VictoriaMetrics

Anomaly Detection

Optimised Observability with AI

What is VictoriaMetrics Anomaly Detection?

VictoriaMetrics Anomaly Detection is a component of VictoriaMetrics Enterprise, which enhances your observability framework by identifying irregularities within metrics data.
Properly handles seasonality, trends and the other unique characteristics of metrics data, generating unified anomaly scores
Scores can be seamlessly integrated with alerting services, enabling the creation of comprehensive, action-oriented pipelines
Designed to bolster the stability of essential operational areas through automation
Minimizes the need for manual intervention, reducing operational costs and decreasing Mean Time to Resolution (MTTR)

What does it do?

VictoriaMetrics Anomaly Detection offers a lightweight service, characterised by flexible configuration and machine learning capabilities.
Designed to periodically scan new data points across selected metrics, it forecasts unified anomaly scores.
Scores are recorded back to VictoriaMetrics TSDB for utilization in subsequent applications, such as alerting services.
Simplified alerting rules can be established and observability insights received, enhancing your operational efficiency.

How does it work?

At its core, VictoriaMetrics Anomaly Detection autonomously re-trains either pre-defined machine learning models or custom models tailored to your business needs on your data.
ML models are employed to calculate anomaly scores for newly collected data points, as per a predefined schedule.
Alerts can be triggered based on simplified thresholds (i.e. anomaly_score > 1) that simplify and automate your observability setup.
Ongoing evaluations, presented either as specific point estimates or as ranges of confidence intervals, are designed to integrate seamlessly with downstream applications.
Tailored to Your Business Needs
Seamless Integration with Downstream Applications
Simplified & Automated Observability Setups

Getting Started

Getting started is simple: Follow the QuickStart guide, where you can find instructions on how to run VictoriaMetrics Anomaly Detection in Docker or Kubernetes.

Key Features & Benefits

Machine learning powered anomaly detection with the support of models, tailored to your business needs
Runs in Your Cloud and Kubernetes
Integrates well with existing solutions
Operational simplicity: Easy to set up & operate
Multi-tenancy support for VictoriaMetrics TSDB
Full support of MetricsQL for data ingestion
Easy maintenance & enhanced reliability (aka “automated thresholds”) *
* All produced anomaly scores are unified in a way that a value less than 1 means “likely normal”, while values greater than 1 means “likely anomalous” - automate and maintain your alerting, so your alerting rules may look as simple as “anomaly_score > 1”.

Customisable Configurations

In a single config define where to read data from, how often and what models to use on what data subsets, how and where to write produced metrics back.

Self-monitoring

VictoriaMetrics Anomaly Detection produces its own metrics for meta-observability.

Anomaly Detection Types Covered

VictoriaMetrics Anomaly Detection supports complex cases including

Contextual Anomalies

Such as seasonal variations, ensuring that fluctuations due to predictable patterns are accurately interpreted.

2021

2022

2023

2024

Time Series
Contextual Anomaly
Contextually Normal

Collective Anomalies

Collective anomalies in time series data are like a group of seemingly normal events that, when looked at together, stand out as unusual

Jan 2022

July 2022

Jan 2023

July 2023

Jan 2024

Time Series
Collective Anomaly

Multivariate Anomalies

Where the interaction between multiple metrics is analysed to identify complex patterns of irregular behaviour.

Time Series
Multivariate Anomaly

Typical Use Cases

Making observability team members’ job easier
VictoriaMetrics Anomaly Detection is a great fit  for SREs, DevOps and SE Engineers who need to provide improved observability for their entire company or for given teams.
Reducing human involvement in alerting rules/threshold maintenance by producing continuous, interpretable anomaly scores with unified alerting thresholds atop of.
IReducing MTTR for common observability use cases through presets and causal incident drilldown.
Handling complex metrics where the usage of constant thresholds or simpler models doesn’t make much sense.

Getting Started

Getting started is simple: Follow the QuickStart guide, where you can find instructions on how to run VictoriaMetrics Anomaly Detection in Docker or Kubernetes.

Some of Our Users & Customers

24h Support

Need Support?

Ground Control Is Here to Help With Support From The Core Team.

Ground Control is here to provide enhanced technical support, architectural and security guidance, help with adoption and integration and performance analytics.