Introduction to Cassandra and Data Modeling
- 3. ©2012 DataStax Why does Cassandra Exist? 3 Thursday, May 30, 13
- 4. ©2012 DataStax Analytics + Real Time 4 Big Data Thursday, May 30, 13
- 7. ©2012 DataStax Why do people like Cassandra? 7 Thursday, May 30, 13
- 10. ©2012 DataStax 10 Thursday, May 30, 13
- 11. ©2012 DataStax Performance 11 Thursday, May 30, 13
- 12. ©2012 DataStax 12 Thursday, May 30, 13
- 13. ©2012 DataStax Multi Datacenter Support 13 Thursday, May 30, 13
- 14. ©2012 DataStax 14 Thursday, May 30, 13
- 15. ©2012 DataStax Hadoop Support 15 Thursday, May 30, 13
- 16. ©2012 DataStax Hadoop Support • InputFormat • Run tasktrackers/datanodes locally • Run namenode/jobtracker anywhere 16 Thursday, May 30, 13
- 17. ©2012 DataStax Data Locality Workload Partitioning 17 Thursday, May 30, 13
- 18. ©2012 DataStax Data Modeling 18 Thursday, May 30, 13
- 19. ©2012 DataStax Keyspace, Column Families 19 Thursday, May 30, 13
- 21. ©2012 DataStax Column Family = Row Key + Columns (name, value) ... 21 Thursday, May 30, 13
- 22. ©2012 DataStax Static Column Families Dynamic Column Families 22 Thursday, May 30, 13
- 23. ©2012 DataStax Static - Users Column Family 23 Row Key g_m_bluth password: banana stand name: George Michael tobias_f password: c_weathers name:Tobias phone: 512-7777 Thursday, May 30, 13
- 24. ©2012 DataStax Dynamic - Friend Column Family 24 Row Key g_m_bluth <date>:ann_v <date>:maeby tobias_f <date>:barry_z <date>:carl_w <date>:lindsay ... Thursday, May 30, 13
- 25. ©2012 DataStax Time Series Data • Event logs • Metrics • Sensor Data • Etc 25 Thursday, May 30, 13
- 26. ©2012 DataStax Time Series - Login CF 26 Row Key g_m_bluth 1369633061: United States 1369625839: Mexico ... tobias_f 1369932413: Canada 1369681738: United States ... Thursday, May 30, 13
- 27. ©2012 DataStax What Else? 27 Thursday, May 30, 13
- 28. ©2012 DataStax Counter Columns 28 • Inc/Dec operations • Not idempotent • Possibility for over counting Thursday, May 30, 13
- 29. ©2012 DataStax Expiring Columns 29 • TTL - Time to live • Set per column • Possibly an anti-pattern (we’ll get to that later) Thursday, May 30, 13
- 30. ©2012 DataStax Secondary Indexes 30 • Select * from Users where name=Nick; • Only support ‘=’ clauses (for first condition) • Often misused Thursday, May 30, 13
- 31. ©2012 DataStax CQL Cassandra Query Language 31 Thursday, May 30, 13
- 32. ©2012 DataStax 32 CREATE COLUMNFAMILY songs ( id uuid PRIMARY KEY, title text, album text, artist text, data blob); INSERT INTO songs (id, title, artist, album) VALUES ('a3e64f8f...', 'La Grange', 'ZZ Top', 'Tres Hombres'); SELECT * FROM songs; id | album | artist | title -------------+--------------+----------------+---------------- 2b09185b... | Roll Away | Back Door Slam | Outside Woman... 8a172618... | We Must Obey | Fu Manchu | Moving in Ste... a3e64f8f... | Tres Hombres | ZZ Top | La Grange Thursday, May 30, 13
- 33. ©2012 DataStax How do I start? 33 Thursday, May 30, 13
- 34. ©2012 DataStax Define your questions 34 Thursday, May 30, 13
- 35. ©2012 DataStax SELECT time, location FROM logins WHERE user = ‘nickmbailey’ ORDER BY time DESC LIMIT 10; 35 Thursday, May 30, 13
- 36. ©2012 DataStax WHERE user = ‘nickmbailey’ Row Key 36 Thursday, May 30, 13
- 37. ©2012 DataStax ORDER BY time DESC LIMIT 10; Store columns in chronological order 37 Thursday, May 30, 13
