Skip to main content
Version: 1.0

Development Best Practices for Cost Optimization

This guide covers best practices for minimizing infrastructure costs during feature development, testing, and deployment.

Development vs. Live Workspaces​

Tecton provides separate workspace types to support cost-effective development:

Development Workspaces​

Development workspaces allow feature iteration without incurring materialization costs:

  • No automatic materialization jobs
  • Notebook-driven development
  • Feature testing without infrastructure costs
  • No real-time endpoints
# Create a development workspace
$ tecton workspace create my_new_feature_ws
$ tecton apply

Live Workspaces​

Use live workspaces only when you need:

  • Training data with incremental backfills (requires materialization)
  • REST API testing
  • Materialization testing
  • Stream source testing
  • Production feature serving

Testing Features in Development​

import tecton

# Use development workspace for testing
ws = tecton.get_workspace("my_new_feature_ws")
fv = ws.get_feature_view("user_transaction_metrics")

# Test feature logic without materialization
df = fv.get_features_in_range(
    start_time=datetime.now() - timedelta(days=7),
    end_time=datetime.now(),
    from_source=True,  # Compute directly from source
)

Cost-Effective Feature Development​

Start with Recent Data​

Begin feature development with a limited time range:

@batch_feature_view(
    ...,
    # Start with recent data
    feature_start_time=datetime.now() - timedelta(days=7),
    # Extend after validation
    # feature_start_time=datetime(2022, 1, 1),
)
def my_feature_view():
    return ...

Use Sampled Data Sources​

During development:

  • Create smaller test datasets
  • Use data sampling where appropriate
  • Consider using get_current_workspace() for environment-specific sources
@pandas_batch_config
def sampled_data():
    # Use sampled data in development
    if get_current_workspace().name == "dev":
        return small_sample_df
    return full_data_df

Minimize Feature Values​

Reduce materialization costs by:

  • Writing only necessary feature values
  • Setting appropriate TTLs
  • Using filtered data sources
  • Limiting time windows where possible
  • Reducing aggregation interval where possible

Configure Materialization Strategically​

  • Start with offline=False until online serving is needed
  • Set appropriate batch schedules
  • Use incremental materialization where possible
  • Consider feature freshness requirements
tip

We suggest first testing various configurations of feature freshness in offline training to understand the impact on model performance.

Integration Testing in Live Workspaces​

When testing in live workspaces:

  • Set recent feature_start_time
  • Use sampled data sources
  • Limit feature value cardinality
  • Consider setting offline=False
  • Reuse existing feature views
  • Reduce materialization frequency

Preventing Unnecessary Costs​

Protect Critical Objects​

Use prevent_destroy to protect production features:

@batch_feature_view(..., prevent_destroy=True)  # Prevent accidental deletion
def critical_feature():
    return ...

Minimize Rematerialization​

  • Build logical feature store model upfront
  • Use CI/CD for controlled deployments
  • Restrict workspace writes to authorized users
  • Consider using --suppress-materialization

Monitor Development Costs​

  • Review materialization job status
  • Track failed jobs
  • Monitor resource utilization
  • Set up cost alerts

Best Practices for Teams​

Development Workflow​

  1. Start in development workspace
  2. Use sampled data
  3. Validate feature logic
  4. Test in live workspace
  5. Deploy to production

Access Control​

  • Limit live workspace access
  • Use CI/CD for deployments
  • Implement review processes
  • Document cost implications

Best Practices Checklist​

  • Use development workspaces for iteration
  • Start with recent/sampled data
  • Minimize materialized values
  • Configure appropriate TTLs
  • Protect production objects
  • Monitor development costs
  • Implement CI/CD controls

Next Steps​

Was this page helpful?