Migrate Redis.
Zero downtime.
Exodus deploys as a RESP wire-protocol proxy with concurrent dual-write, adaptive throttling, and instant rollback. No code changes, no maintenance windows.
Supported Systems
Redis Cloud
ElastiCacheMemoryDB
Azure Cache
Google Memorystore
DigitalOcean
Valkey
KeyDBSelf-HostedHow it works
11 steps to a zero-downtime migration
Diagrams show an Azure example, but the process is identical across all cloud providers and systems.
Watch our founder walk through a live migration demo
Current State/ Direct Connection
Your applications connect directly to your existing database. No proxy, no intermediary.
Deploy Exodus/ Passive Listener
Exodus deploys alongside the existing stack. Passively listens on the wire with no connection to your database and no production impact.
Analyze/ Traffic + Metadata
Exodus runs a full analysis of traffic patterns from the wire capture plus database metadata (memory, schema, config, replication state). Builds a complete profile of the system.
Recommend Target/ Optimal Deployment
Based on traffic analysis, Exodus recommends the optimal target database with the right configuration and sizing.
Connect Target/ Establish Link
Exodus establishes connections to both databases. The target is provisioned and ready. Data replication begins during migration to maintain consistent state.
Shadow Test/ Shadow Validation
Exodus replays production traffic against the target and compares responses to the source. Validates correctness and latency with zero customer impact.
Redirect to Exodus/ Proxy Activated
Update your connection string to point at Exodus. Exodus proxies everything to the source database while mirroring all writes to the target.
Canary Migration/ Gradual Traffic Shift
Exodus replicates data and shifts reads together: 5% to 20% to 40% to 80%. Both happen in lockstep to maintain consistent state. Instant rollback at every step.
Complete Migration/ Full Cutover
Migration complete. All traffic flows through Exodus to the target database. The source is idle, still running as a fallback.
Decommission Source/ Clean Up
Source database decommissioned. Exodus stays as the data orchestration layer, monitoring, optimizing, and ready for the next migration.
Remove Exodus/ Optional
Optionally remove Exodus from the stack. Your applications connect directly to the target database. No lock-in, no residual dependencies.
Built for production Redis
Every feature driven by real-world migration challenges
Full Data Type Coverage
Strings, Lists, Sets, Sorted Sets, and Hashes with type-specific merge and conflict resolution. Hash fields skip-merged, lists concatenated, sets unioned, sorted set scores updated.
Concurrent Dual-Write
Writes fire to both endpoints simultaneously, minimizing latency overhead. Six write consistency policies from BestEffort to BothRequired control how results are adjudicated.
Queue Conflict Prevention
Every dual-write is checked against the active migration pipeline for key conflicts. Exponential backoff prevents data races between application writes and data movement.
Six Read Routing Modes
Old, New, Ratio-based split, FallbackOnMiss, VersionCompare, and Replicated. Replicated mode compares responses byte-for-byte and logs divergence at WARN level.
RESP Wire Protocol
Full RESP2 and RESP3 support at the wire level. Commands classified as read or write via static lookup with unknown commands defaulting to write for safety.
TTL Preservation
Source key TTLs captured via PTTL and included in RESTORE commands on the target. Key expiration semantics are preserved exactly across the migration.
Divergence Detection
Replicated read routing performs full byte comparison between both endpoints. Divergence logged at WARN level with response sizes for operator visibility.
Error Export and Replay
Keys that fail migration after all retries are exported for later playback and reconciliation. Full error context preserved, no silent data loss.
Tunable Batch Sizing
Configure batch limits by key count and byte size. Fine-tune batch dimensions to match your instance capacity and network throughput.
Adaptive Throttling
Data movement speed adjusts to leave CPU overhead for live traffic. Migration never starves production workloads.
Configuration Replication
Source Redis configuration is analyzed and replicated to the target. Memory policies, eviction rules, and server settings carried over automatically.
Datadog Observability
Per-request metrics exported to Datadog for full migration visibility. Track latency, error rates, and throughput across both endpoints in real time.
Multiple migration strategies
Choose the right approach for your risk tolerance and timeline
Big Bang
All traffic switches to the target at once. Fastest cutover for systems that can tolerate a brief maintenance window.
Canary
Gradual traffic shift from 5% to 100% with instant rollback at every step. The safest path for most production systems.
Blue/Green
Run both environments in parallel, validate the target under real load, then switch over with a single routing change.
User-by-User
Migrate individual users or tenants incrementally. Ideal for multi-tenant systems where per-tenant rollback is required.
Shadow
Mirror production traffic to the target for validation without affecting live users. Proves correctness before any cutover.
Every correctness gap, documented
16 gaps identified in correctness review. Here's how each one is handled.