Catalog
method#Delivery#Reliability#Integration

Parallel Run

A risk-reducing rollout method where the new system runs concurrently with the legacy system to validate functionality and data reconciliation.

Parallel run is a migration and rollout technique where the new system operates alongside the legacy system for a defined period to validate functionality and data reconciliation.
Established
Medium

Classification

  • Medium
  • Technical
  • Organizational
  • Intermediate

Technical context

Monitoring and observability tools (e.g. Prometheus)CI/CD pipelines for reproducibilityData replication or ETL tools

Principles & goals

Define clear validation and rollback criteria.Ensure comprehensive monitoring and comparison metrics.Establish communication and responsibilities in advance.
Run
Domain, Team

Use cases & scenarios

Compromises

  • Inconsistent data due to incomplete replication.
  • Missing test scenarios may overlook critical faults.
  • Operational mistakes during cutover can cause outages.
  • Automate comparisons and create clear dashboards.
  • Start with representative pilot areas before broad rollout.
  • Document all discrepancies and remediation steps thoroughly.

I/O & resources

  • Legacy and new system with defined APIs
  • Test data and live sample data
  • Monitoring, logging and reconciliation tools
  • Reconciliation logs and migration reports
  • Cutover or fallback decision
  • Improved remediation and rollback procedures

Description

Parallel run is a migration and rollout technique where the new system operates alongside the legacy system for a defined period to validate functionality and data reconciliation. It reduces outage risk and provides controlled fallback options. Coordination effort and operational overhead increase with system complexity.

  • Reduces risk of production outages during cutover.
  • Enables real tests under load with real data.
  • Provides clear fallback options in case of issues.

  • Increased operational overhead due to dual operation.
  • Not always practical due to compute or licensing costs.
  • Complexity in data consistency and synchronization.

  • Discrepancy rate

    Percentage of transactions that differ between systems.

  • Time to stability

    Time until all critical checks run without discrepancies.

  • Operational overhead

    Additional person-days and costs for running in parallel.

ERP rollout at an SMB

A new ERP was initially run in parallel to correct posting and reporting discrepancies.

Core banking system migration

Bank ran a parallel operation for weeks to ensure transaction consistency.

E-commerce platform changeover

Product data and order flows were validated in parallel before decommissioning the old system.

1

Assess: Identify affected components, data flows and stakeholders.

2

Plan: Define duration, rollback criteria, test cases and responsibilities.

3

Prepare: Set up replication, monitoring and reporting.

4

Execute: Start parallel run, perform comparison tests and validations.

5

Decide: Based on metrics perform cutover or extend/rollback.

⚠️ Technical debt & bottlenecks

  • Ad-hoc replication scripts that remain technical debt.
  • Unclear interface contracts after the parallel period.
  • Monitoring solutions configured temporarily instead of embedded sustainably.
Data synchronizationMonitoring capacityRelease coordination
  • Permanent parallel operation used as a long-term architecture instead of a migration step.
  • Omitting monitoring and relying solely on sampling.
  • Parallel operation without reconciling sensitive data fields.
  • Underestimated synchronization effort for transactional systems.
  • Ignoring latency differences between systems.
  • Lack of automation leads to human errors.
Operations and release engineeringData migration and reconciliationMonitoring, alerting and incident handling
Data consistency and integrityOperational availabilityInterface and integration compatibility
  • Limited compute or licensing resources for parallel systems.
  • Parallel-run windows must not disrupt business processes.
  • Regulatory rules may restrict data duplication.