Skip to main content
GIFRÖST / Compare

Database consistency without downtime.

Compare verifies whether source and target databases stay aligned across replication, migration, backup, reporting, and active-active topologies, while both systems remain online.

Zero downtimeRun periodic consistency checks on live systems without taking source or target offline.
Heterogeneous databasesCompare Oracle, PostgreSQL, Informix, Db2, and mixed-engine environments in one workflow.
Built for replicationConfirm out-of-sync rows even when replication is still in flight and data keeps changing.
Best fitMigration, replication, disaster recovery, and reporting copies.
Works onLive source and target databases with continuous change.
Main outcomeFast confirmation of persistent out-of-sync rows.
Database mixOracle, PostgreSQL, Informix, Db2, and heterogeneous pairs.

What is Compare?

Compare is built for teams that need a dependable answer to one operational question: is the target still faithful to the source, even while replication and business traffic continue?

Compare is a module of the GIFRÖST platform that ensures data consistency between a source database and target databases — backups, reporting databases, active-active configurations, and other systems holding a redundant copy of the data. Throughout this documentation, we refer to these systems collectively as target databases.

The need for high data availability and near 24/7/365 access has driven organizations to maintain distributed, redundant copies of their data. In a complex IT environment, however, keeping them fully consistent is difficult, and discrepancies are a real risk. "Bad" data that goes undetected leads to poor decisions and missed SLAs, and ultimately to operational, financial, and legal risk.

Compare enables periodic comparisons — as frequent as needed — between source and target without taking either system offline. It is an easy-to-use yet high-performance tool for detecting out-of-sync data before it affects the business. It can be deployed alongside a real-time replication mechanism or independently.

Use Compare when the target is operationally important.

Replication is running, but trust is not enough. You need proof that the copy still matches the source.

Downtime is not acceptable. Validation must happen while both sides remain online and useful.

The target influences decisions or recovery. Reporting, DR, migration, and active-active setups all need verified consistency.

Challenges in maintaining data consistency

Replication can be healthy and still leave the target incorrect. The most common failure modes are operational, architectural, and human at the same time.

01Migration mismatch

Initial load tools and replication tools may serialize the same value differently, or miss open transactions during cutover.

02Cloud transition risk

Lift-and-shift programs introduce new cross-location consistency and compliance concerns between on-premises and cloud environments.

03Engine differences

Character sets, locales, endianness, or incompatible date and time handling can quietly distort the target copy.

04Schema preparation mistakes

Missing primary keys, unique keys, jobs, scripts, or triggers can create duplicates or missing behavior on the target side.

05Configuration drift

Replication can behave exactly as configured and still violate data-quality expectations, without showing errors in replication logs.

06Missing captured events

Some operations, such as bulk loads that bypass transaction logging, may never reach the replication stream.

07Latency windows

Asynchronous replication always leaves a delay between source change and target visibility, which can break SLAs and audits.

08Manual intervention

Reporting or standby databases are often opened for writes, and user or administrator changes can corrupt the expected copy.

09Application-side writes

A target-connected application can introduce new inconsistency later, even if the environment is aligned today.

Requirements for the solution

A useful consistency-management tool is not just accurate. It also has to be fast, selective, safe for production, and clear enough to support audit and remediation work.

PerformanceFast enough for real systems
Fast, low-impact comparisons across large data volumes.
Low impact on hardware and network resources.
Automatic and manual partitioning for very large tables.
ScopeFlexible enough for mixed environments
Support for heterogeneous databases and multiple comparison modes.
Flexible options for choosing what is compared and how.
Ability to compare only changed data in continuous replication.
OperationsSafe enough for production traffic
Operation on live databases with continuously changing data.
Minimal intrusiveness and zero downtime for source and target systems.
Clear identification of true inconsistencies instead of transient replication delay.
GovernanceUseful enough after the comparison ends
Audit-ready comparison reports and actionable inconsistency reports.
Flexible reporting for different roles and access levels.
Data security plus ease of use, deployment, troubleshooting, and configuration.
Why Compare fits this model

Compare combines low-impact execution with selective comparison logic, live-replication awareness, and production-ready reporting. It complements replication by adding verified consistency rather than assuming it.

How does Compare work?

Compare avoids the brute-force “read everything, compare everything” model. It narrows the work down to the data that matters and separates temporary delay from actual inconsistency.

