Netflix Atlas

What it does

Alerting

Atlas's documented common API includes Holt-Winters for alerting.checked 11 Oct 2026

Backend queries

Atlas's query layer can combine views across backends and distribute query and aggregation operations across a hierarchy.checked 11 Oct 2026

CloudWatch

The overview says Netflix wanted common query capability for data from third-party services such as CloudWatch.checked 11 Oct 2026

Data retention

The documented Netflix policy keeps all data for under six hours, rolls up older data, and uses an explicit whitelist for older-than-16-day data.checked 11 Oct 2026

Graph access

Atlas graph images can be accessed using a GET request and embedded or linked from other resources.checked 11 Oct 2026

Historical storage

The documentation says data is flushed to long-term storage in S3, with Hadoop used for corrective merging, reports, and rollups.checked 11 Oct 2026

Instrumentation

The documentation points developers to Spectator for instrumenting their own code.checked 11 Oct 2026

Integrations

The overview says Netflix wanted common query capability for third-party services such as CloudWatch.checked 11 Oct 2026

Intended use

The documentation describes Atlas as a system for operational insight into what is currently happening within a system.checked 11 Oct 2026

License

The repository states that Atlas is licensed under the Apache License, Version 2.0.checked 11 Oct 2026

Memory tradeoff

The documentation says Atlas stores most data in memory, which is expensive as data volume grows.checked 11 Oct 2026

Purpose

Atlas manages dimensional time-series data for near-real-time operational insight.checked 11 Oct 2026

Query features

The stack language supports querying, aggregation, time consolidation, math, boolean operations, and graph settings.checked 11 Oct 2026

Query language

Atlas uses a URL-friendly stack language for encoding graph expressions and supports query, aggregation, consolidation, math, Boolean, and graph-setting operations.checked 11 Oct 2026

Query layer

Atlas has a hierarchically composable query layer designed to support different backends and merged views across backends.checked 11 Oct 2026

Resilience

Netflix's internal deployment uses multiple mirrors for its six-hour clusters so data is replicated and an instance can be lost.checked 11 Oct 2026

Retention limits

Netflix's described internal policy keeps all data under six hours, rolls up older data, and uses an explicit whitelist for data older than 16 days.checked 11 Oct 2026

Scale

Atlas was designed to scale with the hardware used to analyze and store its data.checked 11 Oct 2026

Storage

Atlas uses in-memory data storage to gather and report very large numbers of metrics quickly.checked 11 Oct 2026

Support

The repository links to a mailing list and GitHub issues for community contact and issue reporting.checked 11 Oct 2026