Principle:SeldonIO Seldon core Usage Metrics Data Model
| Knowledge Sources | |
|---|---|
| Domains | Observability, Data_Model, Usage_Telemetry |
| Last Updated | 2026-02-13 14:00 GMT |
Overview
A hierarchical data schema that defines the structure and categories of anonymous usage metrics collected from a Seldon Core 2 deployment.
Description
The Usage Metrics Data Model defines a four-category schema for ML platform telemetry: Collector Metrics (version and identity of the collection agent), Cluster Metrics (cluster identity, platform versions), Resource Metrics (counts of deployed models, pipelines, experiments, servers), and Feature Metrics (server configuration flags and resource allocations). The schema uses structural embedding to compose these categories into a flat, JSON-serializable object suitable for event-based analytics systems.
This separation of concerns allows metrics to be collected and published at different granularity levels without restructuring the data model.
Usage
Use this principle when defining the data contract between a metrics collector and a metrics publisher in an ML platform telemetry system. The schema should be flat enough for analytics ingestion while maintaining logical groupings for code clarity.
Theoretical Basis
The data model follows a hierarchical composition pattern:
Pseudo-code Logic:
# Abstract data structure
UsageMetrics = {
CollectorMetrics: {version, git_commit},
ClusterMetrics: {cluster_id, seldon_version, k8s_version},
ResourceMetrics: {model_count, pipeline_count, experiment_count, server_count},
FeatureMetrics: {multimodel_enabled, overcommit_enabled, gpu_enabled, memory_gb}
}
# Serialized as flat JSON via struct tag reflection