Glossary

A

Aggregation

Reduction operation on a collection of events (sum, avg, count, etc.).

B

Backpressure

Flow control mechanism when a consumer is slower than the producer.

C

CEP (Complex Event Processing)

Complex event processing - pattern detection in real-time event streams.

Checkpoint

System state snapshot enabling recovery after crash.

Circuit Breaker

Resilience pattern that "opens the circuit" after multiple failures to avoid overload.

D

DAG (Directed Acyclic Graph)

Directed graph without cycles, used to represent the execution plan.

DLQ (Dead Letter Queue)

Queue for messages that could not be processed.

E

EPL (Event Processing Language)

Event processing language, generic term (used notably by Apama/Cumulocity).

Event

Immutable record representing a timestamped fact.

F

Flink

Apache Flink - distributed stream processing framework. See comparison.

H

Hypertree

Data structure optimized for pattern matching on events.

I

IR (Intermediate Representation)

Intermediate code representation between parsing and execution.

K

Kafka Streams

Stream processing library for Apache Kafka. See comparison.

L

Latency

Time elapsed between receiving an event and producing the result.

P

Partition

Logical division of a stream to enable parallel processing.

Pattern

Detection rule defining a sequence or combination of events to identify.

Pest

PEG (Parsing Expression Grammar) parser generator for Rust, used by VPL.

R

RocksDB

Embedded key-value database, used for state persistence.

S

Output Connector

Final destination of processed events, routed via .to() (Kafka, HTTP, file, etc.).

Source

Origin of events (Kafka, file, HTTP, etc.).

Stream

Potentially infinite sequence of typed events.

T

Throughput

Number of events processed per unit of time.

Tumbling Window

Fixed temporal window without overlap.

V

VPL (Varpulis Pipeline Language)

The pipeline and stream definition language of Varpulis.

Varpulis

Streaming analytics engine. Named after the Slavic wind spirit.

W

Watermark

Temporal marker indicating that all prior events have been received.

Window

Mechanism to bound computations on infinite streams.

• Tumbling: Non-overlapping windows
• Sliding: Overlapping windows
• Session: Inactivity-based windows