GeeCON 2025: Victor Gamov - Stream Processing As You’ve Never Seen Before (Seriously): Apache Flink

## Stream Processing Fundamentals

Confluent's Victor Gamov introduces Apache Flink at GeeCON through a coffee-shop analogy: stateful systems should remember your order like a regular barista, not treat every visit as stateless. Flink is a distributed stream-processing framework for high-throughput, low-latency analytics over unbounded data.

## Architecture and Use Cases

Flink executes dataflow programs modeled as directed acyclic graphs, processing events with exactly-once semantics where configured. It integrates with Kafka, data lakes (Iceberg, Paimon), and batch pipelines, serving fraud detection, real-time dashboards, and event-driven microservices at scale.

## When Not to Use Flink

Gamov cautions against deploying distributed stream processors for tasks CLI tools handle faster—grep and awk beat a cluster for simple file parsing. Network overhead and operational complexity demand genuinely large-scale, continuous workloads before Flink's benefits justify the infrastructure.

## Stateful Versus Stateless Processing

The talk contrasts stateless per-event handling with stateful operators maintaining windows, aggregations, and keyed state across sessions. Understanding this distinction guides architects choosing between Flink, Kafka Streams, and batch systems for coffee-order memory—or enterprise event pipelines.