Apache Cassandra. Inception - all you need to know by Mikhail Dubkov
- 2. Why Cassandra ? Architecture & Design Use cases Summary Q & A Agenda
- 9. DataStax Java Driver Astyanax (Netflix) ODBC Driver with SQL Connector Hector (NO LONGER ACTIVE ) Driver
- 10. value1 value2 value3 JohnDoe * 27abc@d.com column1 column3column2 user nameageemail Row key user123 Table (Column Family) KEYSPACE Data Model
- 11. Usual types : boolean, text, int, bigint ( long ), float, double, decimal , blob Additional types : counter, varint, inet(IPv4 or IPv6), timestamp, timeuuid, uuid Collection types : list, set, map Data Types
- 12. Basic component of Cassandra Main configuration cassandra.yaml Restart after changing cassandra.yaml Node
- 13. key value ts Commit log Memtable Write flow
- 14. key value ts Flusher Thread SSTable1 SSTable2 SSTable3 SSTable4 Memtable Background Write flow
- 15. key value ts Flusher Thread SSTable1 SSTable2 SSTable3 SSTable4 Memtable Compaction Thread SSTable5 Background Write flow
- 16. Writes new record if not exists Writes record with more recent timestamp if exists UPDATE leads to new record if not exists Upsert
- 17. Delete / TTL DELETE -> Tombstone TTL -> Tombstone Compaction removes Tombstone’s data
- 18. Tombstone Tombstones live time defined by gc_grace_seconds Tombstone’s limit (100K) Lot Tombstones -> slow compaction
- 19. result SSTable6 SSTable7 SSTable5 Memtable Read flow Bloom filter
- 21. PRIMARY KEY ( country ) PRIMARY KEY ( country, state ) PRIMARY KEY ( (country, state), city ) * Partition key Row key
- 25. Consistency level Tunable ( 1, 2, 3, Quorum, All etc. ) Strong consistency ( R + W > N ) Weak consistency ( R + W <= N )
- 26. RP: 3 A E B C D CL: TWO Strong consistency WRITE
- 27. RP: 3 A E B C D CL: TWO Strong consistency READ
- 28. RP: 3 A E B C D CL: TWO Weak consistency WRITE
- 29. RP: 3 A E B C D CL: ONE Weak consistency READ
- 30. RP: 3 A E C D CL: ONE Node E store hint for Node B Fall down WRITE Hinted Handoff
- 31. RP: 3 A E C D Fall down Alive Hinted Handoff
- 32. RP: 3 A E B C D CL: TWO Read repair READ
- 33. Cache Row cache Partition key cache
- 34. Row cache Not write-through Off-heap Misuse exhausts the JVM heap
- 35. Partition key cache Where the partition is located on disk Decreasing seek times Enabled by default
- 36. CQL ALTER ( keyspace, table, type, user ) CREATE (keyspace, table, trigger, user ) SELECT INSERT UPDATE DELETE TRUNCATE DROP ( keyspace, table, trigger, user ) GRANT REVOKE
- 37. CREATE TABLE user ( user_id int, . . . group text, PRIMARY KEY(user_id) ); CREATE INDEX ON user (group) ; SELECT * FROM user WHERE group = 'specified_group’ ; Secondary index
- 39. CREATE TABLE car_owner ( group text , owner text, model text, . . . PRIMARY KEY ( group, owner )); group | owner | model major | Ben Affleck | Ferrari major | Vin Diesel | Bugatti minor | Sebastian Druid | Opel CREATE INDEX ON car_owner ( model ) ; SELECT * FROM car_owner where group = 'major' and model= 'Ferrari'; Secondary index
- 41. Association Table Mapping CREATE TABLE email_to_user_id ( email text, user_id int, PRIMARY KEY(email)) CREATE TABLE email_to_user ( email text, user_name text, . . . PRIMARY KEY(email))
- 42. Error handling NoHostAvailableException UnavailableException ( Write | Read ) TimeoutException
- 44. RF: 3 E B D Error handling A UnavailableException CL: ALL R1 R2 R3
- 45. RF: 3 E B D Error handling A ( Write | Read ) TimeoutException CL: ALL R1 R2 R3 C
- 46. Use cases
- 47. Product catalogs & Playlists
- 48. Finance
- 49. Event log
- 51. Social network
- 54. Apple – 75 000+ Nodes 10s of PBs Millions ops/s Netflix – 90+ Clusters 2700+ Nodes >1 Trillion ops/day eBay – 100+ Clusters 20+ Billion read/writes per day Motivation
- 55. www.macys.com 4 nodes 99.9 % availability De facto motivation Macy‘s
- 56. Q & A