Incremental data loading is a cornerstone of modern data engineering pipelines. Instead of re-ingesting entire datasets on each execution, incremental strategies focus on retrieving only records that are new or modified since the last load. This approach reduces latency, improves efficiency, and lowers infrastructure costs.

When designing incremental loads, a common dilemma arises: should the pipeline rely on system-provided cursors such as resource_version, or on timestamp fields like created_at and updated_at? Both approaches are valid, but the choice has significant implications for consistency and reliability.

1. Using resource_version

Some APIs and platforms, such as Chargebee and Kubernetes, expose a resource_version (or an equivalent cursor). This value is a monotonically increasing token that represents the state of a resource at a particular moment in time.

Advantages

• Consistency and ordering: Ensures no gaps or overlaps in data retrieval.
• Resilient to collisions: Captures all updates even if they occur within the same timestamp.
• Purpose-built for synchronization: Specifically designed for incremental replication.

When to Use

• The API explicitly provides resource_version or a cursor-based mechanism.
• Incremental consistency is critical (e.g., billing, compliance, auditing).
• Resources are frequently updated, requiring a robust mechanism to track changes.

2. Using created_at and updated_at

Most APIs and databases provide timestamp fields. These are often used for incremental ingestion when a cursor is not available.

Advantages

• Universality: Almost every system provides timestamp metadata.
• Ease of implementation: Simple to understand and apply.

Challenges

• Resolution issues: Updates within the same second may lead to missed records.
• Retroactive updates: Historical records modified after their initial creation can be overlooked.
• Time synchronization risks: Clock drift in distributed systems can cause inconsistencies.

When to Use

• The system does not support cursors or resource_version.
• Update frequency is low, and risks of timestamp collisions are minimal.
• Defensive measures are implemented, such as:
  • Querying with >= instead of >.
  • Storing the maximum updated_at value from the last run.
  • Deduplicating records by primary key.

4. Best Practice Recommendation

• Prefer resource_version (or equivalent cursors) whenever available. These mechanisms are designed for incremental synchronization and provide strong consistency guarantees.
• Fallback to updated_at with safeguards if cursors are unavailable. Ensure robust handling of potential timestamp collisions and retroactive updates.

This hybrid approach ensures pipelines are both efficient and resilient, regardless of the API or system constraints.