Exodus Migrations

Exodus is Eden's migration product. It uses Eve endpoints, interlays, telemetry, and workflow state to move live database workloads from a source system to a target system while keeping validation and rollback in the operator's hands.

Use Exodus when a database migration cannot be treated as a one-time copy. The product is designed for migrations that need staged testing, traffic observation, live validation, dual operation, and rollback before commit.

What Exodus Does

Exodus puts Eve in the migration path:

text
Applications
    |
    v
Exodus interlay and migration workflow
    |
    +-- Source database
    +-- Target database
    +-- Observability and validation

The source remains available until the operator commits the migration. The target is prepared, backfilled, tested, compared, and promoted through explicit workflow phases.

Migration Phases

The customer-facing migration model has seven phases.

PhaseTraffic pathPurpose
Pre-testApplications still talk directly to source; traffic is copied to Exodus in parallel.Prove that Exodus can observe production-shaped traffic without changing the authoritative path.
AnalyzeApplications cut over to Exodus; Exodus forwards to source.Measure workload shape, compatibility, throughput, command mix, and migration risk.
CompareSource remains authoritative.Mirror safe operations to target and compare results without serving target responses.
TestExodus uses the observed workload and target configuration.Run validation, smoke tests, compatibility checks, and rollback rehearsal.
MigrateApplications continue through Exodus.Move historical data while coordinating live writes according to the selected strategy.
RollbackApplications continue through Exodus.Return traffic to source when validation, latency, or business checks fail.
CommitTarget becomes authoritative.Complete the migration and retire source only when operational criteria are met.

Supported Endpoint Families

Endpoint familyProduct status in docsNotes
Redis and Valkey-compatible systemsGuided workflow availableIncludes Redis Cloud, ElastiCache, MemoryDB, Azure Cache, Valkey, and compatible systems.
PostgreSQLEarly access planning docsRequires transaction-aware cutover, schema-aware comparison, and careful handling of COPY, prepared statements, triggers, and side effects.
MongoDBPreview planning docsRequires document comparison, change stream awareness, session handling, and provider compatibility checks.

See Migrations for the operator runbook and endpoint-specific docs for Redis, PostgreSQL, and MongoDB setup details.

Core Objects

ObjectRole in Exodus
Source endpointThe current authoritative database.
Target endpointThe destination database that will become authoritative after commit.
InterlayThe listener applications connect to so Exodus can forward, mirror, compare, roll back, or cut over traffic.
Workflow runThe guided state machine for setup, validation, approval, execution, retry, and completion.
Migration recordThe lower-level migration object that stores strategy, runtime state, tests, data movement, and compatibility information.
Telemetry streamEvents, metrics, queue depth, divergence, latency, and error details used by dashboards and operators.

Operator Responsibilities

Before a production migration, the operator should:

  • Inventory source and target connection details, protocol versions, TLS settings, credentials, and provider limits.
  • Confirm application traffic can be redirected to an interlay.
  • Establish rollback criteria before execution.
  • Confirm dashboards or SSE subscribers receive migration events.
  • Run pre-test and analyze long enough to capture representative workload behavior.
  • Validate target sizing, command compatibility, and provider feature differences.
  • Keep source available until commit is complete and business owners approve decommissioning.

Strategy Decisions

The Redis workflow exposes several strategy decisions:

DecisionWhy it matters
Migration strategyControls whether the run behaves as big-bang, canary, or blue/green.
Conflict policyControls how backfill handles existing target keys.
Data movement modeControls ordinary source-to-target copy versus bidirectional repair behavior.
Write consistencyControls whether source, target, either side, or both sides determine write success during dual operation.
TTL preservationControls whether Redis key expiration metadata is retained.
Manual approvalKeeps execution behind an explicit human checkpoint.

Keep manual approval enabled for production. Automated runs are useful for repeatable test environments, but production migrations need a named decision owner.

Observability

Exodus should be monitored with:

  • Phase and workflow status.
  • Source and target health.
  • Data movement progress.
  • Mirrored write latency.
  • Replay queue depth.
  • Divergence count and unresolved keys.
  • Compatibility warnings and critical issues.
  • Rollback readiness and commit status.

The migration dashboard should make it clear which system is authoritative at every phase.

Last updated: October 20, 2018
    Eden | Govern AI Access