- Blog /
- VictoriaMetrics Machine Learning takes monitoring to the next level
Today we’re happy to announce our new VictoriaMetrics Anomaly Detection solution, which harnesses machine learning to make database alerts more relevant, accurate and actionable for enterprise customers.
VictoriaMetrics Anomaly Detection lightens the load on overworked data engineers, focusing their scarce resources on the alerts that matter most to their organization.
By unifying anomalies under a simple scoring system, VictoriaMetrics continues its mission to make monitoring even the most complex data sets simpler, more reliable, and more efficient.
As monitoring has spiralled in complexity, with databases becoming more interconnected and co-dependent, engineering teams can quickly become overwhelmed with alerts. Simplistic alerting is unable to distinguish minor performance concerns from potentially mission-critical outages.
To simplify monitoring for the very large datasets enterprises rely on today, we’ve developed VictoriaMetrics Anomaly Detection from scratch to identify data trends and alert only for signals that matter. Now, for the first time using neural networks, it is possible to set alerts which ‘understand’ data and so can draw conclusions from the data’s context. This turns the challenges of time-series data into a strength.
"Machine Learning is famously energy intensive, as a company that prides itself on efficiency we had to balance energy usage with the value it created for businesses. VictoriaMetrics Anomaly Detection is designed to be as efficient as the rest of our product range once calibrated to make sure businesses see a clear return on their investment"
- Roman Khavronenko, co-founder VictoriaMetrics
Most alert systems use threshold alerting, where an alert is sent only if a value exceeds, or falls below, a predetermined range to signal systems operating outside normal tolerances. With the scalability and seasonality of modern real-time and distributed systems, alert thresholds need to be more complex if they are to offer control of an ever-evolving and scaling database.
Instead, VictoriaMetrics Anomaly Detection analyzes historical data and attaches an anomaly score to each data point indicating how far a signal deviates from the expected value or pattern. For engineers, it couldn’t be simpler, whenever the anomaly value exceeds 1, an alert can be generated, taking the cognitive load off of engineers so they can focus on what matters.
VictoriaMetrics monitoring tools are already used by some of the largest databases on the planet counting Grammarly, Wix and CERN among their users. Anomaly detection can ingest the historic data businesses are already generating to calibrate itself, with minimal oversight.
As part of its commitment to efficiency, the VictoriaMetrics team designs all new technologies with the aim of reducing database workloads. Previously, database monitoring at this level required engineering teams continuously on-call; now, monitoring teams are augmented with an AI-like tool continuously observing the system.
VictoriaMetrics Anomaly Detection can account for:
Machine learning understands context and intelligently adapts to changing data dynamics.
Anomaly Detection trained with historic data, allowing it to identify anomalies that would otherwise require an engineer familiar with the data set.
In isolation, concerning signals can go under the radar, continuously analyzing entire datasets to detect patterns all but senior engineers would miss.
Anomaly Detection can detect ‘novelties’, or significant changes in the underlying system, intelligently adjusting to a ‘New Normal’.
Getting started is simple: Follow the QuickStart guide, where you can find instructions on how to run VictoriaMetrics Anomaly Detection in Docker or Kubernetes.
The OpenTelemetry Astronomy Shop demo has long served as a reference environment for exploring observability in distributed systems, but until now it shipped with only a Prometheus datasource. VictoriaMetrics forked the demo and extended it with VictoriaMetrics, VictoriaLogs, and VictoriaTraces, providing insights into VictoriaMetrics’ observability stack where metrics, logs, and traces flow into a unified backend.
Proper alerting is an art. It is all about foreseeing bad scenarios before they happen, so you can prepare for them. In this article, we go through practical recommendations of approaching the alerting to reduce alerting fatigue and avoid false positives.
Tech Talk: In this post, we explore vmanomaly through the eyes of its creators. Learn how this AI-powered alerting system helps cut through noise, avoid static rule spaghetti, and deliver actionable insights directly from your monitoring data.
VictoriaLogs uses three core concepts: message, time, and stream fields to structure log data. Too few streams create fat streams that are slow to query, while too many unique stream combinations create high cardinality problems…