feat[kafka]: avro offline schema handling

aws_lambda_powertools/utilities/kafka/consumer_records.py

@@ -41,17 +41,20 @@ def key(self) -> Any:
         schema_type = None
         schema_value = None
         output_serializer = None
+        key_schema_wire_format = None
 
         if self.schema_config and self.schema_config.key_schema_type:
             schema_type = self.schema_config.key_schema_type
             schema_value = self.schema_config.key_schema
             output_serializer = self.schema_config.key_output_serializer
+            key_schema_wire_format = self.schema_config.key_schema_wire_format
 
         # Always use get_deserializer if None it will default to DEFAULT
         deserializer = get_deserializer(
             schema_type=schema_type,
             schema_value=schema_value,
             field_metadata=self.key_schema_metadata,
+            wire_format=key_schema_wire_format,
         )
         deserialized_value = deserializer.deserialize(key)
 
@@ -69,19 +72,22 @@ def value(self) -> Any:
         schema_type = None
         schema_value = None
         output_serializer = None
+        value_schema_wire_format = None
 
         logger.debug("Deserializing value field")
 
         if self.schema_config and self.schema_config.value_schema_type:
             schema_type = self.schema_config.value_schema_type
             schema_value = self.schema_config.value_schema
             output_serializer = self.schema_config.value_output_serializer
+            value_schema_wire_format = self.schema_config.value_schema_wire_format
 
         # Always use get_deserializer if None it will default to DEFAULT
         deserializer = get_deserializer(
             schema_type=schema_type,
             schema_value=schema_value,
             field_metadata=self.value_schema_metadata,
+            wire_format=value_schema_wire_format,
         )
         deserialized_value = deserializer.deserialize(value)
 

aws_lambda_powertools/utilities/kafka/deserializer/avro.py

@@ -2,7 +2,7 @@
 
 import io
 import logging
-from typing import Any
+from typing import Any, Literal
 
 from avro.io import BinaryDecoder, DatumReader
 from avro.schema import parse as parse_schema
@@ -16,6 +16,9 @@
 
 logger = logging.getLogger(__name__)
 
+_CONFLUENT_HEADER_SIZE = 5
+_CONFLUENT_MAGIC_BYTE = 0x00
+
 
 class AvroDeserializer(DeserializerBase):
     """
@@ -25,16 +28,43 @@ class AvroDeserializer(DeserializerBase):
     a provided Avro schema definition.
     """
 
-    def __init__(self, schema_str: str, field_metadata: dict[str, Any] | None = None):
+    def __init__(
+        self,
+        schema_str: str,
+        field_metadata: dict[str, Any] | None = None,
+        wire_format: Literal["CONFLUENT"] | None = None,
+    ):
         try:
             self.parsed_schema = parse_schema(schema_str)
             self.reader = DatumReader(self.parsed_schema)
             self.field_metatada = field_metadata
+            self.wire_format = wire_format
         except Exception as e:
             raise KafkaConsumerAvroSchemaParserError(
                 f"Invalid Avro schema. Please ensure the provided avro schema is valid: {type(e).__name__}: {str(e)}",
             ) from e
 
+    def _strip_wire_format_header(self, value: bytes) -> bytes:
+        if self.wire_format is None:
+            return value
+
+        if self.wire_format != "CONFLUENT":
+            raise KafkaConsumerDeserializationError(f"Unsupported Avro wire format: {self.wire_format}")
+
+        if len(value) < _CONFLUENT_HEADER_SIZE:
+            raise KafkaConsumerDeserializationError(
+                "Invalid Confluent wire format: payload must contain a 5-byte header",
+            )
+
+        if value[0] != _CONFLUENT_MAGIC_BYTE:
+            raise KafkaConsumerDeserializationError(
+                "Invalid Confluent wire format: expected magic byte 0x00",
+            )
+
+        schema_id = int.from_bytes(value[1:_CONFLUENT_HEADER_SIZE], byteorder="big")
+        logger.debug("Deserializing Confluent payload with schema ID %s", schema_id)
+        return value[_CONFLUENT_HEADER_SIZE:]
+
     def deserialize(self, data: bytes | str) -> object:
         """
         Deserialize Avro binary data to a Python dictionary.
@@ -75,9 +105,12 @@ def deserialize(self, data: bytes | str) -> object:
 
         try:
             value = self._decode_input(data)
+            value = self._strip_wire_format_header(value)
             bytes_reader = io.BytesIO(value)
             decoder = BinaryDecoder(bytes_reader)
             return self.reader.read(decoder)
+        except KafkaConsumerDeserializationError:
+            raise
         except Exception as e:
             raise KafkaConsumerDeserializationError(
                 f"Error trying to deserialize avro data - {type(e).__name__}: {str(e)}",

