# Specify Iceberg Schema

> For the complete documentation index, see [llms.txt](https://docs.redpanda.com/llms.txt). Component-specific: [cloud-data-platform-full.txt](https://docs.redpanda.com/cloud-data-platform-full.txt)

---
title: Specify Iceberg Schema
latest-operator-version: v26.2.1
latest-console-tag: v3.9.0
latest-connect-version: 4.104.0
latest-redpanda-tag: v26.2.1
docname: iceberg/specify-iceberg-schema
page-component-name: cloud-data-platform
page-version: master
page-component-version: master
page-component-title: Cloud
page-relative-src-path: iceberg/specify-iceberg-schema.adoc
page-edit-url: https://github.com/redpanda-data/cloud-docs/edit/main/modules/manage/pages/iceberg/specify-iceberg-schema.adoc
description: Learn about supported Iceberg modes and how you can integrate schemas with Iceberg topics.
page-git-created-date: "2025-07-31"
page-git-modified-date: "2026-05-26"
---

<!-- Source: https://docs.redpanda.com/cloud-data-platform/manage/iceberg/specify-iceberg-schema.md -->

In [Iceberg-enabled clusters](https://docs.redpanda.com/cloud-data-platform/manage/iceberg/about-iceberg-topics/#enable-iceberg-integration), the `redpanda.iceberg.mode` topic property determines how Redpanda maps topic data to the Iceberg table structure. You can have the generated Iceberg table match the structure of a schema in Schema Registry, or you can use the `key_value` mode where Redpanda stores the record values as-is in the table.

After reading this page, you will be able to:

-   Configure the redpanda.iceberg.mode property when you create or update a topic

-   Choose the Iceberg mode that produces the table structure your data consumers need

-   Apply independent translation for record keys, values, and headers


## [](#supported-iceberg-modes)Supported Iceberg modes

Redpanda supports the following modes for Iceberg topics:

### [](#key_value)key_value

Creates an Iceberg table using a simple schema, consisting of two columns, one for the record metadata including the key, and another binary column for the record’s value.

### [](#value_schema_id_prefix)value_schema_id_prefix

Creates an Iceberg table whose structure matches the Redpanda schema for the topic, with columns corresponding to each field. You must register a schema in [Schema Registry](https://docs.redpanda.com/cloud-data-platform/manage/schema-reg/schema-reg-overview/) and producers must write to the topic using the Schema Registry wire format.

In the [Schema Registry wire format](https://docs.redpanda.com/cloud-data-platform/manage/schema-reg/schema-reg-overview/#wire-format), a "magic byte" and schema ID are embedded in the message payload header. Producers to the topic must use the wire format in the serialization process so Redpanda can determine the schema used for each record, use the schema to define the Iceberg table, and store the topic values in the corresponding table columns.

### [](#value_schema_latest)value_schema_latest

Creates an Iceberg table whose structure matches the latest schema registered for the subject in Schema Registry. You must register a schema in Schema Registry.

Producers cannot use the wire format in `value_schema_latest` mode. Redpanda expects the serialized message as-is without the magic byte or schema ID prefix in the record value.

> 📝 **NOTE**
>
> The `value_schema_latest` mode is not compatible with the [`rpk topic produce`](#reference:rpk/rpk-topic/rpk-topic-produce) command which embeds the wire format header. You must use your own producer code to produce to topics in `value_schema_latest` mode.

The latest schema is cached periodically. The cache period is defined by the cluster property `iceberg_latest_schema_cache_ttl_ms` (default: 5 minutes).

### [](#disabled)disabled

Default for `redpanda.iceberg.mode`. Disables writing to an Iceberg table for the topic.

> 📝 **NOTE**
>
> The following modes are compatible with producing to an Iceberg topic using Redpanda Console:
>
> -   `key_value`
>
> -   Starting in version 25.2, `value_schema_latest` with a JSON schema
>
>
> Otherwise, records may fail to write to the Iceberg table and instead write to the [dead-letter queue](https://docs.redpanda.com/cloud-data-platform/manage/iceberg/iceberg-troubleshooting/#dead-letter-queue).

## [](#configure-iceberg-mode-for-a-topic)Configure Iceberg mode for a topic

You can set the Iceberg mode for a topic when you create the topic, or you can update the mode for an existing topic.

