airbyte-cdk 6.21.1.dev0__py3-none-any.whl → 6.26.0.dev4103__py3-none-any.whl

airbyte_cdk/sources/declarative/incremental/concurrent_partition_cursor.py

@@ -0,0 +1,334 @@
+#
+# Copyright (c) 2023 Airbyte, Inc., all rights reserved.
+#
+
+import copy
+import logging
+import threading
+from collections import OrderedDict
+from copy import deepcopy
+from datetime import timedelta
+from typing import Any, Callable, Iterable, Mapping, MutableMapping, Optional
+
+from airbyte_cdk.sources.connector_state_manager import ConnectorStateManager
+from airbyte_cdk.sources.declarative.incremental.global_substream_cursor import (
+    Timer,
+    iterate_with_last_flag_and_state,
+)
+from airbyte_cdk.sources.declarative.partition_routers.partition_router import PartitionRouter
+from airbyte_cdk.sources.message import MessageRepository
+from airbyte_cdk.sources.streams.checkpoint.per_partition_key_serializer import (
+    PerPartitionKeySerializer,
+)
+from airbyte_cdk.sources.streams.concurrent.cursor import ConcurrentCursor, Cursor, CursorField
+from airbyte_cdk.sources.streams.concurrent.partitions.partition import Partition
+from airbyte_cdk.sources.types import Record, StreamSlice, StreamState
+
+logger = logging.getLogger("airbyte")
+
+
+class ConcurrentCursorFactory:
+    def __init__(self, create_function: Callable[..., ConcurrentCursor]):
+        self._create_function = create_function
+
+    def create(
+        self, stream_state: Mapping[str, Any], runtime_lookback_window: Optional[timedelta]
+    ) -> ConcurrentCursor:
+        return self._create_function(
+            stream_state=stream_state, runtime_lookback_window=runtime_lookback_window
+        )
+
+
+class ConcurrentPerPartitionCursor(Cursor):
+    """
+    Manages state per partition when a stream has many partitions, preventing data loss or duplication.
+
+    Attributes:
+        DEFAULT_MAX_PARTITIONS_NUMBER (int): Maximum number of partitions to retain in memory (default is 10,000).
+
+    - **Partition Limitation Logic**
+      Ensures the number of tracked partitions does not exceed the specified limit to prevent memory overuse. Oldest partitions are removed when the limit is reached.
+
+    - **Global Cursor Fallback**
+      New partitions use global state as the initial state to progress the state for deleted or new partitions. The history data added after the initial sync will be missing.
+
+    CurrentPerPartitionCursor expects the state of the ConcurrentCursor to follow the format {cursor_field: cursor_value}.
+    """
+
+    DEFAULT_MAX_PARTITIONS_NUMBER = 10000
+    _NO_STATE: Mapping[str, Any] = {}
+    _NO_CURSOR_STATE: Mapping[str, Any] = {}
+    _GLOBAL_STATE_KEY = "state"
+    _PERPARTITION_STATE_KEY = "states"
+    _KEY = 0
+    _VALUE = 1
+
+    def __init__(
+        self,
+        cursor_factory: ConcurrentCursorFactory,
+        partition_router: PartitionRouter,
+        stream_name: str,
+        stream_namespace: Optional[str],
+        stream_state: Any,
+        message_repository: MessageRepository,
+        connector_state_manager: ConnectorStateManager,
+        cursor_field: CursorField,
+    ) -> None:
+        self._global_cursor: Optional[StreamState] = {}
+        self._stream_name = stream_name
+        self._stream_namespace = stream_namespace
+        self._message_repository = message_repository
+        self._connector_state_manager = connector_state_manager
+        self._cursor_field = cursor_field
+
+        self._cursor_factory = cursor_factory
+        self._partition_router = partition_router
+
+        # The dict is ordered to ensure that once the maximum number of partitions is reached,
+        # the oldest partitions can be efficiently removed, maintaining the most recent partitions.
+        self._cursor_per_partition: OrderedDict[str, ConcurrentCursor] = OrderedDict()
+        self._semaphore_per_partition: OrderedDict[str, threading.Semaphore] = OrderedDict()
+        self._finished_partitions: set[str] = set()
+        self._lock = threading.Lock()
+        self._timer = Timer()
+        self._new_global_cursor: Optional[StreamState] = None
+        self._lookback_window: int = 0
+        self._parent_state: Optional[StreamState] = None
+        self._over_limit: int = 0
+        self._partition_serializer = PerPartitionKeySerializer()
+
+        self._set_initial_state(stream_state)
+
+    @property
+    def cursor_field(self) -> CursorField:
+        return self._cursor_field
+
+    @property
+    def state(self) -> MutableMapping[str, Any]:
+        states = []
+        for partition_tuple, cursor in self._cursor_per_partition.items():
+            if cursor.state:
+                states.append(
+                    {
+                        "partition": self._to_dict(partition_tuple),
+                        "cursor": copy.deepcopy(cursor.state),
+                    }
+                )
+        state: dict[str, Any] = {self._PERPARTITION_STATE_KEY: states}
+
+        if self._global_cursor:
