Kafka Events

26 Kafka event topics produced by the EventML Service, organized by category.

Event Format

All events use the CloudEvents 1.0 envelope and are partitioned by tenant_id. Events are published to Amazon MSK Serverless (Kafka).

CloudEvents envelope formatjson

{
  "specversion": "1.0",
  "type": "com.eventzr.ml.model-registered.v1",
  "source": "eventml-svc",
  "id": "550e8400-e29b-41d4-a716-446655440000",
  "time": "2026-02-28T12:00:00.000Z",
  "datacontenttype": "application/json",
  "partitionkey": "tenant-uuid",
  "data": {
    "modelId": "model-uuid",
    "name": "event-recommender-v1",
    "framework": "PYTORCH",
    "tenantId": "tenant-uuid"
  }
}

MSK broker configurationbash

# Amazon MSK Serverless (staging)
KAFKA_BROKERS=boot-gpm4u49z.c2.kafka-serverless.us-east-1.amazonaws.com:9098
KAFKA_SASL_MECHANISM=aws
KAFKA_SECURITY_PROTOCOL=SASL_SSL

Model Registry4 events

Topic	Description
`eventzr.ml.model-registered`	Emitted when a new ML model is registered in the model registry
`eventzr.ml.model-updated`	Emitted when model metadata (name, description, tags) is updated
`eventzr.ml.model-archived`	Emitted when a model is archived (soft-deleted)
`eventzr.ml.model-version-created`	Emitted when a new version is created for an existing model

Training4 events

Topic	Description
`eventzr.ml.training-started`	Emitted when a training job begins execution
`eventzr.ml.training-completed`	Emitted when a training job completes successfully with metrics
`eventzr.ml.training-failed`	Emitted when a training job fails with error details
`eventzr.ml.training-stopped`	Emitted when a training job is manually stopped by a user

Inference & Serving4 events

Topic	Description
`eventzr.ml.endpoint-created`	Emitted when a new inference endpoint is provisioned
`eventzr.ml.endpoint-scaled`	Emitted when an endpoint auto-scales (up or down)
`eventzr.ml.endpoint-deleted`	Emitted when an inference endpoint is decommissioned
`eventzr.ml.inference-completed`	Emitted after each inference request with latency and result metadata

Experiments3 events

Topic	Description
`eventzr.ml.experiment-started`	Emitted when an A/B experiment begins accepting traffic
`eventzr.ml.experiment-stopped`	Emitted when an experiment is stopped and traffic routing reverts
`eventzr.ml.experiment-result`	Emitted when experiment results are computed (winner, metrics)

Drift & Monitoring3 events

Topic	Description
`eventzr.ml.drift-detected`	Emitted when data drift exceeds the configured threshold
`eventzr.ml.drift-resolved`	Emitted when drift levels return below threshold after retraining
`eventzr.ml.monitor-created`	Emitted when a new drift monitor is configured

Feature Store2 events

Topic	Description
`eventzr.ml.feature-group-created`	Emitted when a new feature group is registered
`eventzr.ml.feature-ingested`	Emitted when feature data is ingested into the store

Governance3 events

Topic	Description
`eventzr.ml.model-approved`	Emitted when a model version passes governance review
`eventzr.ml.model-rejected`	Emitted when a model version fails governance review
`eventzr.ml.release-promoted`	Emitted when a model release is promoted to production stage

AutoML3 events

Topic	Description
`eventzr.ml.automl-started`	Emitted when an AutoML pipeline begins searching for best model
`eventzr.ml.automl-completed`	Emitted when AutoML finishes with best model selection
`eventzr.ml.automl-failed`	Emitted when AutoML pipeline fails with error details