Option 1. Create a new topic and set `redpanda.iceberg.mode`:

```bash
rpk topic create <topic-name> --topic-config=redpanda.iceberg.mode=<iceberg-mode>
```

Option 2. Set `redpanda.iceberg.mode` for an existing topic:

```bash
rpk topic alter-config <topic-name> --set redpanda.iceberg.mode=<iceberg-mode>
```

### [](#override-value-schema-latest-default)Override `value_schema_latest` default

In `value_schema_latest` mode, you only need to set the property value to the string `value_schema_latest`. This enables the default behavior of `value_schema_latest` mode, which determines the subject for the topic using the TopicNameStrategy. For example, if your topic is named `sensor` the schema is looked up in the `sensor-value` subject. For Protobuf data, the default behavior also deserializes records using the first message defined in the corresponding Protobuf schema stored in Schema Registry.

If you use a different strategy other than the topic name to derive the subject name, you can override the default behavior of `value_schema_latest` mode and explicitly set the subject name.

To override the default behavior, use the following optional syntax:

```bash
value_schema_latest:subject=<subject-name>,protobuf_name=<protobuf-message-full-name>
```

-   For both Avro and Protobuf, specify a different subject name by using the key-value pair `subject=<subject-name>`, for example `value_schema_latest:subject=sensor-data`.

-   For Protobuf only:

    -   Specify a different message definition by using a key-value pair `protobuf_name=<message-full-name>`. You must use the fully qualified name, which includes the package name, for example, `value_schema_latest:protobuf_name=com.example.manufacturing.SensorData`.

    -   To specify both a different subject and message definition, separate the key-value pairs with a comma, for example: `value_schema_latest:subject=my_protobuf_schema,protobuf_name=com.example.manufacturing.SensorData`.


    > 📝 **NOTE**
    >
    > If you don’t specify the fully qualified Protobuf message name, Redpanda pauses the data translation to the Iceberg table until you fix the topic misconfiguration.


## [](#configure-key-value-and-header-translation)Configure key, value, and header translation

In addition to the [supported modes](#supported-iceberg-modes), `redpanda.iceberg.mode` also accepts a section-based syntax that lets you independently configure how Redpanda translates the record key, value, and headers into the Iceberg table. The `key_value`, `value_schema_id_prefix`, and `value_schema_latest` modes are shorthand for common combinations of these sections (see [Iceberg mode shorthands](#iceberg-mode-shorthands)).

The `key` and `headers` sections change fields inside the `redpanda` system struct column (`redpanda.key` and the `value` field of each entry in `redpanda.headers`), while the `value` section changes the columns outside that struct. See [How Iceberg modes translate to table format](#how-iceberg-modes-translate-to-table-format) for the base row structure that every generated table includes.

Use the following syntax to configure one or more sections:

```bash
<section>:<option>=<value>,<option>=<value>;<section>:<option>=<value>
```

-   Valid sections are `key`, `value`, and `headers`.

-   Separate sections with `;`.

-   Separate options within a section with `,`.

-   Sections can appear in any order. Any section you omit uses its defaults.


### [](#key-and-value-section-options)Key and value section options

The `key` and `value` sections accept the same options, except for `layout`, which is available only in the `value` section.

| Option | Values | Default | Notes |
| --- | --- | --- | --- |
| mode | binary, schema_id_prefix, schema_latest, string | binary | string stores the field as a UTF-8 string, replacing invalid bytes with the Unicode replacement character (U+FFFD).See Resulting redpanda.key type by key mode and Resulting value field type by value mode for the resulting column types in the generated table. |
| subject | Subject name | Empty | Requires mode=schema_latest. When empty, Redpanda derives the subject using the <topic-name>-key or <topic-name>-value naming convention (TopicNameStrategy). |
| protobuf_name | Fully qualified Protobuf message name | Empty | Requires mode=schema_latest. When empty, Redpanda uses the first message definition in the schema. |
| layout | flat, nested | flat | Value section only. Requires mode=schema_id_prefix or mode=schema_latest. |
| Mode | redpanda.key type | Description |
| --- | --- | --- |
| binary (default) | binary | Raw key bytes; no decoding. |
| schema_id_prefix | struct (from the schema) | Decoded using the schema ID embedded in the key, in Schema Registry wire format. |
| schema_latest | struct (from the schema) | Decoded using the latest schema registered for the subject. |
| string | string | UTF-8 decoded. Invalid bytes are replaced with U+FFFD. |