+            state[self._GLOBAL_STATE_KEY] = self._global_cursor
+        if self._lookback_window is not None:
+            state["lookback_window"] = self._lookback_window
+        if self._parent_state is not None:
+            state["parent_state"] = self._parent_state
+        return state
+
+    def close_partition(self, partition: Partition) -> None:
+        # Attempt to retrieve the stream slice
+        stream_slice: Optional[StreamSlice] = partition.to_slice()  # type: ignore[assignment]
+
+        # Ensure stream_slice is not None
+        if stream_slice is None:
+            raise ValueError("stream_slice cannot be None")
+
+        partition_key = self._to_partition_key(stream_slice.partition)
+        self._cursor_per_partition[partition_key].close_partition(partition=partition)
+        with self._lock:
+            self._semaphore_per_partition[partition_key].acquire()
+            cursor = self._cursor_per_partition[partition_key]
+            if (
+                partition_key in self._finished_partitions
+                and self._semaphore_per_partition[partition_key]._value == 0
+            ):
+                if (
+                    self._new_global_cursor is None
+                    or self._new_global_cursor[self.cursor_field.cursor_field_key]
+                    < cursor.state[self.cursor_field.cursor_field_key]
+                ):
+                    self._new_global_cursor = copy.deepcopy(cursor.state)
+        self._emit_state_message()
+
+    def ensure_at_least_one_state_emitted(self) -> None:
+        """
+        The platform expect to have at least one state message on successful syncs. Hence, whatever happens, we expect this method to be
+        called.
+        """
+        if not any(
+            semaphore_item[1]._value for semaphore_item in self._semaphore_per_partition.items()
+        ):
+            self._global_cursor = self._new_global_cursor
+            self._lookback_window = self._timer.finish()
+            self._parent_state = self._partition_router.get_stream_state()
+        self._emit_state_message()
+
+    def _emit_state_message(self) -> None:
+        self._connector_state_manager.update_state_for_stream(
+            self._stream_name,
+            self._stream_namespace,
+            self.state,
+        )
+        state_message = self._connector_state_manager.create_state_message(
+            self._stream_name, self._stream_namespace
+        )
+        self._message_repository.emit_message(state_message)
+
+    def stream_slices(self) -> Iterable[StreamSlice]:
+        if self._timer.is_running():
+            raise RuntimeError("stream_slices has been executed more than once.")
+
+        slices = self._partition_router.stream_slices()
+        self._timer.start()
+        for partition in slices:
+            yield from self._generate_slices_from_partition(partition)
+
+    def _generate_slices_from_partition(self, partition: StreamSlice) -> Iterable[StreamSlice]:
+        # Ensure the maximum number of partitions is not exceeded
+        self._ensure_partition_limit()
+
+        cursor = self._cursor_per_partition.get(self._to_partition_key(partition.partition))
+        if not cursor:
+            cursor = self._create_cursor(
+                self._global_cursor,
+                self._lookback_window if self._global_cursor else 0,
+            )
+            self._cursor_per_partition[self._to_partition_key(partition.partition)] = cursor
+            self._semaphore_per_partition[self._to_partition_key(partition.partition)] = (
+                threading.Semaphore(0)
+            )
+
+        for cursor_slice, is_last_slice, _ in iterate_with_last_flag_and_state(
+            cursor.stream_slices(),
+            lambda: None,
+        ):
+            self._semaphore_per_partition[self._to_partition_key(partition.partition)].release()
+            if is_last_slice:
+                self._finished_partitions.add(self._to_partition_key(partition.partition))
+            yield StreamSlice(
+                partition=partition, cursor_slice=cursor_slice, extra_fields=partition.extra_fields
+            )
+
+    def _ensure_partition_limit(self) -> None:
+        """
+        Ensure the maximum number of partitions is not exceeded. If so, the oldest added partition will be dropped.
+        """
+        while len(self._cursor_per_partition) > self.DEFAULT_MAX_PARTITIONS_NUMBER - 1:
+            self._over_limit += 1
+            oldest_partition = self._cursor_per_partition.popitem(last=False)[
+                0
+            ]  # Remove the oldest partition
+            logger.warning(
+                f"The maximum number of partitions has been reached. Dropping the oldest partition: {oldest_partition}. Over limit: {self._over_limit}."
+            )
+
+    def _set_initial_state(self, stream_state: StreamState) -> None:
+        """
+        Initialize the cursor's state using the provided `stream_state`.
+
