Iceberg Partitioning¶

pg2iceberg can materialize tables into partitioned Iceberg tables. Partitions are defined per table in the config and are applied during materialization — the staging layer is unaffected.

Configuration¶

Add a partition list under each table's iceberg block:

tables:
  - name: public.orders
    iceberg:
      partition:
        - "day(created_at)"
        - "region"

Multiple partition fields are supported and are applied in the order listed.

Transforms¶

Time-based¶

Apply to date, timestamp, or timestamptz columns.

partition:
  - "day(created_at)"

Identity¶

Partitions by the raw column value. Works with any column type. Use the column name without a transform function.

partition:
  - "region"
  - "status"

Bucket¶

Hashes the column value into N buckets using Murmur3 (per the Iceberg spec). Useful for high-cardinality columns like UUIDs or string IDs where identity partitioning would produce too many small partitions.

Supported column types: integer, bigint, text, varchar, uuid, date, timestamp, timestamptz, numeric, bytea.

partition:
  - "bucket[16](customer_id)"

Truncate¶

Truncates the column value to a width W. For strings, W is the number of UTF-8 code points. For integers and decimals, W is a rounding modulus (values are rounded down to the nearest multiple of W).

Supported column types: integer, bigint, numeric, text, varchar, char, bytea.

partition:
  - "truncate[4](name)"    # first 4 characters of name
  - "truncate[1000](id)"   # id rounded down to nearest 1000

Combining transforms¶

Partition fields can be freely combined:

tables:
  - name: public.events
    iceberg:
      partition:
        - "day(ts)"
        - "bucket[8](user_id)"

Compaction and partitioning¶