Iceberg feature support

Current support matrix for Apache Iceberg sources, sinks, internal tables, catalogs, storage, and maintenance in RisingWave.

RisingWave supports Apache Iceberg in three main workflows:

• Read external Iceberg tables with CREATE SOURCE, ad hoc queries, materialized views, or refreshable CREATE TABLE.
• Write to external Iceberg tables with CREATE SINK.
• Create internal Iceberg tables with CREATE TABLE ... ENGINE = iceberg, where RisingWave manages the table lifecycle and stores data in Iceberg format.

This page summarizes the current feature support. For syntax and examples, follow the linked pages in each section.

Per-table metadata relations

You can query metadata for a single Iceberg object through relation names with an Iceberg metadata suffix:

SELECT * FROM "orders$snapshots";
SELECT * FROM "orders$manifests";
SELECT * FROM "orders$files";

These relations are supported for:

• Iceberg engine tables created with CREATE TABLE ... ENGINE = iceberg
• Iceberg sources created with CREATE SOURCE ... WITH (connector = 'iceberg', ...)
• Standalone Iceberg sinks created with CREATE SINK ... WITH (connector = 'iceberg', ...)

You can use normal batch SELECT operations such as projection, filters, and LIMIT when querying these relations. Time travel is supported for $manifests and $files with FOR SYSTEM_TIME AS OF or FOR SYSTEM_VERSION AS OF, but not for $snapshots.

Support overview

Area	Support	Notes
External Iceberg sources	Supported	Use CREATE SOURCE for continuous ingestion from append-only tables. Schema is inferred from Iceberg metadata.
Refreshable external Iceberg tables	Supported	Use CREATE TABLE ... WITH (connector = 'iceberg', refresh_mode = 'FULL_RELOAD') for on-demand or scheduled full reloads.
External Iceberg sinks	Supported	Use CREATE SINK to write append-only or upsert streams to Iceberg tables.
Internal Iceberg tables	Supported	Use CREATE TABLE ... ENGINE = iceberg to store RisingWave-managed tables in Iceberg format.
Per-table metadata relations	Supported	Query <name>$snapshots, <name>$manifests, and <name>$files for one Iceberg engine table, Iceberg source, or standalone Iceberg sink.
Automatic Iceberg maintenance	Supported	Available for Iceberg sinks and internal Iceberg tables when a dedicated Iceberg compactor is deployed.
Manual maintenance	Supported	Use VACUUM to expire snapshots and VACUUM FULL to compact files and expire snapshots.
Built-in catalog	Supported	Uses RisingWave’s metastore as a JDBC-compatible Iceberg catalog. PostgreSQL or MySQL metastore is required.
External catalogs	Supported	Supports storage, REST, Glue, JDBC, Hive, S3 Tables, Lakekeeper, Snowflake, and Databricks Unity Catalog with catalog-specific limits.

Reading from Iceberg

Capability	Support	Notes
Schema inference	Supported	Iceberg source schemas are inferred automatically. Do not define columns in CREATE SOURCE.
Ad hoc snapshot queries	Supported	Query an Iceberg source directly to read the current table snapshot.
One-time loading	Supported	Append-only and mutable Iceberg tables are supported. RisingWave applies Iceberg equality deletes and position deletes when reading mutable tables.
Continuous ingestion from append-only tables	Supported	Use CREATE SOURCE and build materialized views on top of the source.
Continuous ingestion from mutable tables	Not supported for update/delete propagation	CREATE SOURCE is designed for append-only continuous ingestion. For mutable Iceberg tables, use refreshable CREATE TABLE with FULL_RELOAD when downstream results must reflect updates and deletes.
On-demand full reload	Supported	Use REFRESH TABLE on a refreshable Iceberg table.
Scheduled full reload	Supported	Set refresh_interval_sec together with refresh_mode = 'FULL_RELOAD'.
Equality deletes	Supported	Supported when reading mutable tables in snapshot/full-reload paths.
Position deletes	Supported	Supported when reading mutable tables in snapshot/full-reload paths.
Predicate pushdown	Supported	Filters can be pushed into Iceberg scans where applicable.
Time travel	Supported	Query committed Iceberg snapshots by timestamp or snapshot ID where the target object supports time travel.

For configuration details, see Ingest data from Iceberg tables.

Writing to Iceberg

Capability	Support	Notes
Append-only sink	Supported	Set type = 'append-only'. Append-only sinks must use merge-on-read mode.
Upsert sink	Supported	Set type = 'upsert' and provide primary_key.
Force append-only	Supported	Set force_append_only = 'true' to write an upsert stream as inserts while ignoring deletes.
Exactly-once delivery	Supported	Iceberg sinks default to exactly-once in the normal sink-decoupled mode. If sink_decouple is disabled, exactly-once is disabled automatically unless configured otherwise.
Commit frequency	Supported	Use commit_checkpoint_interval. The default is 60.
Commit retry	Supported	Use commit_retry_num. The default is 8.
Auto-create target table	Supported	Set create_table_if_not_exists = 'true'.
Partitioning	Supported	Use partition_by with column names or Iceberg transforms such as bucket, truncate, year, month, day, and hour.
Sort order keys	Supported	Use order_key when creating the sink.
Merge-on-read	Supported	Default write mode for Iceberg sinks and tables.
Copy-on-write	Supported for upsert only	Append-only sinks and tables must use merge-on-read.
Additive schema evolution	Supported	auto.schema.change = 'true' can propagate upstream ADD COLUMN changes for exactly-once Iceberg sinks.
Iceberg format v2	Supported	Default format version.
Parquet writer settings	Supported	Configure output file size and compression with parameters such as compaction.target_file_size_mb, compaction.write_parquet_compression, and compaction.write_parquet_max_row_group_bytes.

