Scalarium and CouchDB

Jonathan Weiss @jweiss

@scalarium

Peritor GmbH

Scalarium.com

EC2 Cluster Management  Automatic & dynamic configuration over lifetime

of a cluster via Chef runs

  Self healing clusters

 Auto scaling clusters

 One click deployment

 One click cluster cloning

 Monitoring & alerting

  Logs & accounting

  Firewalls & backups

 Access & rights management

2

Use case wooga.com

One of wooga’s Facebook games   1.5 million DAU (Daily Active Users)

  ~ 130 servers

  ~ 150.000 requests per minute on average

  Complex stack with different databases and replication

  Multiple deployments of the game on multiple clusters

3

Scalarium.com

4 http://www.flickr.com/photos/av8pix/3391322123/

Data

Rather static but important   Cluster configuration

  Server descriptions and current state

  Lifecycle events: setup, deploy, configure, shutdown

 Applications and deployment definitions

  EC2 assets like EBS, elastic IPs, SSH keys

5

Dynamic and non-critical  Accounting & events

 Monitoring and metering

Data

Rather static but important

6

Dynamic and non-critical

Why CouchDB?

 High Availability

 Easy Replication

 Clustering

 Robustness

 Short Recovery Time

 Jan Lehnardt

7

Architecture

8

Architecture

9

Interaction

10

SimplyStored

Ruby library with ActiveRecord inspired capabilities  Models

 Associations Generates map & reduce

  Validations

  Callbacks

 Dynamic finders

  S3 attachments

  Paranoid delete

http://github.com/peritor/simply_stored

11

SimplyStored

12

SimplyStored

13

Conflict & Resolution

14

Conflict & Resolution

Re-load and retry: Automatically done by SimplyStored

Discard and report: Internal report

15

RockingChair

In-memory CouchDB as a Ruby library   Just a big Hash

 Understands views generated by SimplyStored

  Speeds up your tests

  Tests can run in parallel

 Nice for debugging

http://github.com/jweiss/rocking_chair

16

Operations

17

Backup

Slaves in different EC2 availability zones and regions

Offsite copy of data and index files

18

Hot Standby

19

Cold Standby

20 http://www.flickr.com/photos/dnkemontoh/2680822579

Keeping views fresh

 Cronjobs for crawling all design docs and calling all views

 On master & standby!

 Rollout new views without using them first

 And while we are there, compact views&dbs

21

Performance

22

I/O

CouchDB heavily depends on file system I/O   EBS RAIDs for better performance

 Different volumes for data and indices

  Be generous with RAM & CPU

23 http://www.flickr.com/photos/walkn/5471945439

include_docs

 Nice feature to load related data in one go…but slow random I/O on read

 Emit the data you need to the view index

24

Design Documents

 Maintained in a framework/model-layer

 Using md5 of content to decide if update needed

 Whenever one view definition changed, all views got recomputed

25

Design Documents

Separate design document per view

Fine granular control  Update

 Delete

  Compact

26

One DB to rule them all

27

Separate DBs

28

Wishlist

 Build-in caching

 include_docs to load from index

 Improved performance of building views

 Build-in solution for

automatic sharding

 Automatic compaction

 Metrics & monitoring

29

Couchbase 2.0

30

Q&A

Talk to us if you want to use Scalarium 3 month for free

@scalarium

31