Skip to content

Exposures

Exposures document how your data is consumed by downstream systems. They don't deploy anything — they provide lineage documentation, SLA tracking, and impact analysis.

What is an Exposure?

An exposure represents:

  • An application that consumes your data
  • A dashboard that displays your metrics
  • An ML model that trains on your streams
  • Any downstream dependency
exposures:
  - name: fraud_detection_service
    type: application
    role: consumer
    consumes:
      - ref: transactions_clean

Why Document Exposures?

  1. Lineage — See the full picture of where data goes
  2. Impact Analysis — Know what breaks when you change a model
  3. SLA Tracking — Document and monitor service agreements
  4. Ownership — Know who to contact when things break

Basic Exposure

exposures/services.yml
exposures:
  - name: billing_service
    type: application
    description: Processes confirmed orders for invoicing
    owner: billing-team

    role: consumer
    consumer_group: billing-service-cg

    consumes:
      - ref: orders_confirmed

Exposure Types

Type Description
application Backend service or microservice
dashboard Visualization or reporting
ml_training ML model training pipeline
ml_inference Real-time ML inference
api External API endpoint

Complete Exposure Reference

exposures:
  - name: customer_dashboard
    type: dashboard
    description: |
      Real-time customer analytics dashboard showing
      engagement metrics, revenue trends, and churn risk.

    # Ownership
    owner: analytics-team
    tags: [analytics, customer, kpi]

    # URL to the exposure
    url: https://grafana.company.com/d/customer-metrics

    # What it consumes
    role: consumer
    consumer_group: grafana-customer-dashboard

    consumes:
      - ref: customer_metrics
      - ref: customer_segments
      - source: user_events

    # What it produces (if bi-directional)
    produces:
      - ref: dashboard_events

    # SLA expectations
    sla:
      latency_p99_ms: 5000      # 5 second max latency
      availability: 99.9         # 99.9% uptime
      throughput_per_second: 50  # Expected consumption rate

    # Contacts
    contacts:
      - name: Jane Smith
        email: jane@company.com
      - name: Analytics Team
        slack: "#analytics-team"

Roles

Consumer

The exposure reads from your data:

- name: reporting_service
  role: consumer
  consumer_group: reporting-cg
  consumes:
    - ref: daily_metrics

Producer

The exposure writes to your pipeline (usually a source):

- name: checkout_service
  role: producer
  produces:
    - source: orders_raw

Both

The exposure both reads and writes:

- name: enrichment_service
  role: both
  consumes:
    - ref: events_raw
  produces:
    - source: events_enriched

SLA Configuration

Document service level expectations:

sla:
  # Latency requirements
  latency_p50_ms: 100
  latency_p99_ms: 500

  # Availability
  availability: 99.95    # Percentage

  # Throughput
  throughput_per_second: 1000
  max_lag_seconds: 60

  # Data freshness
  freshness_seconds: 300

Lineage Visualization

With exposures defined, streamt lineage shows the complete picture:

$ streamt lineage

orders_raw (source) → checkout_service (producer)
    └── orders_clean (topic)
            ├── order_metrics (flink)
            │       └── operations_dashboard (dashboard)
            ├── orders_warehouse (sink)
            └── billing_service (application)

Impact Analysis

See what's affected by changes:

$ streamt lineage --model orders_clean --downstream

orders_clean (topic)
    ├── order_metrics (flink)
    │       └── operations_dashboard (dashboard)
    ├── orders_warehouse (sink)
    └── billing_service (application)

⚠️  3 downstream consumers will be affected

Examples

Application Exposure

- name: fraud_detection_service
  type: application
  description: |
    Real-time fraud detection using ML models.
    Scores transactions within 100ms.
  owner: fraud-team
  url: https://wiki.company.com/fraud-service

  role: consumer
  consumer_group: fraud-detection-cg

  consumes:
    - ref: transactions_enriched

  sla:
    latency_p99_ms: 100
    throughput_per_second: 5000
    availability: 99.99

  contacts:
    - name: Fraud Team
      slack: "#fraud-alerts"

Dashboard Exposure

- name: executive_dashboard
  type: dashboard
  description: C-suite real-time business metrics
  owner: bi-team
  url: https://tableau.company.com/exec-dashboard

  role: consumer
  consumer_group: tableau-exec

  consumes:
    - ref: revenue_hourly
    - ref: customer_metrics
    - ref: product_performance

  sla:
    latency_p99_ms: 10000
    availability: 99.5

ML Training Pipeline

- name: recommendation_training
  type: ml_training
  description: |
    Trains recommendation models daily.
    Consumes 7 days of interaction data.
  owner: ml-platform

  role: consumer
  consumer_group: ml-training-recommendations

  consumes:
    - ref: user_interactions
    - ref: product_views
    - ref: purchase_history

  sla:
    throughput_per_second: 10000  # High throughput for backfill

ML Inference

- name: recommendation_api
  type: ml_inference
  description: Real-time product recommendations
  owner: ml-platform
  url: https://api.company.com/recommendations

  role: consumer
  consumer_group: recommendation-inference

  consumes:
    - ref: user_features
    - ref: product_features

  sla:
    latency_p99_ms: 50
    throughput_per_second: 2000
    availability: 99.9

External API

- name: partner_data_feed
  type: api
  description: Partner data syndication API
  owner: partnerships-team

  role: consumer
  consumer_group: partner-feed-cg

  consumes:
    - ref: product_catalog
    - ref: inventory_updates

  sla:
    latency_p99_ms: 1000
    availability: 99.5
    throughput_per_second: 100

File Organization

Organize exposures by domain:

exposures/
├── services.yml        # Backend services
├── dashboards.yml      # Reporting dashboards
├── ml.yml              # ML systems
└── external.yml        # External consumers

Or by team:

exposures/
├── analytics-team.yml
├── fraud-team.yml
├── billing-team.yml
└── partners.yml

Best Practices

1. Document All Consumers

Every consumer should have an exposure:

# Don't leave mystery consumer groups
- name: unknown_consumer
  description: "TODO: Identify this consumer"
  consumer_group: legacy-consumer-123

2. Include Contact Information

contacts:
  - name: Primary Owner
    email: owner@company.com
    slack: "#team-channel"
  - name: Oncall
    pagerduty: team-oncall

3. Set Realistic SLAs

# Based on actual requirements
sla:
  latency_p99_ms: 500     # Not aspirational, achievable
  availability: 99.9       # Match your infrastructure

url: https://wiki.company.com/services/fraud-detection

5. Tag Appropriately

tags: [critical, tier-1, pci-compliant]

Validation

With governance rules, validate exposure completeness:

stream_project.yml
rules:
  exposures:
    require_owner: true
    require_sla: true
    require_contacts: true

Next Steps