Principle:SeldonIO Seldon core Monitoring Pipeline Validation
| Property | Value |
|---|---|
| Principle Name | Monitoring Pipeline Validation |
| Overview | Deploying and validating that a monitoring pipeline and all its component models are operational |
| Domains | MLOps, Kubernetes |
| Related Implementation | SeldonIO_Seldon_core_Seldon_Pipeline_Load_And_Status |
| Knowledge Sources | Repo (https://github.com/SeldonIO/seldon-core), Doc (https://docs.seldon.io/projects/seldon-core/en/v2/) |
| Last Updated | 2026-02-13 00:00 GMT |
Description
After defining the monitoring pipeline, it must be deployed and validated. Validation confirms that all four component steps are ready:
- Classifier (income) - Model loaded and serving
- Preprocessor (income-preprocess) - Model loaded and transforming
- Drift Detector (income-drift) - Model loaded and ready for batch aggregation
- Outlier Detector (income-outlier) - Model loaded and ready for anomaly scoring
The pipeline status shows each step's readiness state. The pipeline transitions through states: PipelineCreating to PipelineReady (or PipelineFailed if any component is unavailable).
Theoretical Basis
Monitoring pipeline validation is a composite check that verifies the entire signal chain:
- Input to classifier - Direct path for predictions
- Input to preprocessor to outlier detector - Chained path for anomaly detection
- Input to drift detector (batched) - Aggregated path for distribution testing
Partial Failure Modes
A partial failure (e.g., drift detector not ready) may still allow the pipeline to function for predictions but disables monitoring. The validation check provides visibility into which specific components are operational:
| Scenario | Predictions | Outlier Detection | Drift Detection |
|---|---|---|---|
| All ready | Available | Available | Available |
| Drift detector down | Available | Available | Unavailable |
| Outlier detector down | Available | Unavailable | Available |
| Preprocessor down | Available | Unavailable (depends on preprocessor) | Available |
| Classifier down | Unavailable | Available | Available |
Kafka Data Flow Validation
The Seldon scheduler creates Kafka topics for inter-step communication during pipeline deployment. Validation also implicitly confirms that:
- Kafka topics are created for each step
- Message routing between steps is configured
- Batch aggregation buffers are initialized for the drift detector
Usage
Use this principle after deploying all monitoring components and the pipeline definition. The validation sequence is:
- Submit the pipeline CRD using
seldon pipeline load - Wait for the pipeline to reach PipelineReady state using
seldon pipeline status -w PipelineReady - Inspect the status response to verify all four steps are listed and active
- Check the pipeline version number to confirm the expected version is deployed
Related Pages
- SeldonIO_Seldon_core_Seldon_Pipeline_Load_And_Status (implements this principle) - Concrete CLI tools for deploying and validating monitoring pipelines
- SeldonIO_Seldon_core_Monitoring_Pipeline_Definition (prerequisite) - Defining the monitoring pipeline CRD
- SeldonIO_Seldon_core_Seldon_Pipeline_CRD_Monitoring (prerequisite) - Concrete pipeline CRD pattern
- SeldonIO_Seldon_core_Production_Traffic_Monitoring (next step) - Sending production traffic after validation
- SeldonIO_Seldon_core_Pipeline_Version_Progression (related) - Validating across pipeline version updates
Implementation:SeldonIO_Seldon_core_Seldon_Pipeline_Load_And_Status