aws_lambda_powertools/utilities/kafka/deserializer/deserializer.py

@@ -1,7 +1,7 @@
 from __future__ import annotations
 
 import hashlib
-from typing import TYPE_CHECKING, Any
+from typing import TYPE_CHECKING, Any, Literal
 
 from aws_lambda_powertools.utilities.kafka.deserializer.default import DefaultDeserializer
 from aws_lambda_powertools.utilities.kafka.deserializer.json import JsonDeserializer
@@ -13,7 +13,12 @@
 _deserializer_cache: dict[str, DeserializerBase] = {}
 
 
-def _get_cache_key(schema_type: str | object, schema_value: Any, field_metadata: dict[str, Any]) -> str:
+def _get_cache_key(
+    schema_type: str | object,
+    schema_value: Any,
+    field_metadata: dict[str, Any],
+    wire_format: Literal["CONFLUENT"] | None,
+) -> str:
     schema_metadata = None
 
     if field_metadata:
@@ -30,10 +35,15 @@ def _get_cache_key(schema_type: str | object, schema_value: Any, field_metadata:
         # For objects like Protobuf, use the object id
         schema_hash = f"{str(id(schema_value))}_{schema_metadata}"
 
-    return f"{schema_type}_{schema_hash}"
+    return f"{schema_type}_{schema_hash}_{wire_format}"
 
 
-def get_deserializer(schema_type: str | object, schema_value: Any, field_metadata: Any) -> DeserializerBase:
+def get_deserializer(
+    schema_type: str | object,
+    schema_value: Any,
+    field_metadata: Any,
+    wire_format: Literal["CONFLUENT"] | None = None,
+) -> DeserializerBase:
     """
     Factory function to get the appropriate deserializer based on schema type.
 
@@ -81,7 +91,7 @@ def get_deserializer(schema_type: str | object, schema_value: Any, field_metadat
     """
 
     # Generate a cache key based on schema type and value
-    cache_key = _get_cache_key(schema_type, schema_value, field_metadata)
+    cache_key = _get_cache_key(schema_type, schema_value, field_metadata, wire_format)
 
     # Check if we already have this deserializer in cache
     if cache_key in _deserializer_cache:
@@ -93,7 +103,11 @@ def get_deserializer(schema_type: str | object, schema_value: Any, field_metadat
         # Import here to avoid dependency if not used
         from aws_lambda_powertools.utilities.kafka.deserializer.avro import AvroDeserializer
 
-        deserializer = AvroDeserializer(schema_str=schema_value, field_metadata=field_metadata)
+        deserializer = AvroDeserializer(
+            schema_str=schema_value,
+            field_metadata=field_metadata,
+            wire_format=wire_format,
+        )
     elif schema_type == "PROTOBUF":
         # Import here to avoid dependency if not used
         from aws_lambda_powertools.utilities.kafka.deserializer.protobuf import ProtobufDeserializer

aws_lambda_powertools/utilities/kafka/schema_config.py

@@ -26,6 +26,14 @@ class SchemaConfig:
         Schema definition for message keys. Required when key_schema_type is 'AVRO' or 'PROTOBUF'.
     key_output_serializer : Any, optional
         Custom serializer for message keys. Supports Pydantic classes, Dataclasses and Custom Class
+    value_schema_wire_format : {'CONFLUENT', None}, default=None
+        Set this when a Confluent schema-registry-aware serializer produced the value payload
+        but you are supplying the Avro schema offline rather than using the ESM Schema Registry integration.
+        Only applies to AVRO values.
+    key_schema_wire_format : {'CONFLUENT', None}, default=None
+        Set this when a Confluent schema-registry-aware serializer produced the key payload
+        but you are supplying the Avro schema offline rather than using the ESM Schema Registry integration.
+        Only applies to AVRO keys.
 
