Companies running Apache Flink in production include Netflix, which runs more than 30,000 Flink jobs across ad tech, recommendations, and a real-time distributed graph, evolved out of Keystone, its original stream processing platform; Uber, which runs more than 2,000 Flink SQL jobs across ML feature pipelines, ad billing, and streaming ingestion into its data lake; Lyft, which runs Flink across dynamic pricing, fraud detection, and a large-scale event persistence pipeline; Booking.com, whose Security Platform Services team runs Flink as the engine behind an internal security-as-a-service platform; Pinterest, which runs 130+ Flink jobs across image-similarity detection, experiment analytics, ad-budget enforcement, and CDC ingestion into Iceberg; Airbus, whose AirSense unit fuses multi-provider aircraft-position feeds into real-time global flight tracking; and Alibaba, which built Blink, its own internal Flink fork, to run Double 11 at a peak of 4 billion records a second before merging much of it back into Apache Flink itself. This page indexes production Apache Flink use cases, growing as we research and publish more.
Apache Flink use cases
Netflix
Netflix runs Apache Flink at large scale for stream processing as a service (Keystone), a real-time distributed graph connecting member activity across services, SQL-based stream transformations (the Data Mesh SQL Processor), and real-time replacements for batch metrics such as take fraction. Sourced from Netflix’s own engineering blog and conference talks by named Netflix engineers.
Uber
Uber runs Apache Flink across more than 2,000 SQL jobs processing over 4 trillion messages a day, spanning near-real-time ML feature pipelines (Michelangelo Palette), exactly-once ad event billing, geospatial anomaly detection, and IngestionNext, its Flink-to-Hudi streaming data lake. Sourced from Uber’s own engineering blog and conference talks by named Uber engineers.
Lyft
Lyft runs Apache Flink, sometimes through Apache Beam’s Flink runner, across at least five systems: real-time pricing and matching features, fraud detection, a shared ML feature-serving layer, a self-service streaming framework, and a large-scale event persistence pipeline moving 80 gigabytes of events a minute from Kinesis to S3. Sourced from Lyft’s own engineering blog and a conference talk by a named Lyft engineer.
Booking.com
Booking.com’s Security Platform Services team runs Apache Flink as the engine behind an internal security-as-a-service platform, built so that other internal security teams can run their own Flink pipelines on a shared, multi-tenant platform. Sourced from a conference keynote by two named Booking.com engineers, with scale figures corroborated only by a vendor case study.
Pinterest runs more than 130 Apache Flink jobs across eight multitenant YARN clusters: near-real-time image-similarity detection, experiment analytics, ad-budget enforcement, its Kappa-architecture visual signal platform, and a next-generation change-data-capture pipeline into Apache Iceberg. Sourced from 18 pieces, almost all Pinterest’s own engineering blog, plus conference talks by named Pinterest engineers.
Airbus
Airbus’s AirSense unit runs Apache Flink to fuse aircraft-position reports from multiple independent ADS-B data providers into a single real-time global flight-tracking service, processing more than 2 billion events a day. Sourced from a conference talk by a named Airbus engineer, Airbus’s own newsroom announcement, and a case study from its platform vendor.
Alibaba
Alibaba built Blink, its own internal fork of Apache Flink, to run real-time search, recommendation, and its annual Double 11 shopping festival at a peak of 4 billion records a second, then merged Blink’s job-isolation, checkpointing, and SQL innovations back into Apache Flink 1.9 and 1.10. Sourced from Alibaba’s own engineering blog, the official Apache Flink blog, and conference talks by named Alibaba engineers.
FAQ
What companies use Apache Flink in production?
Netflix runs Apache Flink at large scale, with more than 30,000 jobs spanning a self-serve stream processing platform, a real-time distributed graph, and SQL-based stream transformations, as detailed in our own use-case writeup on Netflix. Uber runs more than 2,000 Flink SQL jobs processing over 4 trillion messages a day, spanning ML feature pipelines, ad billing, and data lake ingestion, as detailed in our own use-case writeup on Uber. Lyft runs Flink across dynamic pricing, fraud detection, and one of its largest event-persistence pipelines, as detailed in our own use-case writeup on Lyft. Booking.com’s Security Platform Services team runs Flink as the engine behind an internal security-as-a-service platform, as detailed in our own use-case writeup on Booking.com. Pinterest runs more than 130 Flink jobs across image-similarity detection, experiment analytics, ad-budget enforcement, and CDC ingestion into Iceberg, as detailed in our own use-case writeup on Pinterest. Airbus’s AirSense unit fuses multi-provider aircraft-position feeds into real-time global flight tracking, as detailed in our own use-case writeup on Airbus. Alibaba built Blink, its own internal Apache Flink fork, to run real-time search, recommendation, and its Double 11 shopping festival at a peak of 4 billion records a second, then merged much of it back into Apache Flink itself, as detailed in our own use-case writeup on Alibaba. We are adding further named production deployments to this page as we research and publish them.
What is Apache Flink used for?
Apache Flink is used for stateful, low-latency stream processing: enrichment, joins across streams, event-time windowing, and continuous SQL over unbounded data, typically fed by a system like Apache Kafka.
Related reading
- Apache Flink: the complete guide
- How Netflix uses Apache Flink in production
- How Uber uses Apache Flink in production
- How Lyft uses Apache Flink in production
- How Booking.com uses Apache Flink in production
- How Pinterest uses Apache Flink in production
- How Airbus uses Apache Flink in production
- How Alibaba uses Apache Flink in production