Step 01Read and normalize rows

Rows are fetched from source and target, and cross-engine data is normalized so that the comparison is meaningful.

Step 02Run a hash-first comparison

Primary keys are compared value-by-value, while non-key columns use a signature by default to reduce network transfer.

Step 03Isolate uncertain rows

In replicated environments, suspicious rows are queued because the target may simply be behind the source for a moment.

Step 04Confirm after the latency window

Rows are rechecked after the configured replication threshold so that Compare can tell delay apart from persistent drift.

StatusIn-flight

The row changed after the preliminary comparison, so Compare cannot confirm a stable inconsistency yet.

StatusIn-sync

The row is confirmed as consistent after the follow-up check.

StatusPersistently out-of-sync

The row remains inconsistent even after the latency-aware confirmation step.

Compare Steps

Hash-first comparison keeps network overhead low, while the confirmation stage filters out rows that are only temporarily out of sync because replication is still catching up.

Architecture

A typical Compare deployment is intentionally modular. Components may run on one host or many, all links are bidirectional, and the repository stores configuration only — not user data.

TopologyOne control plane, distributed execution.
PlacementAgents stay close to source and target to keep reads efficient.
Repository roleConfiguration only. No business payload is stored there.
CoreCompare Server

Configures jobs, launches comparisons, and exposes reports through the web interface.

ExecutionRuntime CLI

Runs comparison jobs from repository configuration and fits scheduled or scripted execution.

Data accessAgents

Sit close to the source and target databases, fetch rows, and pass them to the comparison process.

Control planeRepository and helpers

Store configuration, support import and export, and keep operational tooling separate from business data.

Web clientHTTPCompare Servercompares hashesAgent (source)fetches, hashesAgent (target)fetches, hashesSourcedatabaseTargetdatabaseRepositoryJDBCJDBCTCP/IP + SSLJDBC

Compare components

Databases

JDBC Direct database connectivity

TCP/IP + SSL Encrypted agent-to-server traffic

Supported databases

Compare is aimed at enterprise databases that commonly appear in replication, modernization, and disaster-recovery programs.

Compare currently supports the following enterprise-class databases:

PostgreSQL logo

PostgreSQL

Oracle logo

Oracle

Informix logo

Informix

IBM Db2 logo

IBM DB2

Homogeneous and heterogeneous comparisons

The comparison model works both for same-engine validation and for cross-platform migrations where normalization is required before any result is trustworthy.

Same engineHomogeneous comparison

Use it when source and target run the same database type, such as Oracle to Oracle, and you need confirmation that replication or recovery stayed aligned.

Cross-engineHeterogeneous comparison

Use it when moving between engines, such as Oracle to PostgreSQL or Informix to Db2, and Compare must normalize values before checking them.

Why this matters

The same workflow can validate steady-state replication, cloud migration, and gradual platform modernization without forcing a separate toolchain for each scenario.

Benefits

The main business value is not the comparison itself. It is the confidence to keep critical copies online, auditable, and safe to use in real decisions.

OutcomeCatch issues early

Detect migration gaps and replication errors before they turn into reporting mistakes or recovery failures.

OperationsKeep systems online

Validate consistency without long re-instantiation windows or disruptive manual copy procedures.

RepairLower MTTR

Shorten the time between discovering an inconsistency and understanding exactly where action is needed.

ReadinessVerify before reuse

Confirm that production or test databases are safe to bring back into service after incidents or migration work.

RiskReduce business exposure

Cut operational, financial, and legal risk created by inaccurate copies of critical data.

ConfidenceTrust the copy you operate

Use reporting, standby, or cloud targets with confirmed consistency instead of assumption-based confidence.

How it works

Prefer a short walkthrough before diving into configuration details? This video summarizes the product flow and the confirmation logic on live data.

Product walkthrough

Watch the end-to-end flow before you configure anything.

Live systemsWorkflow overviewConfirmation logic
What you will seeA concise walkthrough of the product surface, task flow, and what happens after a potential mismatch is detected.
Why it helpsIt gives readers a fast mental model before they move into deployment, configuration, and operations details.
Best moment to watchUse it as an orientation pass for new stakeholders, technical evaluators, or teams planning rollout workshops.

Short walkthrough of the Compare workflow, configuration surface, and confirmation stage for live data streams.