Cassandra is a partitioned row store, where rows are organized into tables with a required primary key.

The first component of a table’s primary key is the partition key; within a partition, rows are clustered by the remaining columns of the PK. Other columns may be indexed independent of the PK.

This allows pervasive denormalization to "pre-build" resultsets at update time, rather than doing expensive joins across the cluster.

https://github.com/jaegertracing/jaeger/blob/master/plugin/storage/cassandra/schema/v001.cql.tmpl

Stackoverflow

Cassandra UUID vs TimeUUID benefits and disadvantages?

Sample Tables:

CREATE TABLE sensor_readings ( sensorID uuid, time_bucket int, timestamp bigint, reading decimal, PRIMARY KEY ((sensorID, time_bucket), timestamp) ) WITH CLUSTERING ORDER BY (timestamp DESC);

SELECT * FROM sensor_readings WHERE sensorID = 53755080-4676-11e4-916c-0800200c9a66 AND time_bucket IN (1411840800, 1411844400) AND timestamp >= 1411841700 AND timestamp ⇐ 1411845300;

CREATE TABLE IF NOT EXISTS ${keyspace}.traces ( trace_id blob, span_id bigint, span_hash bigint, parent_id bigint, operation_name text, flags int, start_time bigint, duration bigint, tags list<frozen<keyvalue>>, logs list<frozen<log>>, refs list<frozen<span_ref>>, process frozen<process>, PRIMARY KEY (trace_id, span_id, span_hash) ) WITH compaction = { 'compaction_window_size': '1', 'compaction_window_unit': 'HOURS', 'class': 'org.apache.cassandra.db.compaction.TimeWindowCompactionStrategy' } AND dclocal_read_repair_chance = 0.0 AND default_time_to_live = ${trace_ttl} AND speculative_retry = 'NONE' AND gc_grace_seconds = 10800; — 3 hours of downtime acceptable on nodes

CREATE TABLE IF NOT EXISTS ${keyspace}.duration_index ( service_name text, // service name operation_name text, // operation name, or blank for queries without span name bucket timestamp, // time bucket, - the start_time of the given span rounded to an hour duration bigint, // span duration, in microseconds start_time bigint, trace_id blob, PRIMARY KEY ((service_name, operation_name, bucket), duration, start_time, trace_id) ) WITH CLUSTERING ORDER BY (duration DESC, start_time DESC) AND compaction = { 'compaction_window_size': '1', 'compaction_window_unit': 'HOURS', 'class': 'org.apache.cassandra.db.compaction.TimeWindowCompactionStrategy' } AND dclocal_read_repair_chance = 0.0 AND default_time_to_live = ${trace_ttl} AND speculative_retry = 'NONE' AND gc_grace_seconds = 10800; — 3 hours of downtime acceptable on nodes

Sequential writes can cause hot spots: If the application tends to write or update a sequential block of rows at a time, the writes will not be distributed across the cluster. They all go to one node. This is frequently a problem for applications dealing with timestamped data