1. Robert Koletka

2. What is Cassandra
● Basically a key value store
● With some stuff.
● It is a NoSQL database that is
● Decentralized : No single point of failure
● Elastic : Linear Scalability
● Fault Tolerant : Replication
● Optimized for writes, reads don't do badly at all
though.

3. What is Cassandra
● Based on two papers
● Bigtable: Google
● Dynamo: Amazon
● Dynamo partitioning and replication
● Bigtable data model
● CAP Theorem
● Consistent NO
● Available YES
● Partition Tolerant YES

4. What is Cassandra
● Uses consistent hashing

5. Data Model
● Cluster
● Keyspace : like a DB
● Column Families : like a Table
● Super Columns (optional)
● Columns
● Values

6. Data Model
● Keyspace groups column families together
● Column Family groups data together
● Example :
● User Keyspace has
– UserProfiles Column Family
– Friends Column Family

7. Data Model
● Cassandra doesn't require schema's like
traditional DB's
● UserProfiles Example
● Rk = {Name:Robert, Surname:Koletka,
Gender:Male}
● Js = {Name:John, Surname:Smith, Location:WC}
● Rk & Js both valid entries in UserProfiles
Column Family even though different columns.

8. Data Model
● Think about QUERIES not de-normalizing data.
● Use Case: “I want to get friends name's and
surname's for a given UserID”
● Name & Surname needs to be in the friend
column family.
● Js = {Ac:{Name:Alice,Surname:Cook}, Bb:
{Name:Betty,Surname:Blah}} (Super)
● Js = {Ac:”Alice Bob”, Bb:”Betty Blah”}

9. Data Model
● Column
● Rowkey = {ColumnName:Value,CN=V,CN=V}
● Super Column
● Rowkey = {SuperColumnName:{CN=V,CN=V},
SCN:{CN=V,CN=V}}
● Super Columns group columns together
● Cannot index on a Sub column.

10. Define Keyspace
● create keyspace <keyspace> with <att1>=<value1> and
<att2>=<value2> ...;
● create keyspace UserKeyspace with placement_strategy
= 'org.apache.cassandra.locator.SimpleStrategy' and
strategy_options = {replication_factor:2};
● Simple Strategy – place replica on next node
● NetworkTopologyStrategy – for multiple data centers
● OldNetworkToplogyStrategy – different data centers and
different racks

11. Define Column Family
● create column family <name> with
<att1>=<value1> and <att2>=<value2>...;
● create column family UserProfiles with
comparator=UTF8Type and
default_validation_class=UTF8Type and
key_validation_class=UTF8Type and
column_metadata=[{column_name:Location,
validation_class:UTF8Type,
Index_Type:KEYS}];

12. Define Column Family
● Comparator = Column Name validator and
compare column names
● default_validation_class = Validation for values
in columns which are not listed in
column_metadata
● key_validation_class = Validate key
● Default is BytesType

13. Define Column Family
● Other Available Types
● AsciiType
● BytesType
● CounterColumnType (distributed counter column, a CF
either contains counters or non at all)
● Int32Type
● IntegerType (a generic variable-length integer type)
● LexicalUUIDType
● LongType
● UTF8Type

14. Define Column Family
● Many more options
● bloom_filter_fp_chance : false positives
● gc_grace : garbage collection
● keys_cached
● row_cache_save_period
● max_compaction_threshold
● ...

15. Read and Writes
● Cassandra is optimized for writes
● First written to a commitlog
● Then to an in-memory table (memtable)
● Then periodically written to disk (SStable)
● Reads
● Read from all SStables and memtables
● Bloom filters used to speed up Sstable lookups
● Compaction
● Periodically Cassandra merges SStables

16. Indexes
● Row Keys
● Cassandra keeps and index of its Row keys
● Column Indexes
● Known as Secondary Indexes, build an index on
column values.
● Indexes existing data in the background
● Query by using equality predicates
– Then additional filters

17. Indexes
● Get userprofiles where location = 'WC'
● Get userprofiles where location = 'WC' and age
> 18
● NOT
● Get userprofiles where age > 18

18. Consistency
● Allows for configurable consistency settings
● Read
– One, Quorum((Replication Factor / 2) +1), Local/Each_Quorum
(Data Centers), All.
● Write
– Any, One, Quorum, Local/Each_Quorum (Data Centers), All.
● Any means that data can be written to co-ordinator if
replica's are down till replica's come back up.
● Quorum allows for some consistency and tolerating
some failures.
● All replica's must be up.

19. Consistency
● Read
● At least one node needs to be up to read data from,
obvious.
● Reads from a number of replicas returning the
latest data, based on timestamp.
● Read repair ensures data remains consistent,
updates out of date nodes with latest data. Runs in
background.

20. Cassandra Query Language
● Allows for
● Select
– SELECT [FIRST N] [REVERSED] <SELECT EXPR> FROM <COLUMN FAMILY> [USING
<CONSISTENCY>] [WHERE <CLAUSE>] [LIMIT N];
– SELECT [FIRST N] [REVERSED] name1..nameN FROM
– Unlike SQL, no guarantee that columns will be returned
– SELECT ... WHERE KEY >= startkey and KEY =< endkey AND name1 = value1
● Insert
● Delete
● Update
● Batch
● Truncate
● Create Keyspace
● Create Column Family
● Create Index
● Drop

21. Other Stuff
● Cassandra stores columns in sorted order
● Allows you to get the first or last X number of columns
● Potentially store historical user data
● Single column cannot hold more than 2gb
● Max number of columns per row is 2 billion
● Key and Column Names must be <64kb
● Most Languages have client libraries (Python,
Java, Scala, Node.js, PHP, C++...)
● Try not to use raw thrift.

22. Last Example
● User Statuses
● Columns stored in sorted order... use timestamp as column
name
● Rk = {1:'Good morning all',2:'lunch was good',3:'time to get
drunk',4:'so many regrets from last night'}
● Create column family UserStatuses with comparator =
LongType and Key_validation_class=UTF8Type and
default_validation_class=UTF8Type
● Get last X number of Columns, Get first X number of
columns