+        This method supports global and per-partition state initialization.
+
+        - **Global State**: If `states` is missing, the `state` is treated as global and applied to all partitions.
+          The `global state` holds a single cursor position representing the latest processed record across all partitions.
+
+        - **Lookback Window**: Configured via `lookback_window`, it defines the period (in seconds) for reprocessing records.
+          This ensures robustness in case of upstream data delays or reordering. If not specified, it defaults to 0.
+
+        - **Per-Partition State**: If `states` is present, each partition's cursor state is initialized separately.
+
+        - **Parent State**: (if available) Used to initialize partition routers based on parent streams.
+
+        Args:
+            stream_state (StreamState): The state of the streams to be set. The format of the stream state should be:
+                {
+                    "states": [
+                        {
+                            "partition": {
+                                "partition_key": "value"
+                            },
+                            "cursor": {
+                                "last_updated": "2023-05-27T00:00:00Z"
+                            }
+                        }
+                    ],
+                    "state": {
+                        "last_updated": "2023-05-27T00:00:00Z"
+                    },
+                    lookback_window: 10,
+                    "parent_state": {
+                        "parent_stream_name": {
+                            "last_updated": "2023-05-27T00:00:00Z"
+                        }
+                    }
+                }
+        """
+        if not stream_state:
+            return
+
+        if self._PERPARTITION_STATE_KEY not in stream_state:
+            # We assume that `stream_state` is in a global format that can be applied to all partitions.
+            # Example: {"global_state_format_key": "global_state_format_value"}
+            self._global_cursor = deepcopy(stream_state)
+            self._new_global_cursor = deepcopy(stream_state)
+
+        else:
+            self._lookback_window = int(stream_state.get("lookback_window", 0))
+
+            for state in stream_state[self._PERPARTITION_STATE_KEY]:
+                self._cursor_per_partition[self._to_partition_key(state["partition"])] = (
+                    self._create_cursor(state["cursor"])
+                )
+                self._semaphore_per_partition[self._to_partition_key(state["partition"])] = (
+                    threading.Semaphore(0)
+                )
+
+            # set default state for missing partitions if it is per partition with fallback to global
+            if self._GLOBAL_STATE_KEY in stream_state:
+                self._global_cursor = deepcopy(stream_state[self._GLOBAL_STATE_KEY])
+                self._new_global_cursor = deepcopy(stream_state[self._GLOBAL_STATE_KEY])
+
+        # Set initial parent state
+        if stream_state.get("parent_state"):
+            self._parent_state = stream_state["parent_state"]
+
+        # Set parent state for partition routers based on parent streams
+        self._partition_router.set_initial_state(stream_state)
+
+    def observe(self, record: Record) -> None:
+        if not record.associated_slice:
+            raise ValueError(
+                "Invalid state as stream slices that are emitted should refer to an existing cursor"
+            )
+        self._cursor_per_partition[
+            self._to_partition_key(record.associated_slice.partition)
+        ].observe(record)
+
+    def _to_partition_key(self, partition: Mapping[str, Any]) -> str:
+        return self._partition_serializer.to_partition_key(partition)
+
+    def _to_dict(self, partition_key: str) -> Mapping[str, Any]:
+        return self._partition_serializer.to_partition(partition_key)
+
+    def _create_cursor(
+        self, cursor_state: Any, runtime_lookback_window: int = 0
+    ) -> ConcurrentCursor:
+        cursor = self._cursor_factory.create(
+            stream_state=deepcopy(cursor_state),
+            runtime_lookback_window=timedelta(seconds=runtime_lookback_window),
+        )
+        return cursor
+
+    def should_be_synced(self, record: Record) -> bool:
+        return self._get_cursor(record).should_be_synced(record)
+
+    def _get_cursor(self, record: Record) -> ConcurrentCursor:
+        if not record.associated_slice:
+            raise ValueError(
+                "Invalid state as stream slices that are emitted should refer to an existing cursor"
+            )
+        partition_key = self._to_partition_key(record.associated_slice.partition)
+        if partition_key not in self._cursor_per_partition:
+            raise ValueError(
+                "Invalid state as stream slices that are emitted should refer to an existing cursor"
+            )
+        cursor = self._cursor_per_partition[partition_key]
+        return cursor

