Skip to content

Latest commit

 

History

History
63 lines (47 loc) · 1.86 KB

datamodeling-in-cassandra-brad.md

File metadata and controls

63 lines (47 loc) · 1.86 KB
Raw

Matthew F. Dennis

DataStax

Goals

  • The data you're querying needs to stay together.
    • Avoids seeks
    • Relational DBs are bad at this, rows get intersperesed.
  • Start with the questions you want to answer, then figure out how you would store that efficiently in Cassandra.
  • Don't normalize data too much.
  • Better to keep a record that something happened as opposed to a changing value.

Use Case: ClickStream data

What do you want to do?

  • Record actions within a session

How would you structure it?

  • Column family for session a user had.

    • Row key is user name or user id
    • Column Name is session id
    • Column Value is empty.

  • Column family for actual sessions

    • Row key is TimeUUID session id
    • Column name is timestamp
    • Column value is a JSON or XML blob
      • Why JSON or XML? You don't want to use SuperColumns. SuperColumns have a 10-15% performance penalty on read or writes. Also, this data is immutable. No need to change the data!
      • Maybe we aren't doing it so "wrong" then?

  • Append columns to UserSessions

    • userId: session_01, session_02, session_03
    • Note: If you use TimeUUID comparator, it will bring them out in the right order.
      • We should do this for our run_ids

Partitioning by Time

"Bucketing" data: Grouping it logically. i.e. "everything that happened in January"

  • Row key is composite of userid and time bucket
  • Column name is TimeUUID of Click
  • Column Value is serialized click data
  • Can aggregate these.

Use Case: Rolling Time Window of Data Points

Data point is a value or a set of descrete values that corresponds with a specific point in time or an interval.

I got lost here, I have the information but I don't have time to write it now.

Eventually Atomic

How do you cope with losing ACID? Example: User uploads a bunch of pictures but they're all related to another user. Answer: Tune consistency!