What Tools Does Netflix Use?

Netflix's main tools are Spinnaker for deployment, Titus for containers, Atlas and Mantis for monitoring, Kafka and Flink for data, and Metaflow for machine learning. Most of them were built by Netflix and released as open source, which is why the list is public. The rest of this page groups the tools by job and marks which ones Netflix still maintains.

Two cautions apply. Netflix does not publish its internal ticketing, chat or payment vendors, so this page does not name them. And several older Netflix projects are retired, so a tool list copied from a few years ago will be wrong.

Netflix tools by area

Deployment and infrastructure tools

Spinnaker is how code reaches production at Netflix. It is a continuous delivery tool, meaning it takes a built artifact and rolls it out across regions in stages. Its canary feature sends a small share of traffic to the new version, compares its metrics with the old one, and stops the rollout if they differ.

Titus runs the containers. A container packages a service with everything it needs so it runs the same on any host. Netflix built Titus on top of AWS and has described moving it onto Amazon EKS, the managed Kubernetes service, and adopting Kueue for batch job queueing. Underneath all of it are standard AWS services: EC2 for compute, S3 for storage, and several regions for failover.

Netflix monitoring tools

Searches for "Netflix monitoring tools" usually want three names.

Atlas stores metrics. It is a time-series database, meaning it stores numbers stamped with a time, built to take in billions of data points per minute and answer dashboard queries in near real time. Alerts are defined as Atlas queries.

Mantis handles operational events rather than counters. Services publish streams of events, and engineers write jobs that filter or aggregate them as they arrive, which lets someone watch playback failures for one device type without storing every event first.

Edgar is tracing. It follows a single request as it crosses dozens of services and shows where the time went. Netflix has written about Edgar but has not open sourced it.

Netflix has not stated in public whether it also uses commercial monitoring products such as Datadog.

Data and machine learning tools

Events from every app and service flow into Kafka. From there, Flink jobs compute results as events arrive, and Spark jobs run scheduled batches. Both read and write tables stored in the Iceberg format, which Netflix created to make very large tables safe to update and later donated to the Apache Software Foundation.

Pipelines are scheduled by Maestro, Netflix's workflow orchestrator. Data scientists build models in Metaflow, a Python framework that tracks each run and moves work between a laptop and the cloud with no code change. Notebooks run in Jupyter and in Polynote, a Netflix notebook that supports Scala next to Python.

Resilience and security tools

Chaos Monkey turns off production servers at random during working hours so that teams design for failure. ChAP, the Chaos Automation Platform, goes further by injecting failures into a small slice of traffic and comparing the result with a control group.

For security, Lemur issues and tracks TLS certificates, and Dispatch coordinates incident response through chat and tickets. Both are open source.

Tools Netflix has retired

Hystrix, the circuit breaker library that many companies copied, is in maintenance mode. Netflix replaced it internally with adaptive concurrency limits, a method that lowers the number of requests a service accepts when its latency rises. Conductor, the workflow engine, is no longer maintained by Netflix, and a community continues it separately. Ribbon, the client-side load balancer, is also in maintenance mode.