Documentation
Overview
TSP (Time Series Patterns) - analytical backend for searching patterns in high-volume time-series data.
Key Properties:
- Agile - flexible DSL for expressing unique patterns
- Fast - optimized stream-based processing engine on top of Apache Flink
Example patterns:
- Invalid use of engine - “5 seconds average engine speed greater than 5000 for 10 min then maximum oil pump for 20 sec is bigger than zero”:
avg(engineSpeed, 5 sec) > 5000 for 10 min andThen max(oilPump, 20 sec) > 0 -
Sessionization by user activity - “Time between any user actions is less than 10 min”:
anyOf(click, move, ...) for 10 minutes > 0 times -
Behavioral analytics - “User spent too little time on advertisement page”:
activePage = "AdPage" for lessThen 0 sec - More about writing patterns
Deployment options:
It can be used as the service and library:
- As a service it:
- Takes as input: source and sink type in URL (JDBC, InfluxDB or Kafka topic for now), then configuration and patterns to search in the body (endpoints details)
- Write incidents (occurrences of the pattern) in the sink as they’re found, with neighbours united together
- Provide monitoring end-points