For configuration details, see Deliver data to Iceberg tables and Iceberg write modes.

Internal Iceberg tables

Capability	Support	Notes
CREATE TABLE ... ENGINE = iceberg	Supported	Creates a RisingWave-managed table backed by Iceberg storage.
Primary-key tables	Supported	Upsert semantics are written to Iceberg through the table’s internal sink.
Append-only tables	Supported	Use APPEND ONLY with ENGINE = iceberg.
Inserts and streaming writes	Supported	Write with INSERT or stream data into the table with CREATE SINK ... INTO.
Materialized views on internal Iceberg tables	Supported	Use internal Iceberg tables as inputs to materialized views.
Partitioning	Supported	Use partition_by in table WITH (...) options. The partition key must be compatible with the table primary key.
Commit frequency	Supported	Use commit_checkpoint_interval; the default is 60.
Time travel	Supported	Query committed snapshots with FOR SYSTEM_TIME AS OF or FOR SYSTEM_VERSION AS OF.
Query storage selection	Supported	Use iceberg_query_storage_mode to choose Iceberg columnar reads, Hummock row reads, or optimizer-selected mode.
Built-in catalog	Supported	Set hosted_catalog = true in the Iceberg connection.
Automatic maintenance	Supported	enable_compaction controls background compaction and snapshot expiration.
Manual maintenance	Supported	Use VACUUM or VACUUM FULL on the table.

For configuration details, see Create and manage internal Iceberg tables.

Catalog and storage support

Catalog or storage	Support	Notes
Storage catalog	Supported	Uses the warehouse path directly. Supports object stores such as S3, GCS, and Azure Blob where configured.
REST catalog	Supported	Works with REST-compatible catalogs. Supports OAuth2-style credentials, bearer tokens, custom headers, SigV4 signing, and vended credentials.
Lakekeeper	Supported	Use Lakekeeper as a self-hosted REST catalog.
Amazon S3 Tables	Supported	Uses the REST catalog path with SigV4 signing.
AWS Glue	Supported	Supports separate Glue credentials and assuming IAM roles.
JDBC catalog	Supported	Can be used for external catalogs and for RisingWave’s built-in catalog.
Hive Metastore	Supported	Available as an external Iceberg catalog.
Snowflake catalog	Supported for Iceberg sources	Use Snowflake-managed Iceberg tables as read-only sources.
Databricks Unity Catalog	Supported for Iceberg sinks	Use Unity Catalog for Databricks-managed external Iceberg tables. CREATE SOURCE and CREATE CONNECTION are not supported for this integration.
S3 and S3-compatible storage	Supported	Configure S3 credentials, endpoint, region, path-style access, or config loading.
Google Cloud Storage	Supported	Configure GCS credentials directly or use REST catalog credential vending where applicable.
Azure Blob Storage and ADLS Gen2	Supported	Supported through Iceberg storage/catalog configuration where applicable.

For complete catalog parameters, see Iceberg catalog configuration. For storage parameters, see Object storage configuration.

Maintenance support

Capability	Support	Notes
Automatic compaction	Supported	Set enable_compaction = true. Requires a dedicated Iceberg compactor node.
Snapshot expiration	Supported	Controlled by enable_snapshot_expiration and snapshot expiration parameters.
VACUUM	Supported	Expires old snapshots for an Iceberg sink or internal Iceberg table.
VACUUM FULL	Supported	Runs compaction and then expires old snapshots.
Full compaction	Supported	Rewrites all data files.
Small-files compaction	Supported	Set compaction.type = 'small-files'.
Files-with-delete compaction	Supported	Set compaction.type = 'files-with-delete'.
Dynamic maintenance updates	Supported	Many compaction and snapshot-expiration parameters can be updated with ALTER SINK.

For setup and sizing, see Deploy a dedicated Iceberg compactor and Iceberg table maintenance.

Important limitations

• CREATE SOURCE for Iceberg is intended for append-only continuous ingestion. Use refreshable CREATE TABLE with FULL_RELOAD for mutable external Iceberg tables.
• Do not define columns in an Iceberg CREATE SOURCE; RisingWave infers the schema from Iceberg metadata.
• refresh_mode is supported for CREATE TABLE, not CREATE SOURCE.
• Append-only Iceberg sinks and append-only internal Iceberg tables must use merge-on-read mode.
• Additive schema evolution currently covers ADD COLUMN; column drops, renames, and type changes require separate handling.
• Catalog integrations can have role-specific limits. For example, Snowflake catalog is read-only in RisingWave, while the Databricks Unity Catalog integration is documented for Iceberg sinks.