Atlassian's Multi-Signal Approach Speeds Up Root Cause Investigation in Cloud Incidents

When incidents strike large distributed systems, engineers typically face hours of manual investigation—jumping between dashboards, trawling logs, and mentally stitching together pieces of telemetry to understand what went wrong. Atlassian has published a fresh framework for automating much of this detective work, treating root cause analysis (RCA) as a problem of correlating signals across time, signal type, and service topology.

The Core Challenge#

Modern cloud systems generate vast amounts of observability data—metrics, traces, and logs—but connecting these disparate signals into a coherent narrative remains labor-intensive. Many organizations compound the problem by running multiple observability tools in parallel, forcing engineers to bridge data across platforms by hand. The real constraint isn't data collection; it's the speed and accuracy with which teams can reason over what the data means.

How Atlassian's System Works#

Rather than analyzing every service in a production environment, the approach starts by narrowing focus. The system uses OpenTelemetry-derived service maps to identify which services are involved in a customer's affected journey, then builds a dependency graph from actual traffic patterns—not static architectural diagrams.

Anomaly detection runs independently on each telemetry type:

• Metrics are scanned for unusual changes in rates, error counts, and latencies
• Traces are examined for exceptions, slowdowns, and structural deviations
• Logs are grouped to spot new or rare error patterns

Once anomalies surface across these channels, the system normalizes them into a common format and groups nearby events into potential failure sequences. Time proximity alone isn't enough to prove causation, however. The system layers in service dependency information and temporal ordering, tracing upstream from affected services to pinpoint likely origins. The result is a ranked list of hypotheses—each showing a suspected cause, the services it affected, and the evidence backing that claim.

Context: A Crowded Landscape#

Atlassian isn't alone in tackling automated RCA. Grafana Cloud is building knowledge-graph approaches to surface correlations, and correlates multi-source events with telemetry to identify problems. Atlassian's angle emphasizes modularity—individual anomaly detectors can improve independently while the same reasoning engine evaluates their collective findings.