Type mapping reference¶

This page covers how identifiers are converted from PostgreSQL to Snowflake on mirror targets, and how PostgreSQL data types map to Iceberg.

Name mapping¶

Schema, table, and column names from the source PostgreSQL database are converted to uppercase in the target database. For example, the PostgreSQL column created_at is written as CREATED_AT in the target table. Replicating Postgres tables with the same name in different case isn’t supported.

Type mapping¶

PostgreSQL types are classified as either supported or unsupported. Supported types are replicated directly (native) or cast to text/binary. Unsupported types are blocked when you add a table to the publication.

Native types¶

These types have direct Iceberg equivalents and are replicated without conversion:

Fallback types¶

Types not listed above are cast to string or binary when written to Iceberg. This includes types like json, jsonb, hstore, and vector.

Some fallback types have restrictions:

PostGIS geometry values are stored as WKB in a BINARY column rather than a Snowflake GEOMETRY. Cast at query time with TO_GEOMETRY if you need Snowflake geospatial functions.

Unsupported types¶

These types are blocked when you add a table to the publication:

Domains are transparently resolved to their base types. A domain over an unsupported type is also blocked.

Nested type handling¶

Arrays and composite types have special rules depending on their element or field types:

Arrays:

Composite types:

uuid is an exception to the “supported type → yes” rule: it’s supported as a standalone column but is blocked when nested inside an array or composite type. Tables with nested UUID columns are rejected at publication time.

Value clamping¶

Even for supported types, certain special values can’t be represented in Iceberg. These values are silently clamped at write time.

Temporal values (date, timestamp, timestamptz):

Infinity and out-of-range temporal values are clamped to the nearest Iceberg boundary:

Numeric NaN:

NaN in bounded numeric columns (for example, numeric(20,5)) is clamped to NULL because Iceberg’s decimal type doesn’t support NaN.

Numeric Infinity (unbounded):

Unbounded numeric and numeric(p,s) with p > 38 or s > 38 are converted to double precision. IEEE 754 Infinity and -Infinity are valid double values and are preserved (not clamped).

Multidimensional arrays:

PostgreSQL allows multidimensional array values (for example, ARRAY[[1,2],[3,4]]), but Iceberg’s list type is single-dimensional. Multidimensional array values are clamped to NULL at write time. A column declared as int[] that contains a 1D value (for example, ARRAY[1,2,3]) is stored normally; only values with more than one dimension are nullified.

Oversize string and binary values:

Snowflake imposes maximum sizes on string and binary columns (16 MiB for strings, 8 MiB for binary). Values that exceed these limits are clamped at write time. The cap is inclusive: a value at exactly the limit is stored verbatim; only values that exceed it are clamped.