airbyte_cdk/sources/declarative/incremental/global_substream_cursor.py

@@ -64,6 +64,9 @@ class Timer:
         else:
             raise RuntimeError("Global substream cursor timer not started")
 
+    def is_running(self) -> bool:
+        return self._start is not None
+
 
 class GlobalSubstreamCursor(DeclarativeCursor):
     """

airbyte_cdk/sources/declarative/incremental/per_partition_cursor.py

@@ -222,6 +222,8 @@ class PerPartitionCursor(DeclarativeCursor):
         next_page_token: Optional[Mapping[str, Any]] = None,
     ) -> Mapping[str, Any]:
         if stream_slice:
+            if self._to_partition_key(stream_slice.partition) not in self._cursor_per_partition:
+                self._create_cursor_for_partition(self._to_partition_key(stream_slice.partition))
             return self._partition_router.get_request_params(  # type: ignore # this always returns a mapping
                 stream_state=stream_state,
                 stream_slice=StreamSlice(partition=stream_slice.partition, cursor_slice={}),
@@ -244,6 +246,8 @@ class PerPartitionCursor(DeclarativeCursor):
         next_page_token: Optional[Mapping[str, Any]] = None,
     ) -> Mapping[str, Any]:
         if stream_slice:
+            if self._to_partition_key(stream_slice.partition) not in self._cursor_per_partition:
+                self._create_cursor_for_partition(self._to_partition_key(stream_slice.partition))
             return self._partition_router.get_request_headers(  # type: ignore # this always returns a mapping
                 stream_state=stream_state,
                 stream_slice=StreamSlice(partition=stream_slice.partition, cursor_slice={}),
@@ -266,6 +270,8 @@ class PerPartitionCursor(DeclarativeCursor):
         next_page_token: Optional[Mapping[str, Any]] = None,
     ) -> Union[Mapping[str, Any], str]:
         if stream_slice:
+            if self._to_partition_key(stream_slice.partition) not in self._cursor_per_partition:
+                self._create_cursor_for_partition(self._to_partition_key(stream_slice.partition))
             return self._partition_router.get_request_body_data(  # type: ignore # this always returns a mapping
                 stream_state=stream_state,
                 stream_slice=StreamSlice(partition=stream_slice.partition, cursor_slice={}),
@@ -288,6 +294,8 @@ class PerPartitionCursor(DeclarativeCursor):
         next_page_token: Optional[Mapping[str, Any]] = None,
     ) -> Mapping[str, Any]:
         if stream_slice:
+            if self._to_partition_key(stream_slice.partition) not in self._cursor_per_partition:
+                self._create_cursor_for_partition(self._to_partition_key(stream_slice.partition))
             return self._partition_router.get_request_body_json(  # type: ignore # this always returns a mapping
                 stream_state=stream_state,
                 stream_slice=StreamSlice(partition=stream_slice.partition, cursor_slice={}),
@@ -341,8 +349,32 @@ class PerPartitionCursor(DeclarativeCursor):
             )
         partition_key = self._to_partition_key(record.associated_slice.partition)
         if partition_key not in self._cursor_per_partition:
-            raise ValueError(
-                "Invalid state as stream slices that are emitted should refer to an existing cursor"
-            )
+            self._create_cursor_for_partition(partition_key)
         cursor = self._cursor_per_partition[partition_key]
         return cursor
+
+    def _create_cursor_for_partition(self, partition_key: str) -> None:
+        """
+        Dynamically creates and initializes a cursor for the specified partition.
+
+        This method is required for `ConcurrentPerPartitionCursor`. For concurrent cursors,
+        stream_slices is executed only for the concurrent cursor, so cursors per partition
+        are not created for the declarative cursor. This method ensures that a cursor is available
+        to create requests for the specified partition. The cursor is initialized
+        with the per-partition state if present in the initial state, or with the global state
+        adjusted by the lookback window, or with the state to migrate from.
+
+        Note:
+            This is a temporary workaround and should be removed once the declarative cursor
+            is decoupled from the concurrent cursor implementation.
+
+        Args:
+            partition_key (str): The unique identifier for the partition for which the cursor
+            needs to be created.
+        """
+        partition_state = (
+            self._state_to_migrate_from if self._state_to_migrate_from else self._NO_CURSOR_STATE
+        )
+        cursor = self._create_cursor(partition_state)
+
+        self._cursor_per_partition[partition_key] = cursor

