.error "site.json - No Elasticsearch servers found" else es_nodes = nodes.map {|n| { 'host' => n['hostname'], 'port' => 9200 } } data["elasticsearch"] = { "connections" => es_nodes.sort_by {|n| n['host'] } } end Wednesday, November 20, 13

Slide 39

Slide 39 text

site.json "elasticsearch": { "connections": [ { “host": "esindex-c1", "port": 9200 }, ... { "host": "esindex-e3", "port": 9200 } ] } Wednesday, November 20, 13

Slide 40

Slide 40 text

You Wednesday, November 20, 13

Slide 41

Slide 41 text

Running Elasticsearch Wednesday, November 20, 13

Slide 42

Slide 42 text

Running ElasticSearch Single Node OOM! Wednesday, November 20, 13

Slide 43

Slide 43 text

Monitor, Monitor, Monitor! Wednesday, November 20, 13

Slide 44

Slide 44 text

Backup Wednesday, November 20, 13

Slide 45

Slide 45 text

100,000,000,000? To infinity and beyond! Wednesday, November 20, 13

Slide 46

Slide 46 text

Questions? Wednesday, November 20, 13