OSO Engineering

A Kafka Schema Registry backup must protect registry state alongside the Kafka records that depend on it. Capture subjects, versions, IDs, references, compatibility settings, and modes through the supported Registry API. Restore that state before starting consumers against recovered records.

Backing up records alone can leave valid bytes that applications cannot decode. Treat Kafka data and Schema Registry as one recovery set with one tested recovery point.

Kafka backup verification requires more than a successful backup job. Inspect the manifest, run quick and deep validation, validate the restore plan, restore into isolation, and test the recovered workload.

Each step finds a different failure class. Only the complete sequence tests stored data, Kafka access, consumer positions, schemas, and application assumptions together.

To backup Kafka to S3, prepare an independent bucket, grant scoped access, select the required topics, and set storage.backend: s3. Then run kafka-backup backup --config s3-backup.yaml and validate the stored result.

This workflow keeps topic data and recovery metadata outside the Kafka cluster. It creates a recovery path that broker replication alone cannot provide.

A Kafka consumer offsets backup must capture committed group positions with the topic records those positions reference. In OSO Kafka Backup, enable include_offset_headers and consumer_group_snapshot, then store both under the same backup ID.

During recovery, restore records before resetting consumer groups. This order keeps a saved position tied to data that actually exists on the target cluster.

Kafka Replicator is Confluent's Kafka Connect connector for asynchronously copying selected topics from one Kafka cluster to another. It can preserve topic settings, translate consumer positions, and fit into an existing Confluent Platform operating model.

Replicator keeps a second Kafka cluster close to the source's current state. It does not create historical restore points, so it cannot recover the state before a bad write or deletion.

A Kafka MirrorMaker 2 setup needs two cluster aliases, a directional flow, topic and group filters, and a running dedicated Connect process. You must then verify remote topics, copied records, translated consumer offsets, and measured replication delay.

MirrorMaker 2 provides asynchronous cross-cluster replication. It does not keep a historical restore point, and a running connector does not prove that an application can fail over safely.

Kafka replication factor is the number of brokers that store each topic partition. A factor of three gives each partition one leader replica and two follower replicas. Kafka can then keep serving the partition when one broker fails, provided another eligible replica is current.

More replicas improve tolerance for broker and disk failure. They also consume more storage, network bandwidth, and recovery capacity. The right value depends on your failure domains and durability target, not a universal rule.

Kafka MirrorMaker 2 is Apache Kafka's framework for copying records and selected metadata between independent Kafka clusters. It runs on Kafka Connect, reads from a source cluster, and writes to a target cluster through a directional replication flow.

MirrorMaker helps with regional availability, cluster migration, and centralized data access. It does not create an independent historical backup, and it cannot restore a topic to a time before a bad record or deletion.

Kafka replication stores copies of each partition on multiple brokers. One replica leads reads and writes while followers copy its ordered log and stand ready to take over. This design keeps partitions available through many broker failures. It does not create a historical backup or protect every copy from a bad delete.

Theory says your Kafka DR plan will work. Production has other ideas. These Kafka disaster recovery stories walk through four failures that actually happen — a region outage, an accidental topic deletion, a multi-datacenter partition, and a Kubernetes broker cascade — and the choices that decided whether recovery took minutes or hours. Learning from someone else's outage is cheaper than living through your own.