Skip to main content
GIFRÖST / Compare / Connections

Push heavy comparison work closer to the data.

Agent-Based Connections place local agents near the source and target databases. That design reduces network pressure, keeps sorting and hashing close to each engine, and scales better when the reconciliation scope becomes large or geographically distributed. It is also the only execution model that can run the Chunk Hash algorithm.

Lower network pressureMore processing happens near the databases instead of across long central links.
Better for large scopeUseful for heavier table volumes and multi-environment reconciliation runs.
Higher operational controlAgents give more control over execution locality, but introduce deployment overhead.

How it works

In agent mode, Compare coordinates the run while local agents handle part of the read, sort, and hash workload close to each database. This architecture lowers central data movement and helps when source and target engines behave differently under direct ordering and encoding rules. It is also the mode required by Chunk Hash, where agents exchange chunk signatures first and return the first mismatching decrypted chunk only when Compare needs an exact diff.

AspectAgent-Based Connections
TopologyLocal agents run near the source and target, while Compare remains the control plane.
Operational advantageLower network transfer, better scalability, and more stable behavior across heterogeneous engines.
Main constraintExtra rollout, lifecycle management, and local infrastructure requirements for each agent.
Exclusive algorithmChunk Hash is available only here. It compares hash windows first and expands only the first mismatching chunk into a detailed diff.
Typical fitLarge tables, remote environments, or cases where direct central reads are too expensive.

Currently supported databases

Agent-Based Connections currently support only PostgreSQL and Oracle.

PostgreSQL logo

PostgreSQL

Currently supported for agent mode

8.4+
Oracle logo

Oracle

Currently supported for agent mode

21c19c18c12.2

Best-fit scenarios

01

The reconciliation scope is large or long-running

Agent mode is the safer default when central reads would create too much data movement or extend runtime unacceptably.

02

Source and target sit in different network zones

Local agents reduce dependency on long network paths between the Compare host and each database endpoint.

03

Sorting and encoding behavior differs between engines

Keeping heavy preparation work close to each engine reduces comparison noise caused by platform-specific ordering differences.

04

You want the first exact difference without moving every row

Chunk Hash is agent-only and keeps payload low by comparing hashes chunk-by-chunk, then requesting full values only for the first differing chunk.

Deployment checklist

01

Place each agent near its database

Deploy agents where database access is local, low-latency, and operationally stable.

02

Validate secure communication paths

Confirm that the Compare control plane and both agents can exchange control traffic reliably.

03

Check agent resources and version alignment

CPU, memory, and agent version consistency matter once hashing and sorting move off the central host.

04

Run endpoint and agent health tests

Validate both database access and agent readiness before defining mappings and execution rules.

Installation and security model

Agent-Based Connections are enabled by installing dedicated agent software. The agent package is available for macOS, Linux, Windows, and Docker deployments.

AspectAgent operating model
Installation methodInstall the agent software on macOS, Linux, Windows, or run it as a Docker deployment.
Logical bindingEach agent is logically associated with exactly one database endpoint.
Communication securityTraffic between the agent and Compare is authenticated with a Bearer Token.
Operational resultThe execution model stays explicit: one agent, one database, one controlled Compare link.
When direct mode is enough

Stay with Direct Database Connections when the workload is modest, the network path is simple, and the team wants the lowest operational overhead.