Netdata

Netdata is an open source, real-time infrastructure monitoring platform written mostly in C and Go, positioned as a lower-cost alternative to Datadog and New Relic. It collects per-second metrics from servers, containers, and applications without requiring the user to define what to monitor first, since it auto-discovers metrics sources on start. The project traces back to 2013, when its creator, frustrated by monitoring tools that provided too few metrics at too low a resolution, built a replacement from scratch. The Agent (GPL v3+) runs the ML pipeline and dashboards locally, while an optional Netdata Cloud layer adds centralized alerting and multi-node views.

Key features

• Per-second data collection: Netdata gathers metrics every second by default rather than the 10-to-60-second intervals typical of most monitoring tools, so short-lived spikes stay visible on the dashboard instead of getting averaged away.
• Zero-configuration auto-discovery: The agent detects running services, containers, and hardware on the host at startup and begins collecting relevant metrics without manual instrumentation or a predefined list of what to watch.
• Unsupervised anomaly detection: Netdata trains multiple machine learning models per metric directly at the edge, on the monitored node itself, to flag anomalies without sending raw data to a central service for analysis.
• Efficient long-term storage: A tiered storage engine compresses samples down to roughly half a byte per data point, letting nodes retain long history locally without a separate archival pipeline.
• Distributed, edge-based architecture: Processing happens on each node rather than centralizing raw data, which keeps overhead low, keeps data local by default, and simplifies compliance for teams that cannot ship metrics off-site.
• Horizontal scaling via Parent-Child: Nodes can stream metrics to Parent nodes for centralization, supporting multi-million-sample-per-second ingestion across large fleets without a bespoke aggregation layer.
• Interactive, query-language-free dashboards: Charts can be sliced and filtered directly in the UI, so operators can drill from a fleet-wide view down to a single container without writing a query.

How it works

Netdata's three components split responsibilities: the Agent handles collection, storage, ML, and exports, and can be deployed on servers, cloud instances, Kubernetes, or IoT devices; Netdata UI renders dashboards and charts served from a CDN; Netdata Cloud, the only closed-source piece, adds user management, RBAC, and cross-node alert centralization on top of the open source Agent, and includes a free community tier. Because the ML models train per metric on the node itself, a newly added server starts building its own anomaly baseline immediately rather than waiting on a shared model to be retrained.

Self-hosting and deployment

The Agent installs with a single shell command or via Docker, and supports Linux, macOS, FreeBSD, and Windows. Because each Agent processes and stores its own metrics, there is no requirement to run a central time-series database, though Parent-Child streaming lets teams centralize retention if needed for fleets that want one place to query history. A University of Amsterdam study cited in the project's README found Netdata to be the most energy-efficient tool for monitoring Docker-based systems, ahead of alternatives measured on CPU usage, RAM usage, and execution time, which matters for teams running the agent on thousands of nodes.