> 📝 **NOTE**
>
> For `schema_id_prefix` and `schema_latest`, every field in the decoded key struct is optional and an absent field is stored as null, so the table can tolerate schema evolution.

| Mode | Value field type | Description |
| --- | --- | --- |
| binary (default) | binary column named value | Raw value bytes (no decoding). |
| schema_id_prefix | Struct fields. Placement depends on layout. | Decoded using the schema ID embedded in the value, in Schema Registry wire format. |
| schema_latest | Struct fields. Placement depends on layout. | Decoded using the latest schema registered for the subject. |
| string | string column named value | UTF-8 decoded; invalid bytes are replaced with U+FFFD. |

#### [](#key-field-examples)Key field examples

The `key` section supports the same schema-decoding modes as the `value` section (`key:mode=schema_id_prefix` and `key:mode=schema_latest`), but decoded key fields always land in the single `redpanda.key` field, and keys have no `layout` option.

For example, given a key schema with fields `user_id` (int) and `region` (string):

```avro
{
    "type": "record",
    "name": "OrderKey",
    "fields": [
        {
            "name": "user_id",
            "type": "int"
        },
        {
            "name": "region",
            "type": "string"
        }
    ]
}
```

With the default `key:mode=binary`, `redpanda.key` is a single binary field. With `key:mode=schema_id_prefix` or `key:mode=schema_latest`, `redpanda.key` becomes a struct whose fields match the decoded schema:

```sql
redpanda struct<
    ...,
    key: struct<
        user_id: int,
        region:  string
    >,
    ...
>
```

#### [](#value-field-examples)Value field examples

The `layout` option (value section only) controls where decoded value fields appear as columns. By default (`layout=flat`), Redpanda places each decoded value field as a top-level column in the generated table, alongside the `redpanda` system struct. If a decoded value field is named `redpanda`, Redpanda moves it into the `redpanda` system struct as a `data` field, to avoid colliding with the record metadata column of the same name.

Set `layout=nested` to nest all decoded value fields inside a single `value` struct column instead. A field named `redpanda` stays nested under the value column and is unaffected.

For example, using a value schema with fields `user_id` (int) and `region` (string), the default `layout=flat` promotes those fields to top-level columns:

```sql
redpanda struct<
    ...
>,
user_id: int,
region:  string
```

You then query the fields as top-level columns, for example `SELECT user_id, region FROM orders`.

With `layout=nested`, the same fields are wrapped inside a single `value` struct column instead:

```sql
redpanda struct<
    ...
>,
value: struct<
    user_id: int,
    region:  string
>
```

You then query the fields through the `value` struct, for example `SELECT value.user_id, value.region FROM orders`.

### [](#headers-section-options)Headers section options

The `headers` section accepts a single option, `value_type`, which controls how header values are stored in the generated table.

| value_type | Header value type | Description |
| --- | --- | --- |
| binary (default) | binary | Raw header value bytes (no decoding). |
| string | string | UTF-8 decoded. Invalid bytes are replaced with U+FFFD. |

> 📝 **NOTE**
>
> Only header values are affected by `value_type`. Header keys are always stored as strings.

#### [](#header-value-examples)Header value examples

Consider a record with two headers: `content-type` with the value `application/json` (valid UTF-8), and `trace-id` with the invalid UTF-8 byte sequence `0xDEADBEEF`.

With the default `headers:value_type=binary`, both header values are stored as raw bytes:

```sql
redpanda.headers = [
    {key: "content-type", value: b"application/json"},
    {key: "trace-id",     value: b"\xDE\xAD\xBE\xEF"}
]
```

With `headers:value_type=string`, both values are decoded as UTF-8. Valid bytes pass through unchanged, and invalid bytes are replaced with `U+FFFD`:

```sql
redpanda.headers = [
    {key: "content-type", value: "application/json"},
    {key: "trace-id",     value: "����"}    -- Each of the four invalid bytes replaced with U+FFFD (�)
]
```

