AWS announced that the Amazon S3 Tables service now supports all data types included in the Apache Iceberg V3 specification, enabling the creation of new V3 tables or the upgrade of existing V2-based tables. The update includes deletion vectors, row lineage, and the variant, nanosecond timestamp, geometry, geography, and unknown data types.
These capabilities target analytics teams managing large tables on data lakes. Instead of representing semi-structured data, geographic coordinates, or ultra-high-precision timestamps as strings or integers, they can be stored as native types inside Iceberg tables, while continuing to use Parquet files on Amazon S3.
What changes in practice?
Deletion vectors replace the positional delete files used in Iceberg V2 with a compact binary representation. According to the example presented by AWS, deleting 50,000 records from a table containing two billion rows can produce a single deletion-vector file instead of thousands of small files, reducing file overhead and subsequent compaction time.
Row lineage automatically adds the _row_id and _last_updated_sequence_number fields. Subsequent processing pipelines can use the sequence number to extract rows that have changed since a previous checkpoint, instead of scanning the entire table on every run.
The variant type stores semi-structured data in a columnar format, allowing the query engine to read the required fields directly and use statistics to reduce input and output operations, rather than parsing JSON text for every query. Iceberg V3 also provides native types for geospatial data and nanosecond-precision timestamps.
Upgrading from Iceberg V2
An existing table can be upgraded atomically by changing the format-version property to 3, without rewriting the data. V2 readers continue to work on tables upgraded to V3 until V3 adoption is complete in the environment, while S3 Tables removes the old delete files during the next compaction cycle and begins initializing row-lineage data upon the first subsequent data modification.
However, this upgrade is one-way; the Apache Iceberg specification does not support reverting from V3 to V2. Therefore, it is necessary to verify in advance that all engines accessing the table support the new version.
Compatibility and limitations
The new data types require an engine built on Apache Spark 4.0 or later, such as AWS Glue 6.0 or later, or Amazon EMR version 8.1 or later. These types are also limited to tables that use Parquet and are not available with ORC or Avro.
variant, geometry, geography, and nanosecond-precision timestamp columns cannot be used in the table ordering used for compaction, although tables containing them remain eligible for compaction when their ordering is based on columns of other types. S3 Tables continues to manage compaction and maintenance automatically, with support for creating tables through the Amazon S3 console, AWS CLI, or any engine that supports the Iceberg REST Catalog interface.
Why does this update matter?
The update gives data teams a path that relies less on custom transformations when handling large-scale deletion, irregular data, and incremental updates. However, the practical benefit depends on the readiness of the organization’s query and write engines; the upgrade does not mean immediate compatibility for every tool, and using the new types restricts the storage format and requires specific versions of Spark and related services.
Apache Iceberg V3 support is now available in all AWS Regions that support S3 Tables, with no additional charges specific to the update, while the standard S3 Tables pricing continues to apply.