Data Modeling Patterns for Time-Series, Events, and Wide-Row Workloads
Study practical Cassandra modeling patterns used in real production systems with large-scale event and activity data.
Inside this chapter
- Time-Series Modeling
- Bucketing to Avoid Oversized Partitions
- Wide Rows and Event Feeds
- Common Modeling Mistakes
Series navigation
Study the chapters in order for the clearest path from beginner Cassandra concepts to advanced distributed operations. Use the navigation at the bottom of each page to move through the full series.
Time-Series Modeling
Cassandra is widely used for time-series style workloads such as metrics, logs, device events, and user activity. These workloads usually need fast writes and reads by a known grouping such as device id, user id, or metric id over time.
Bucketing to Avoid Oversized Partitions
CREATE TABLE metrics_by_device_day (
device_id UUID,
event_day DATE,
event_time TIMESTAMP,
metric_name TEXT,
metric_value DOUBLE,
PRIMARY KEY ((device_id, event_day), event_time)
) WITH CLUSTERING ORDER BY (event_time DESC);Adding a day bucket into the partition key helps avoid one endlessly growing partition. This is a common and important Cassandra modeling pattern.
Wide Rows and Event Feeds
Cassandra can handle wide-row patterns well when partition growth is controlled thoughtfully. Activity feeds, message histories, and telemetry streams often use this style. The design challenge is balancing read convenience against partition size and operational safety.
Common Modeling Mistakes
- Using one partition key value for too much data
- Expecting relational joins later
- Creating tables before defining real queries
- Ignoring partition growth over time
- Depending on unsupported arbitrary filters