### [](#iceberg-mode-shorthands)Iceberg mode shorthands

The `key_value`, `value_schema_id_prefix`, and `value_schema_latest` modes are shorthands for common section-based configurations. Use a mode when you don’t need per-section control, and use the section-based syntax when you do.

| Iceberg mode | Equivalent section-based configuration |
| --- | --- |
| disabled | Not applicable. disabled turns off Iceberg table writes for the topic entirely. No key, value, or headers section applies. |
| key_value | key:mode=binary;value:mode=binary;headers:value_type=binary (every section left at its default) |
| value_schema_id_prefix | value:mode=schema_id_prefix (the key and headers sections stay at their defaults) |
| value_schema_latest[:subject=…​,protobuf_name=…​] | value:mode=schema_latest[,subject=…​,protobuf_name=…​] (the key and headers sections stay at their defaults) |

### [](#validation-rules)Validation rules

Redpanda rejects the following configurations when you create or alter a topic:

-   Unknown section names (anything other than `key`, `value`, or `headers`).

-   Duplicate sections, or duplicate options within a section.

-   Empty keys or values in an `<option>=<value>` pair.

-   `subject` or `protobuf_name` set without `mode=schema_latest`.

-   `layout` set in the `key` section.

-   `layout=nested` set without `mode=schema_id_prefix` or `mode=schema_latest`.


> 📝 **NOTE**
>
> Option values cannot contain `,` or `;`, and whitespace is not trimmed. Avoid extra spaces around subject names or Protobuf message names.

### [](#example-configurations)Example configurations

To decode a schema-encoded key and store header values as strings:

```bash
rpk topic alter-config orders --set redpanda.iceberg.mode="key:mode=schema_id_prefix;headers:value_type=string"
```

-   Key: decoded using the schema ID embedded in the key, stored as a struct in `redpanda.key`

-   Value: raw bytes (default)

-   Headers: decoded to UTF-8 strings


To decode the key and value using the latest schema, override the key’s subject, and nest the value fields under a `value` column:

```bash
rpk topic alter-config orders --set redpanda.iceberg.mode="key:mode=schema_latest,subject=orders-key-v2;value:mode=schema_latest,layout=nested"
```

-   Key: decoded using the latest schema registered for the `orders-key-v2` subject, stored as a struct in `redpanda.key`

-   Value: decoded using the latest schema registered for its subject, with fields nested under a `value` struct column

-   Headers: raw bytes (default)


To store the key as a plain UTF-8 string and decode headers to strings:

```bash
rpk topic create events --topic-config redpanda.iceberg.mode="key:mode=string;headers:value_type=string"
```

-   Key: stored as a UTF-8 string in `redpanda.key`

-   Value: raw bytes (default)

-   Headers: decoded to UTF-8 strings


To verify the current configuration:

```bash
rpk topic describe orders -c | grep redpanda.iceberg.mode
```

> 📝 **NOTE**
>
> If a section-based configuration is equivalent to one of the modes described in [Supported Iceberg modes](#supported-iceberg-modes), `rpk topic describe` displays it using that mode’s name instead of the section syntax.

## [](#resolve-schemas-within-a-context)Resolve schemas within a Schema Registry context

If you use [Schema Registry contexts](https://docs.redpanda.com/cloud-data-platform/manage/schema-reg/schema-reg-contexts/) to isolate schemas (for example, by environment or tenant), set the `redpanda.schema.registry.context` topic property to bind the topic to that context. Redpanda then resolves the schemas referenced by records in the topic against the configured context instead of the default context (`.`). This applies to all schema-decoding modes, for both keys and values: the `value_schema_id_prefix` and `value_schema_latest` shorthand modes, and any `key` or `value` section that uses `mode=schema_id_prefix` or `mode=schema_latest`.

Both modes rely on a schema registered in Schema Registry to determine the Iceberg table structure.

Schema Registry contexts are enabled by default. See [Schema Registry contexts](https://docs.redpanda.com/cloud-data-platform/manage/schema-reg/schema-reg-contexts/) to learn about contexts and qualified subject naming before you configure this property.

Set the Schema Registry context on a new topic

```bash
rpk topic create <topic-name> --topic-config redpanda.schema.registry.context=<context-name>
```

Set the Schema Registry context on an existing topic

```bash
rpk topic alter-config <topic-name> --set redpanda.schema.registry.context=<context-name>
```

The context name must start with a period (`.`), for example `.staging`. If you don’t set this property, Redpanda resolves schemas in the default context (`.`).

