Replicate Kafka across any vendor mix, with offsets every instance agrees on, schemas that travel with records, and failover you can actually rehearse.
Canon doesn't pick sides. Standard Kafka clients on either end mean any vendor that speaks the protocol is a valid source, a valid target, or both, including bidirectional active-active and N-cluster mesh topologies with target-aware loop prevention built in.
Sources & targets:
Apache Kafka · Confluent Cloud & Platform · Amazon MSK · Redpanda · WarpStream · Red Hat AMQ Streams · IBM Event Streams · any Kafka-protocol broker
Replicates:
Records (with offset-pair agreement) · consumer-group offsets · schemas (Confluent-compatible Schema Registry) · topic configs · ACLs (with delete propagation)
Fail over from Confluent to MSK, from Apache to Redpanda, or between regions of either, with a rehearsable per-direction ceremony, fabric-agreed offset translation, and an audited activity log a regulator can actually read.
Move from one Kafka vendor to another at your pace. Run both sides live, replicate either direction, cut consumers over when the offsets, schemas, and ACLs all match, no Connect cluster, no operator-built audit story.
Run bidirectional links with target-aware loop prevention, so a record that's already visited a cluster doesn't visit it again. N-cluster mesh is a first-class topology, not a sharp edge.
Answer the auditor's question, can you evidence cross-vendor failover, schema lineage, and ACL governance? With a verifiable trail of every replication decision. Built for DORA, Basel III, NYDFS Part 500, and APRA CPS 230.
Define a link between any two Kafka clusters in the Symphony UI. Standard Kafka credentials on each side; Canon handles the rest.
Every Canon instance agrees each (source, target) offset pair, schema mapping, and ACL snapshot through a consensus-backed fabric before the source commit lands.
Drain, fail over, restore, and rehearse from the UI, REST, CLI, or any MCP-speaking agent. The same contract drives every operation.
