PATTERN Cited by 1 source

Shadow-Mode Parallel Validation

Definition

Deploy a new system in shadow mode — running in parallel with the production system on real workloads without customer impact — then prove data equivalence through rigorous row-by-row comparison of outputs.

Canonical example

Netflix's Casspactor → Move Data migration (Source: sources/2026-06-19-netflix-the-evolution-of-cassandra-data-movement-at-netflix):

• Let C = rows in legacy Casspactor output (Iceberg table).
• Let M = rows in new Move Data output (Iceberg table).
• Trust metric: prove C = M by checking both (C−M) and (M−C) are empty.
• Any non-zero difference triggered immediate high-priority investigation.
• Target: 100% similarity.

Key insight

Shadow mode became a forensic tool exposing "unknown unknowns" — subtle behavioral differences in TTL handling, consistency levels, backup selection, and business logic that were not bugs in the new system but differences between systems.

Process

1. Run new connector in shadow alongside production jobs.
2. Continuously compare output sets at row level.
3. Maintain an investigation log categorising each disparity.
4. Resolve categories to build measured confidence.
5. Use confidence level to drive migration cohort expansion.

Relationship to other patterns

• Extends patterns/shadow-migration with a formal data-equality proof framework.
• Feeds confidence into patterns/decider-pattern-workflow-routing cohort decisions.

Seen in

• sources/2026-06-19-netflix-the-evolution-of-cassandra-data-movement-at-netflix — Pillar 1 (Validation) of Netflix's three-pillar migration strategy.