ScalingRedisAndMemcachedAtWayfair.pdf

This is a talk I gave at Facebook's Data@Scale Boston in November 2014. Here's the talk on Youtube: https://www.youtube.com/watch?v=oLjryfUZPXU, and the accompanying blog post: http://engineering.wayfair.com/scaling-redis-and-memcached-at-wayfair/

Transcript

1. Scaling Redis and Memcached at Wayfair Ben Clark, Chief Architect November 2014
2. 2 Home goods, Boston, 5 retail brands, 5 countries
3. 3 NYSE:W, as of 10/2/2014
4. Background  Caching  Key-value stores  Consistent hashing at Akamai and after Scaling KV stores in the industry  Redis cluster  Talks, blogs, code by Instagram, Pinterest, Facebook, Twitter Scaling Redis and Memcached at Wayfair
5. Background
6. 6 Caching keys: web server, DB, cache Because the best database query is no query at all. Why and what?
7. 7 Key-value stores, 2003-present • 2003: Memcached, @bradfitz • 2009: Redis, @antirez • 2014: Redis cluster
8. 8 Consistent hashing, 1997-present Diagram from MIT paper, 1997 Keys {1,2,3} on server A, {4,5} on B Wayfair MC cluster, 2013 Down a server, and the right keys migrate
9. Scaling key-value stores in the industry
10. 10 Redis Cluster current state
11. 11 Composite caching systems in the industry Company Components (not comprehensive) Instagram Redis, ketama Pinterest Redis, ZooKeeper, S3 Facebook mcrouter proxying to Memcached, Tao
12. 12 Twitter’s forks and proxies Twemproxy as applied to a Wayfair web server: PHP-MC|Redis connections / 50 Before After
13. Scaling Redis and Memcached at Wayfair
14. 14 2011: Memcached, master and slave
15. 15 2012: Add Redis, ketama, slave writes to SAN
16. 16 2013: Memcached w/Akamai-style ketama
17. 17 2014: Double ketama rings, Twemproxy
18. 18 Configuration, health checks
19. 19 Tools we used, tools we hacked Stock tools Written or modified by us Memcached 1.4 Ketama (RJ) Redis 2.6 Twemproxy (Twitter) ZooKeeper 3.4.5 php-pecl-memcache Kazoo 1.2 (ZK client) zk_failover Phpredis, ‘redis’ on PyPI >=2.7 Utilities, wrapper classes, scripts, etc.
20. 20 Who
21. 21 TODO list • Hot keys (load balancing for the same key), mcrouter or Riak style • Faster Redis failover.
22. 22 References • “Scaling Redis at Twitter”, great talk in July 2014 by Yao Yu (@thinkingfish): https://www.youtube.com/watch?v=rP9EKvWt0zo • Opensourcing of mcrouter, September 2014: https://code.facebook.com/posts/296442737213493/introducing-mcrouter-a- memcached-protocol-router-for-scaling-memcached-deployments/ • Redis at Pinterest: http://blog.pivotal.io/pivotal/case-studies-2/using-redis-at- pinterest-for-billions-of-relationships • Mike Krieger of Instagram at atscale, Storage Systems at a Rapidly Scaling Startup with a Small Team - Data@Scale”: https://www.youtube.com/watch?v=bLyv8zKa5DU • https://github.com/reddit/zookeeper.dsc/blob/master/src/contrib/zkpython/src/p ython/zk.py • github.com/wayfair/ketama (fork of RJ/ketama) • Details on part of this talk: http://engineering.wayfair.com/consistent-hashing- with-memcached-or-redis-and-a-patch-to-libketama/
23. bclark@wayfair.com @clarkjacker, @WayfairEng github.com/wayfair http://engineering.wayfair.com