Apache Ozone

Use it when

• An HDFS estate is being modernized in place; the same data serves both the S3 protocol and the ofs Hadoop filesystem interface, so jobs migrate gradually.
• The NameNode ceiling is the problem: namespace and block management are separated, scaling to billions of objects.
• The surrounding stack is Hadoop: Kerberos authentication, Ranger authorization, and transparent encryption at rest integrate directly.
• Erasure coding on dense commodity nodes matters for capacity economics.

Think twice when

• You need S3 features beyond the basics; versioning, object lock, SSE through the gateway, event notifications, and lifecycle tiering are all missing or on the roadmap.
• Multi-site is required: there is no multisite replication, only snapshot diff and DistCp to another cluster.
• The team is small; Ozone Manager, SCM, DataNodes, and an S3 gateway (three OMs for HA) put its operational complexity at the high end, alongside Ceph.
• There is no Hadoop legacy at all, where MinIO or cloud object storage is simpler.

How it runs

A self-hosted Java system from the Hadoop community: Ozone Manager holds the namespace, Storage Container Manager places data, DataNodes store it, and an S3 gateway translates. Durability is 3x Ratis replication or Reed-Solomon erasure coding per bucket.