The slide deck I've consolidated & streamlined specifically for describing Riak, giving a quick overall high level overview & then jumping into building a cluster & using the database.
Metadata Users/Profiles Object Storage Session Storage Sensor Data Logging Systems Record Systems Notification Systems Use, when & what to use Riak for... Wednesday, March 13, 13
{“Key”:“Value”} • Values are stored against keys • Key/Value + Metadata = Object • Fundamental Unit of Replication • Any Datatype will work • Record to disk in binary format Wednesday, March 13, 13
INTERFACES HTTP API - Via a little piece of magic called Webmachine Protocol Buffers API - Thanks, Google! Largely-faithful REST implementation Compact, binary protocol Wednesday, March 13, 13
MapReduce Distributed processing system using Riak Pipe Efficient for targeted queries over known key range Write jobs in Erlang or JS. (Erlang more performant) Wednesday, March 13, 13
Secondary Indexing (2i) riak_object riak_object X-Riak-Index-email_bin X-Riak-Index-value_int “[email protected]” “42” Tag objects with custom metadata on PUT... Exact match and range queries... No multi-index queries yet... Pagination is on its way... Wednesday, March 13, 13
Riak Search Store and index documents (JSON, text, XML, etc) Current Riak Search supports subset of Solr API Next iteration (Yokozuna; in beta)will implement distributed Solr on Riak. It will be sexy. Looking for beta testers to help harden Yokozuna Wednesday, March 13, 13
ARCHITECTURE The scaleability and ease of operation goals inform architectural decisions. These come with tradeoffs. Consistent Hashing Virtual Nodes Append-only storage Handoff/Rebalancing Vector Clocks Active Anti-Entropy* Wednesday, March 13, 13
Consistent Hashing Location of data in the Riak ring is determined based on hash of bucket + key. Provides even distribution of storage and query load Trades off advantages gained from locality - e.g. Range queries and aggregates Wednesday, March 13, 13
Virtual Nodes Unit of addressing and concurrency in Riak Each physical host manage many vnodes Partition count / physical machines = vnodes/machine* Decouples physical assets from data distribution. This provides: - simplicity in cluster sizing - failure isolation Wednesday, March 13, 13
Handoff/Rebalancing Mechanisms for data rebalancing When nodes join/leave cluster, handoff and rebalancing manage the date shuffling dynamically Trades off speed of convergence vs. effects on cluster performance - causes disk & network load Wednesday, March 13, 13
Vector Clocks VCs used to rectify object consistency at READ time. Lots of knobs to turn; well-documented Trades off space, speed, and complexity for safety - will store all sibling objects until resolved - can lead to object size issues Wednesday, March 13, 13
Append-Only Storage Riak provides a pluggable backend interface. (Write your own; we’ll probably hire you...) Bitcask, LevelDB are most-heavily used. Both are append - only Provides crash safety and speed. Trade off: periodic compaction/merge ops Wednesday, March 13, 13
RIAK 1.3 (AKA “new hotness”) Active Anti Entropy MapReduce Improvements IPv6 Support Riaknostic included by default Much more Riak Control improvements Full release notes: https://github.com/basho/riak/blob/1.3/RELEASE-NOTES.md Wednesday, March 13, 13
FUTURE WORK* (1.4 and beyond) (* all code subject to ship early, late, or not at all) Dynamic Ring Size Yokozuna CRDTs/Data Types Riak Object Consistency 2i Improvements Riak Pipe work Much more Wednesday, March 13, 13
S3-API compatible and supports per-tenant reporting for billing and metering use cases. Additional APIs on the way. Multi-tenant cloud storage software for public and private clouds. Designed to provide simple, available, distributed cloud storage at any scale. Stores files of arbitrary size. Under the hood stores 1MB chunks along side a manifest. Stateless proxy (CS) does chunking. Riak does distribution, storage, etc. Wednesday, March 13, 13
Data transfer is unidirectional (source -> sink). Bidirectional synchronization can be achieved by configuring a pair of connections between clusters. Extends Riak's capabilities with: - multi-datacenter replication - SNMP Configuration - JMX-Monitoring - 24x7 support from Basho Engineers One cluster acts as a "source cluster". The source cluster replicates its data to one or more "sink clusters" using either real-time or full sync. Wednesday, March 13, 13
RIAK COMMUNITY Mailing List - 1300 developers IRC - 200+ people every day yelling about software GitHub - 1000s of watchers; 200+ contributors to all projects Meetups - 10 Countries, 23 Cities, 3700+ Members & growing fast! Deployments - 1000s in production. Wednesday, March 13, 13
May 13-14th in New York City ricon.io/east.html Talks, hacking, parties Dedicated to the future of Riak and distributed systems in production REGISTER NOW! https://ricon-east-2013.eventbrite.com/?discount=lovevnodes Wednesday, March 13, 13
adron
ikezoemakoto
hacomono
cyberarchitect
n1ru4l
line_developers
i524
nagix
masakamayama
yurihondo
tkmnzm
bengo4com
prashantspol
stephaniewalter
afnizarnur
maggiecrowley
rocio
phodgson
dougneiner
hursman
nonsquared
lauravandoore
shpigford
sferik
jensimmons