- 38. ©2012 DataStax CREATE COLUMN FAMILY logins ( ! user, time, location, PRIMARY KEY (user, time)); 38 Thursday, May 30, 13
- 39. ©2012 DataStax What about? 39 Thursday, May 30, 13
- 40. ©2012 DataStax SELECT time FROM logins WHERE user = ‘nickmbailey’ and location = ‘United States’; 40 Thursday, May 30, 13
- 41. ©2012 DataStax 41 g_m_bluth 1369633061: United States 1369625839: Mexico .... 1369622839: Canada 1369422839: Canada 1368422839: Canada .... 1368421839: Canada 1367421839: United States 1367411839: Mexico .... Thursday, May 30, 13
- 42. ©2012 DataStax CREATE COLUMN FAMILY logins (user, time, location, PRIMARY KEY (user, location)); 42 Thursday, May 30, 13
- 43. ©2012 DataStax 43 g_m_bluth United States: 1369633061 Canada: 1369622839 .... Thursday, May 30, 13
- 44. ©2012 DataStax To Normalize or Not 44 Thursday, May 30, 13
- 45. ©2012 DataStax SELECT time, location FROM..... + SELECT city, state, zip.... FROM locations..... 45 Thursday, May 30, 13
- 46. ©2012 DataStax 46 g_m_bluth 1369633061: <United States, Austin, Texas, 78701> 1369625839: <Mexico, Tiajuana, 88191> 1358633061: <United States,Austin, Texas, 78701> Thursday, May 30, 13
- 47. ©2012 DataStax Anti Patterns 47 Thursday, May 30, 13
- 48. ©2012 DataStax Batched Writes • Failure case is suboptimal • Increased chance of failure • Tune to your workload 48 Thursday, May 30, 13
- 49. ©2012 DataStax BOP/OPP • You don’t really need it • Your Ops Team will hate you • Really, you don’t need it. 49 Thursday, May 30, 13
- 50. ©2012 DataStax Super Columns • Performance penalty • Speed • Memory • Replaced by CQL3 50 Thursday, May 30, 13
- 51. ©2012 DataStax Read Before Write • Race conditions • Hurts performance • Cache • IO 51 Thursday, May 30, 13
- 52. ©2012 DataStax Queues • More generally, many deletes within a row • A delete in Cassandra is actually a tombstone • Read 1000 tombstones in order to find 10 columns 52 Thursday, May 30, 13
- 53. ©2012 DataStax Use Cases 53 Thursday, May 30, 13
- 54. ©2012 DataStax Ebay 54 Thursday, May 30, 13
- 55. ©2012 DataStax http://www.youtube.com/ watch?v=F-fYqPu2ciQ 55 Thursday, May 30, 13
- 56. ©2012 DataStax Ebay • dozens of nodes • 200 TB+ of storage 56 Thursday, May 30, 13
- 57. ©2012 DataStax Ebay • Social Signals • Hunch Taste Graph • Various Time Series 57 Thursday, May 30, 13
- 58. ©2012 DataStax Social Signals • Like, Own, Want • Need: • scalable counters • high performance writes • want to find most popular items in a given category 58 Thursday, May 30, 13
- 59. ©2012 DataStax Social Signals 59 Row Key item_id_1 like: 300 own:104 want:105 item_id_2 ... ... ... ItemCount Row Key user_id_1 like: 50 own:10 want:75 user_id_2 ... ... ... UserCount Thursday, May 30, 13
- 60. ©2012 DataStax Social Signals 60 Row Key item_id_1 user_id_1:<time> user_id_2:<time> ... item_id_2 ... ... ... ItemLike Row Key user_id_1 <time>: <item_id> <time>: <item_id> ... user_id_2 ... ... ... UserLike Thursday, May 30, 13
- 61. ©2012 DataStax Social Signals - Possibilities • Store aggregated counts per category • Column names are counts • Get top N items in a category 61 Thursday, May 30, 13
- 62. Questions? Thursday, May 30, 13
- 63. Come to the Summit! Ask me for a discount code June 11-12, 2013 San Francisco, CA http://www.datastax.com/company/news-and-events/events/ cassandrasummit2013 Thursday, May 30, 13