If Redpanda cannot resolve a record’s schema within the configured context, it doesn’t translate the record and instead writes it to a dead-letter queue (DLQ) table. See [Troubleshoot Iceberg Topics](https://docs.redpanda.com/cloud-data-platform/manage/iceberg/iceberg-troubleshooting/).

> ❗ **IMPORTANT**
>
> Redpanda resolves schemas using the topic’s current `redpanda.schema.registry.context` value, not the context that was active when a record was ingested. Changing the context on a topic that has pending, uncommitted Iceberg translation entries can cause resolution errors or unexpected DLQ routing for records still being translated.
>
> To change the Schema Registry context on a topic that is actively translating to Iceberg:
>
> 1.  Disable Iceberg translation for the topic.
>
> 2.  Wait for pending translation entries to commit.
>
> 3.  Change `redpanda.schema.registry.context`.
>
> 4.  Re-enable Iceberg translation.

## [](#how-iceberg-modes-translate-to-table-format)How Iceberg modes translate to table format

Redpanda generates an Iceberg table with the same name as the topic. In each mode, Redpanda writes to a `redpanda` table column that stores a single Iceberg [struct](https://iceberg.apache.org/spec/#nested-types) per record, containing nested columns of the metadata from each record, including the record key, headers, timestamp, the partition it belongs to, and its offset.

For example, if you produce to a topic `ClickEvent` according to the following Avro schema:

```avro
{
    "type": "record",
    "name": "ClickEvent",
    "fields": [
        {
            "name": "user_id",
            "type": "int"
        },
        {
            "name": "event_type",
            "type": "string"
        },
        {
            "name": "ts",
            "type": "string"
        }
    ]
}
```

The `key_value` mode writes to the following table format:

```sql
CREATE TABLE ClickEvent (
    redpanda struct<
        partition:      integer,
        timestamp:      timestamptz,
        offset:         long,
        headers:        array<struct<key: string, value: binary>>,
        key:            binary,
        timestamp_type: integer
    >,
    value binary
)
```

Use `key_value` mode if you want to use the Iceberg data in its semi-structured format.

The `value_schema_id_prefix` and `value_schema_latest` modes can use the schema to translate to the following table format:

```sql
CREATE TABLE ClickEvent (
    redpanda struct<
        partition: integer,
        timestamp:      timestamptz,
        offset:         long,
        headers:        array<struct<key: string, value: binary>>,
        key:            binary,
        timestamp_type: integer
    >,
    user_id integer NOT NULL,
    event_type string,
    ts string
)
```

As you produce records to the topic, the data also becomes available in object storage for Iceberg-compatible clients to consume. You can use the same analytical tools to [read the Iceberg topic data](https://docs.redpanda.com/cloud-data-platform/manage/iceberg/query-iceberg-topics/) in a data lake as you would for a relational database.

If Redpanda fails to translate the record to the columnar format as defined by the schema, it writes the record to a dead-letter queue (DLQ) table. See [Troubleshoot Iceberg Topics](https://docs.redpanda.com/cloud-data-platform/manage/iceberg/iceberg-troubleshooting/) for more information.

> 📝 **NOTE**
>
> By default, Redpanda stores the record key in binary format in the `redpanda.key` column. To decode the key using a schema, or store it as a string, configure the `key` section. See [Configure key, value, and header translation](#configure-key-value-and-header-translation).

### [](#schema-types-translation)Schema types translation

Redpanda supports direct translations of the following types to Iceberg value domains:

#### Avro

| Avro type | Iceberg type |
| --- | --- |
| boolean | boolean |
| int | int |
| long | long |
| float | float |
| double | double |
| bytes | binary |
| string | string |
| record | struct |
| array | list |
| map | map |
| fixed | fixed* |
| decimal | decimal |
| uuid | uuid* |
| date | date |
| time | time* |
| timestamp | timestamp |

\*These types are not currently supported in Unity Catalog managed Iceberg tables.

There are some cases where the Avro type does not map directly to an Iceberg type and Redpanda applies the following transformations:

-   Enums are translated into the Iceberg `string` type.

