PDB - 9. lecture - Cassandra

Exam requirements

• Data model (column families, rows, columns), query patterns, use cases, representatives
• Cassandra: data model (keyspaces, tables, rows, columns), primary keys (partition key, clustering columns), column values (missing; empty; native data types, tuples, user-defined types; collections: lists, sets, maps; frozen mode), additional data (TTL, timestamp); CQL language: DDL statements: CREATE KEYSPACE (replication strategies), DROP KEYSPACE, USE keyspace, CREATE TABLE (column definitions, usage of types, primary key), DROP TABLE, TRUNCATE TABLE; native data types (int, varint, double, boolean, text, timestamp, …); literals (atomic, collections, …); DML statements: SELECT statements (SELECT, FROM, WHERE, GROUP BY, ORDER BY, and LIMIT clauses; DISTINCT modifier; selectors; non/filtering queries, ALLOW FILTERING mode; filtering relations; aggregates; restrictions on sorting and aggregation), INSERT statements (update parameters: TTL, TIMESTAMP), UPDATE statements (assignments; modification of collections: additions, removals), DELETE statements (deletion of rows, removal of columns, removal of items from collections)

Apache Cassandra

• Sloupcové databáze
• Apache Cassandra - cheatsheet
• additional notes from the lecture
  • row is a collection of columns
    • a column is a pair of column name and value (+ additional metadata like TTL, timestamp))
    • additional data are associated with atomic values and also with each element of collection values
  • even tuples and UDT (user defined types) are considered atomic
  • frozen mode = the nested collections are not accesible directly
    • e.g. for nested collections in maps, sets and lists (their values are not accessible directly)
    • for example, when the UDT or a collection is a part of primary key, it needs to be frozen (= saved as one blob, not as a sequence of elements/properties)
      • so the internal parts cannot be edited, just the whole UDT/collection could be replaced/deleted
  • there are no foreign keys
    • we can simulate them with our application logic (just store unique values that act as IDs)
  • TTL is implemented in various NoSQL systems
    • it could be turned off to keep the data forever
    • but it has a lot of use-cases
  • expressive power of CQL is lower than MongoDB, but higher than KV stores
  • Cassandra works even with multiple datacenters that exists on different geographic locations
    • and allows for different replication factors for each data center
    • NetworkTopology replication strategy
  • important is the primary key options (column-level primary keys etc.)