Skip to content

Batch incremental scan: Diff mode for deduplicate PK tables #505

Description

@Pandas886

Describe the feature

Add Diff incremental planning/reading: compare table state at two snapshots and emit the net changes (insert / update-before / update-after / delete) when no pre-materialized changelog is available.

Motivation

Many tables run with changelog-producer=none. Diff is the practical way to still produce an audit/changelog stream for batch incremental jobs.

Proposed scope (phase 1)

  • Diff planning as per-(partition, bucket) before/after pairs
  • Fail loud (no silent wrong answers) for:
    • non-deduplicate merge engines
    • bucket rescale between snapshots
    • deletion vectors / nested/decimal types we can't safely compare yet
  • Audit-style row kinds for the diff result

Depends on the base incremental scan + AuditLog read path.

Out of scope

  • Full DV-aware diff
  • Non-deduplicate merge engines

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Fields

    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions