A unified representation for observable data
UModel provides a standard data modeling language to uniformly describe observable data:
Core Set types:
EntitySet: Defines a collection of entities. It includes information such as primary keys, properties, and status.
TelemetryDataSet: A general representation for observable data. It includes a base `time` field. Other sets, such as MetricSet, LogSet, TraceSet, and EventSet, inherit from this type.
LogSet: A general definition for log data with few constraints.
MetricSet: Defines a collection of metrics. It supports multiple metrics and tag dimensions.
TraceSet: A definition for tracing analysis. It includes fields such as TraceID and SpanID.
EventSet: A definition for event data. It supports alerts and system events.
ProfileSet: A definition for performance profiling data. It supports performance data such as CPU and memory usage.
Layered progression principle:
Field: The most basic field definition.
ActivityData: Observable data that includes a time property.
Entity&TelemetryDataSet: Abstracted entities and observable data.
Link: Connects entities to other entities and to data, forming a complete system of associations.
Core idea of entity association and data fusion
UModel emphasizes interconnection. It builds data graphs and entity graphs through associations:
Data interconnection:
Logs and metrics are associated through entities. This enables association analysis of log anomalies and metric anomalies.
Traces and logs are associated through a TraceID. This fuses distributed tracing with detailed logs.
Events and entities are associated. This provides a unified view of alert events and system status.
Entity interconnection:
The call relationship between a service and a database supports analysis of how database performance affects the service.
The containment relationship between a container and a host supports fault propagation analysis at the resource layer.
The dependency between an application and its infrastructure supports full-stack root cause analysis.
Neuron-like connections: Fault propagation and root cause analysis
UModel uses a connection method similar to a neural network, leveraging an Entity Relationship Diagram to perform intelligent fault propagation and root cause analysis:
Fault propagation model:
Simulates the propagation path of a fault in the system based on the relationship graph defined by EntitySetLink.
Supports multi-level jump analysis to trace from an abnormal entity to the root cause entity.
Provides fault propagation probability analysis by combining historical data with machine learning algorithms.
Root cause analysis capabilities:
Alert events automatically associate with relevant entities to generate a minimum connected subgraph.
Identifies the chronological order of faults based on time series analysis.
Quickly locates faults caused by changes by combining change events, such as CI/CD and configuration changes.
The path from data to insight
UModel provides a complete path to transform raw data into business insights:
Data normalization: Standardizes data formats using Field definitions and Sets.
Entity modeling: Organizes scattered data into entities using EntitySet.
Relational association: Establishes associations between data and between entities using Link.
Graph-based analysis: Performs complex association analysis and pattern recognition based on the graph structure.
Intelligent insights: Achieves automated anomaly detection and root cause analysis by combining AI algorithms.