Feature matrix
What Ursa supports today across formats, clouds and catalogs.
This page summarizes Ursa's supported formats, storage backends, and catalogs. ✔ marks supported and ✖ marks not supported. Support for individual features does not certify every cloud/catalog combination or imply production readiness.
Data formats
Entry format is what's read off the wire; WAL object and compacted object formats are how Ursa stores that data internally, before and after compaction.
Entry format
| Kafka MemoryRecords | Arrow | |
|---|---|---|
| Supported | ✔ | ✖ |
The current reactor does not include the historical Pulsar adapters.
WAL object format
| Ursa WAL Format | |
|---|---|
| Supported | ✔ |
Compacted object format
| Ursa WAL Format | Parquet | Kafka Segment Format | |
|---|---|---|---|
| Supported | ✔ | ✔ | ✖ |
Storage backends
Cloud WAL implementations cover S3, GCS, and Azure Blob. FileStorage also includes a LOCAL backend for development; that backend is not shared, diskless object storage.
WAL storage
Latency-optimized backends:
| Backend | Status |
|---|---|
| AWS FSx | ✖ |
| Distributed Filesystem | ✖ |
| Regional Disks | ✖ |
Cost-optimized backends:
| AWS S3 | GCS | Azure Blob Store | |
|---|---|---|---|
| Supported | ✔ | ✔ | ✔ |
Lakehouse storage
| AWS S3 | GCS | Azure Blob Store | HDFS | |
|---|---|---|---|---|
| Supported | ✔ | ✔ | ✔ | ✖ |
Schema
Ursa's current materialization path supports Kafka schemas.
Kafka Schema
| AVRO | JSON | PROTOBUF | |
|---|---|---|---|
| Supported | ✔ | ✔ | ✔ |
Kafka schema integration works with Confluent-compatible schema registries. The Kafka quickstart runs Karapace in that role.
Lakehouse tables
Table format and catalog support determine which query engines can read Ursa's compacted output without a conversion step.
Table format
| Apache Iceberg | Delta Lake | Apache Hudi | Apache Paimon | Lance | |
|---|---|---|---|---|---|
| Supported | ✔ | ✔ | ✖ | ✖ | ✖ |
Catalog
| Databricks Unity Catalog | Apache Iceberg REST Catalog | Snowflake Open Catalog | AWS Glue Catalog | |
|---|---|---|---|---|
| Supported | ✔ | ✔ | ✔ | ✖ |
Materialization sinks versus compacted storage
The materialization framework includes ICEBERG, DELTA, DELTA_UC, and CLICKHOUSE catalog types, plus NONE for storage-only compaction. ClickHouse is implemented in ursa-storage-clickhouse as an external sink; it is not a lakehouse file format or a WAL backend. External materialization is distinct from the internal compacted objects used for streaming reads. See table catalogs, lakehouse tables and implementation status for which policy fields each materializer applies.
Apache BookKeeper is not a selectable backend in the current FileStorage implementation; the former "in progress" entry is not a current support commitment.