     Raises
     ------
@@ -63,21 +71,39 @@ def __init__(
         key_schema_type: Literal["AVRO", "PROTOBUF", "JSON"] | None = None,
         key_schema: str | None = None,
         key_output_serializer: Any | None = None,
+        value_schema_wire_format: Literal["CONFLUENT"] | None = None,
+        key_schema_wire_format: Literal["CONFLUENT"] | None = None,
     ):
         # Validate schema requirements
         self._validate_schema_requirements(value_schema_type, value_schema, "value")
         self._validate_schema_requirements(key_schema_type, key_schema, "key")
+        self._validate_wire_format(value_schema_wire_format, value_schema_type, "value")
+        self._validate_wire_format(key_schema_wire_format, key_schema_type, "key")
 
         self.value_schema_type = value_schema_type
         self.value_schema = value_schema
         self.value_output_serializer = value_output_serializer
         self.key_schema_type = key_schema_type
         self.key_schema = key_schema
         self.key_output_serializer = key_output_serializer
+        self.value_schema_wire_format = value_schema_wire_format
+        self.key_schema_wire_format = key_schema_wire_format
 
     def _validate_schema_requirements(self, schema_type: str | None, schema: str | None, prefix: str) -> None:
         """Validate that schema is provided when required by schema_type."""
         if schema_type in ["AVRO", "PROTOBUF"] and schema is None:
             raise KafkaConsumerMissingSchemaError(
                 f"{prefix}_schema must be provided when {prefix}_schema_type is {schema_type}",
             )
+
+    def _validate_wire_format(self, wire_format: str | None, schema_type: str | None, prefix: str) -> None:
+        """Validate the wire format for a key or value payload."""
+
+        if wire_format is None:
+            return
+
+        if wire_format != "CONFLUENT":
+            raise ValueError(f"{prefix}_schema_wire_format must be 'CONFLUENT'.")
+
+        if schema_type != "AVRO":
+            raise ValueError(f"{prefix}_schema_wire_format is supported only when {prefix}_schema_type is 'AVRO'.")

docs/utilities/kafka.md

@@ -29,6 +29,7 @@ flowchart LR
 * Support for key and value deserialization
 * Support for custom output serializers (e.g., dataclasses, Pydantic models)
 * Support for ESM with and without Schema Registry integration
+* Support for offline Avro schemas with schema-registry wire-format prefixes (Confluent only)
 * Proper error handling for deserialization issues
 
 ## Terminology
@@ -255,6 +256,45 @@ Each Kafka record contains important metadata that you can access alongside the
 | `value_schema_metadata` | Metadata about the value schema like `schemaId` and `dataFormat` | Data format and schemaId propagated when integrating with Schema Registry |
 | `key_schema_metadata` | Metadata about the key schema like `schemaId` and `dataFormat` | Data format and schemaId propagated when integrating with Schema Registry |
 
+### Using an offline Avro schema with a schema-registry wire-format prefix
+
+When Confluent serializes messages with its schema-registry-aware Avro serializer (for example, `KafkaAvroSerializer`), each payload carries a wire-format header before the Avro body.
+The header is 5 bytes long: 1-byte magic byte (`0x00`) followed by a 4-byte big-endian schema ID.
+
+When the ESM Schema Registry integration is enabled, Lambda strips those bytes and populates the record's schema metadata. When you use an **offline Avro schema** without the ESM Schema Registry integration, the header reaches the function and prevents plain Avro deserialization.
+
+Set `value_schema_wire_format` or `key_schema_wire_format` on `SchemaConfig` to `"CONFLUENT"`. Powertools validates the magic byte and strips the 5-byte header before running the Avro decoder.
+
+???+ info "When do I need this?"
+    Use this option when you supply the Avro schema and the producer uses the Confluent wire format. If ESM Schema Registry integration has already removed the header, leave the option as `None`.
+
+=== "Offline Avro schema with a Confluent prefix"
+
+    ```python hl_lines="10"
+    from aws_lambda_powertools.utilities.kafka import SchemaConfig, kafka_consumer
+    from aws_lambda_powertools.utilities.kafka.consumer_records import ConsumerRecords
+    from aws_lambda_powertools.utilities.typing import LambdaContext
+
+    AVRO_SCHEMA = open("user.avsc").read()
+
+    schema_config = SchemaConfig(
+        value_schema_type="AVRO",
+        value_schema=AVRO_SCHEMA,
+        value_schema_wire_format="CONFLUENT",
+    )
+
+
+    @kafka_consumer(schema_config=schema_config)
+    def lambda_handler(event: ConsumerRecords, context: LambdaContext):
+        for record in event.records:
+            # record.value is the deserialized Avro payload
+            # with the validated 5-byte wire-format header removed.
+            ...
+    ```
+
+???+ warning "Scope"
+    `value_schema_wire_format` and `key_schema_wire_format` apply only to **Avro** payloads. Leave them as `None` when ESM Schema Registry integration has already removed the wire-format header.
+
 ### Custom output serializers
 
 Transform deserialized data into your preferred object types using output serializers. This can help you integrate Kafka data with your domain models and application architecture, providing type hints, validation, and structured data access.