Operational replicas and event streaming
Capture each change once. Give every consumer its own pace.
Omni Loader writes one source stream to a durable spool, then advances relational replicas and event sinks on independent cursors.
Independent fan-outOne committed stream. Every consumer connected.
4 cursors advancing
Durable streamOne ordered history
PostgreSQLIndependent cursor
Azure SQL HyperScaleIndependent cursor
KafkaIndependent cursor
Fabric MirrorOpen Mirroring
1 → many
Independent fan-out
Add destinations without adding source readers.
DB
Operational replicas
Maintain queryable copies with keyed apply.
Events
Kafka and Pub/Sub
Publish table-aware change events.
One durable stream
Every destination reads the same committed history.
Database targets use convergent apply while Kafka and Pub/Sub receive serialized events. Each sink owns its checkpoint, batching, and error policy.
✓ Independent cursor and lag per sink
✓ Pause or add a sink without restarting capture
✓ Table-aware Kafka topic routing
✓ Stable event deduplication identifiers
Mixed consumers
Relational apply and event delivery remain distinct jobs.
Destination-specific workers preserve the shared source order while choosing the apply contract appropriate to each target.
✓ Target-specific batching and connection reuse
✓ JSON and Protobuf event serialization
✓ Retry, skip, fail, and dead-letter policies
✓ Source-only monitoring when no sink is needed
Keep exploring
See the connected parts of the platform.
Give every downstream system the same committed truth.
List the replicas, topics, subscriptions, and SLAs. We will map them into independent delivery contracts.