airbyte_cdk/sources/declarative/manifest_declarative_source.py

@@ -7,6 +7,7 @@ import logging
 import pkgutil
 from copy import deepcopy
 from importlib import metadata
+from types import ModuleType
 from typing import Any, Dict, Iterator, List, Mapping, Optional, Set
 
 import yaml
@@ -32,6 +33,9 @@ from airbyte_cdk.sources.declarative.models.declarative_component_schema import
     DeclarativeStream as DeclarativeStreamModel,
 )
 from airbyte_cdk.sources.declarative.models.declarative_component_schema import Spec as SpecModel
+from airbyte_cdk.sources.declarative.parsers.custom_code_compiler import (
+    get_registered_components_module,
+)
 from airbyte_cdk.sources.declarative.parsers.manifest_component_transformer import (
     ManifestComponentTransformer,
 )
@@ -59,22 +63,29 @@ class ManifestDeclarativeSource(DeclarativeSource):
     def __init__(
         self,
         source_config: ConnectionDefinition,
+        *,
+        config: Mapping[str, Any] | None = None,
         debug: bool = False,
         emit_connector_builder_messages: bool = False,
         component_factory: Optional[ModelToComponentFactory] = None,
     ):
         """
-        :param source_config(Mapping[str, Any]): The manifest of low-code components that describe the source connector
-        :param debug(bool): True if debug mode is enabled
-        :param component_factory(ModelToComponentFactory): optional factory if ModelToComponentFactory's default behaviour needs to be tweaked
+        Args:
+            config: The provided config dict.
+            source_config: The manifest of low-code components that describe the source connector.
+            debug: True if debug mode is enabled.
+            emit_connector_builder_messages: True if messages should be emitted to the connector builder.
+            component_factory: optional factory if ModelToComponentFactory's default behavior needs to be tweaked.
         """
         self.logger = logging.getLogger(f"airbyte.{self.name}")
-
         # For ease of use we don't require the type to be specified at the top level manifest, but it should be included during processing
         manifest = dict(source_config)
         if "type" not in manifest:
             manifest["type"] = "DeclarativeSource"
 
+        # If custom components are needed, locate and/or register them.
+        self.components_module: ModuleType | None = get_registered_components_module(config=config)
+
         resolved_source_config = ManifestReferenceResolver().preprocess_manifest(manifest)
         propagated_source_config = ManifestComponentTransformer().propagate_types_and_parameters(
             "", resolved_source_config, {}

airbyte_cdk/sources/declarative/models/declarative_component_schema.py

@@ -328,6 +328,16 @@ class LegacyToPerPartitionStateMigration(BaseModel):
     type: Optional[Literal["LegacyToPerPartitionStateMigration"]] = None
 
 
+class Clamping(BaseModel):
+    target: str = Field(
+        ...,
+        description="The period of time that datetime windows will be clamped by",
+        examples=["DAY", "WEEK", "MONTH", "{{ config['target'] }}"],
+        title="Target",
+    )
+    target_details: Optional[Dict[str, Any]] = None
+
+
 class Algorithm(Enum):
     HS256 = "HS256"
     HS384 = "HS384"
@@ -496,8 +506,8 @@ class OAuthAuthenticator(BaseModel):
         examples=["custom_app_id"],
         title="Client ID Property Name",
     )
-    client_id: str = Field(
-        ...,
+    client_id: Optional[str] = Field(
+        None,
         description="The OAuth client ID. Fill it in the user inputs.",
         examples=["{{ config['client_id }}", "{{ config['credentials']['client_id }}"],
         title="Client ID",
@@ -508,8 +518,8 @@ class OAuthAuthenticator(BaseModel):
         examples=["custom_app_secret"],
         title="Client Secret Property Name",
     )
-    client_secret: str = Field(
-        ...,
+    client_secret: Optional[str] = Field(
+        None,
         description="The OAuth client secret. Fill it in the user inputs.",
         examples=[
             "{{ config['client_secret }}",
@@ -614,6 +624,16 @@ class OAuthAuthenticator(BaseModel):
         description="When the token updater is defined, new refresh tokens, access tokens and the access token expiry date are written back from the authentication response to the config object. This is important if the refresh token can only used once.",
         title="Token Updater",
     )
+    profile_assertion: Optional[JwtAuthenticator] = Field(
+        None,
+        description="The authenticator being used to authenticate the client authenticator.",
+        title="Profile Assertion",
+    )
+    use_profile_assertion: Optional[bool] = Field(
+        False,
+        description="Enable using profile assertion as a flow for OAuth authorization.",
+        title="Use Profile Assertion",
+    )
     parameters: Optional[Dict[str, Any]] = Field(None, alias="$parameters")
 