-   Different flavors of time (such as `time-millis`) and timestamp (such as `timestamp-millis`) types are translated to the same Iceberg `time` and `timestamp` types, respectively.

-   Avro unions are flattened to Iceberg structs with optional fields. For example:

    -   The union `["int", "long", "float"]` is represented as an Iceberg struct `struct<0 INT NULLABLE, 1 LONG NULLABLE, 2 FLOAT NULLABLE>`.

    -   The union `["int", null, "float"]` is represented as an Iceberg struct `struct<0 INT NULLABLE, 1 FLOAT NULLABLE>`.


-   Two-field unions that contain `null` are represented as a single optional field only (no struct). For example, the union `["null", "long"]` is represented as `long`.


Some Avro types are not supported:

-   The Avro `duration` logical type is ignored.

-   The Avro `null` type is ignored and not represented in the Iceberg schema.

-   Recursive types are not supported.

#### Protobuf

| Protobuf type | Iceberg type |
| --- | --- |
| bool | boolean |
| double | double |
| float | float |
| int32 | int |
| sint32 | int |
| int64 | long |
| sint64 | long |
| sfixed32 | int |
| sfixed64 | long |
| string | string |
| bytes | binary |
| map | map |
| message | struct |

There are some cases where the Protobuf type does not map directly to an Iceberg type and Redpanda applies the following transformations:

-   Repeated values are translated into Iceberg `list` types.

-   Enums are translated into the Iceberg `string` type.

-   `uint32` and `fixed32` are translated into Iceberg `long` types as that is the existing semantic for unsigned 32-bit values in Iceberg.

-   `uint64` and `fixed64` values are translated into their Base-10 string representation.

-   `google.protobuf.Timestamp` is translated into `timestamp` in Iceberg.


Recursive types are not supported.

#### JSON Schema

Requirements:

-   Only JSON Schema Draft-07 is currently supported.

-   You must declare the JSON Schema dialect using the `$schema` keyword, for example `"$schema": "http://json-schema.org/draft-07/schema#"`.

-   You must use a JSON Schema that constrains JSON documents to a strict type so Redpanda can translate to Iceberg. In most cases this means each subschema uses the `type` keyword, but a subschema can also use `$ref` if the referenced schema resolves to a strict type.


Valid JSON Schema example

```json
{
  "$schema": "http://json-schema.org/draft-07/schema#",
  "type": "object",
  "properties": {
    "productId": {
      "type": "integer"
    },
    "tags": {
      "type": "array",
      "items": {
        "type": "string"
      }
    }
  }
}
```

| JSON type | Iceberg type | Notes |
| --- | --- | --- |
| array | list | The keywords items and additionalItems must be used to constrain element types. |
| boolean | boolean |  |
| null |  | The null type is only supported as a nullability marker, either in a type array (for example, ["string", "null"]) or in an exclusive oneOf nullable pattern. |
| number | double |  |
| integer | long |  |
| string | string | The format keyword can be used for custom Iceberg types. See format annotation translation for details. |
| object | struct or map | Use properties to define struct fields and constrain their types. additionalProperties: false is supported for closed objects.If additionalProperties contains a schema, it translates to an Iceberg map<string, T>.You cannot combine properties and additionalProperties in an object if additionalProperties is set to a schema. |
| format value | Iceberg type |
| --- | --- |
| date-time | timestamptz |
| date | date |
| time | time |

The following keywords have specific behavior:

-   The `$ref` keyword is supported for internal references resolved from schema resources declared in the same document (using `$id`), including relative and absolute URI forms. References to external resources and references to unknown keywords are not supported. A root-level `$ref` schema is not supported.

-   The `oneOf` keyword is supported only for the nullable serializer pattern where exactly one branch is `{"type":"null"}` and the other branch is a non-null schema (`T|null`).

-   In Iceberg output, Redpanda writes all fields as nullable regardless of serializer nullability annotations.


The following are not supported for JSON Schema:

-   The `$dynamicRef` keyword

-   The `default` keyword

-   Conditional typing (`if`, `then`, `else`, `dependencies` keywords)

-   Boolean JSON Schema combinations (`allOf`, `anyOf`, and non-nullable `oneOf` patterns)

-   Dynamic object members with the `patternProperties` keyword

-   The `additionalProperties` keyword when set to `true`