Overview

Watchdog Explains is an investigation assistant that guides you to the root cause of anomalies on timeseries graphs.

Watchdog Explains makes investigations more efficient by automatically showing which individual tags could be responsible for a given spike. This allows you to focus your investigation on problematic areas of your infrastructure or software stack.

To disable Watchdog Explains, see Disabling anomaly detection.

Detecting anomalies with seasonality awareness

Watchdog Explains applies anomaly detection to graphs on your dashboard by analyzing both the shape and value of the underlying timeseries. It identifies deviations from historical patterns, flagging spikes, dips, or gradual drifts that don’t align with expected behavior.

To account for seasonality, the algorithm looks back up to three weeks in time. For example, if a spike appears on a Monday at 9:00 a.m., Watchdog compares that datapoint against previous Mondays at the same hour. If similar patterns appear consistently, the spike is treated as seasonal and not flagged as an anomaly. This helps reduce false positives and ensures that only unexpected deviations are surfaced.

Anomalies can be sharp spikes or drops, but may also be more subtle trends like step changes or slope shifts.

Investigating anomalies

You can start your investigation from any timeseries graph that uses metric data. When Watchdog Explains detects an anomaly, it highlights the affected region with a pink box. To begin investigating, click Investigate Anomaly.

This opens a full-screen investigation view. Watchdog analyzes the anomaly and surfaces any tag groups that significantly contributed to the shape or scale of the anomaly. Click on a tag to see how removing or isolating that dimension affects the graph. Use this to identify root causes like specific customers, services, or environments.

Isolating the cause with tag breakdown

Watchdog Explains analyzes the tag dimensions of the query to determine what may have caused the shift. It re-runs the same graph query filtered by each tag group (such as service, availability zone, or customer tier) to isolate the segments most likely responsible.

By comparing the shape of the filtered timeseries with the overall graph, Watchdog Explains can infer which tag group had the greatest impact.

When a tag is flagged, it means filtering by that tag significantly reduced or reshaped the anomaly. Watchdog compares the filtered timeseries to the original graph to pinpoint which tags contributed to the anomaly most. Sometimes, an anomaly may not have an obvious root cause. If no tag group meaningfully contributes to the spike or dip, Watchdog Explains may surface a general anomaly without tag-level attribution.

Example: Surge in traffic from a single tenant

A sudden spike in traffic is detected in the shopist-returns service. Watchdog surfaces a sharp increase in request volume, highlighted on the graph. Upon investigation, it attributes the anomaly to a single tag group: org_id:17728

Further analysis reveals:

• The spike originated entirely from that one customer.
• The traffic volume overwhelmed the rdv.shopist.io data center.
• This incident exposed a gap in tenant isolation and traffic governance mechanisms.

In this example, Watchdog’s tag breakdown makes it clear that this was not a general usage trend, but rather an infrastructure risk caused by one high-volume tenant.

Disabling anomaly detection

To disable anomaly detection on a dashboard, open Anomalies at the top of the dashboard and click Turn Off.

Further reading

추가 유용한 문서, 링크 및 기사:

Learn more about Watchdog InsightsDOCUMENTATION Anomaly detection, predictive correlations - Using AI-assisted metrics monitoringBLOG