 
@@ -719,7 +739,7 @@ class HttpResponseFilter(BaseModel):
 class TypesMap(BaseModel):
     target_type: Union[str, List[str]]
     current_type: Union[str, List[str]]
-    condition: Optional[str]
+    condition: Optional[str] = None
 
 
 class SchemaTypeIdentifier(BaseModel):
@@ -797,14 +817,11 @@ class DpathFlattenFields(BaseModel):
     field_path: List[str] = Field(
         ...,
         description="A path to field that needs to be flattened.",
-        examples=[
-            ["data"],
-            ["data", "*", "field"],
-        ],
+        examples=[["data"], ["data", "*", "field"]],
         title="Field Path",
     )
     delete_origin_value: Optional[bool] = Field(
-        False,
+        None,
         description="Whether to delete the origin value or keep it. Default is False.",
         title="Delete Origin Value",
     )
@@ -1242,9 +1259,6 @@ class LegacySessionTokenAuthenticator(BaseModel):
 
 
 class JsonParser(BaseModel):
-    class Config:
-        extra = Extra.allow
-
     type: Literal["JsonParser"]
     encoding: Optional[str] = "utf-8"
 
@@ -1457,6 +1471,11 @@ class AuthFlow(BaseModel):
 
 class DatetimeBasedCursor(BaseModel):
     type: Literal["DatetimeBasedCursor"]
+    clamping: Optional[Clamping] = Field(
+        None,
+        description="This option is used to adjust the upper and lower boundaries of each datetime window to beginning and end of the provided target period (day, week, month)",
+        title="Date Range Clamping",
+    )
     cursor_field: str = Field(
         ...,
         description="The location of the value on a record that will be used as a bookmark during sync. To ensure no data loss, the API must return records in ascending order based on the cursor field. Nested fields are not supported, so the field must be at the top level of the record. You can use a combination of Add Field and Remove Field transformations to move the nested field to the top.",
@@ -1680,6 +1699,18 @@ class CompositeErrorHandler(BaseModel):
     parameters: Optional[Dict[str, Any]] = Field(None, alias="$parameters")
 
 
+class ZipfileDecoder(BaseModel):
+    class Config:
+        extra = Extra.allow
+
+    type: Literal["ZipfileDecoder"]
+    parser: Union[GzipParser, JsonParser, JsonLineParser, CsvParser] = Field(
+        ...,
+        description="Parser to parse the decompressed data from the zipfile(s).",
+        title="Parser",
+    )
+
+
 class CompositeRawDecoder(BaseModel):
     type: Literal["CompositeRawDecoder"]
     parser: Union[GzipParser, JsonParser, JsonLineParser, CsvParser]
@@ -1886,7 +1917,7 @@ class SessionTokenAuthenticator(BaseModel):
         description="Authentication method to use for requests sent to the API, specifying how to inject the session token.",
         title="Data Request Authentication",
     )
-    decoder: Optional[Union[JsonDecoder, XmlDecoder]] = Field(
+    decoder: Optional[Union[JsonDecoder, XmlDecoder, CompositeRawDecoder]] = Field(
         None, description="Component used to decode the response.", title="Decoder"
     )
     parameters: Optional[Dict[str, Any]] = Field(None, alias="$parameters")
@@ -2092,6 +2123,7 @@ class SimpleRetriever(BaseModel):
             XmlDecoder,
             GzipJsonDecoder,
             CompositeRawDecoder,
+            ZipfileDecoder,
         ]
     ] = Field(
         None,
@@ -2168,6 +2200,8 @@ class AsyncRetriever(BaseModel):
             IterableDecoder,
             XmlDecoder,
             GzipJsonDecoder,
+            CompositeRawDecoder,
+            ZipfileDecoder,
         ]
     ] = Field(
         None,
@@ -2182,6 +2216,8 @@ class AsyncRetriever(BaseModel):
             IterableDecoder,
             XmlDecoder,
             GzipJsonDecoder,
+            CompositeRawDecoder,
+            ZipfileDecoder,
         ]
     ] = Field(
         None,