fix(decoder): auto-detect gzip magic bytes for responses without Content-Encoding header

airbyte_cdk/sources/declarative/decoders/composite_raw_decoder.py

@@ -11,6 +11,8 @@
 from io import BufferedIOBase, TextIOWrapper
 from typing import Any, List, Optional
 
+GZIP_MAGIC_BYTES = b"\x1f\x8b"
+
 import orjson
 import requests
 
@@ -35,15 +37,22 @@ def parse(self, data: BufferedIOBase) -> PARSER_OUTPUT_TYPE:
         """
         Decompress gzipped bytes and pass decompressed data to the inner parser.
 
-        IMPORTANT:
-            - If the data is not gzipped, reset the pointer and pass the data to the inner parser as is.
-
-        Note:
-            - The data is not decoded by default.
+        Auto-detects gzip content by checking for magic bytes (1f 8b) at the start of the data.
+        If the data is not gzip-compressed, it is passed directly to the inner parser as-is.
+        This handles APIs that return gzip-compressed bodies without setting the Content-Encoding header.
         """
+        header = data.read(2)
+        if not header:
+            return
 
-        with gzip.GzipFile(fileobj=data, mode="rb") as gzipobj:
-            yield from self.inner_parser.parse(gzipobj)
+        remaining = data.read()
+        full_data = io.BytesIO(header + remaining)
+
+        if header == GZIP_MAGIC_BYTES:
+            with gzip.GzipFile(fileobj=full_data, mode="rb") as gzipobj:
+                yield from self.inner_parser.parse(gzipobj)
+        else:
+            yield from self.inner_parser.parse(full_data)
 
 
 @dataclass

airbyte_cdk/sources/declarative/parsers/model_to_component_factory.py

@@ -2634,15 +2634,16 @@ def create_gzip_decoder(
         gzip_parser: GzipParser = ModelToComponentFactory._get_parser(model, config)  # type: ignore  # based on the model, we know this will be a GzipParser
 
         if self._emit_connector_builder_messages:
-            # This is very surprising but if the response is not streamed,
-            # CompositeRawDecoder calls response.content and the requests library actually uncompress the data as opposed to response.raw,
-            # which uses urllib3 directly and does not uncompress the data.
-            return CompositeRawDecoder(gzip_parser.inner_parser, False)
+            # When not streaming, CompositeRawDecoder uses response.content which the requests
+            # library auto-decompresses when Content-Encoding is set. However, some APIs return
+            # gzip data without Content-Encoding headers. Using gzip_parser (which auto-detects
+            # gzip magic bytes) ensures decompression works in both cases.
+            return CompositeRawDecoder(gzip_parser, False)
 
         return CompositeRawDecoder.by_headers(
             [({"Content-Encoding", "Content-Type"}, _compressed_response_types, gzip_parser)],
             stream_response=True,
-            fallback_parser=gzip_parser.inner_parser,
+            fallback_parser=gzip_parser,
         )
 
     @staticmethod

unit_tests/sources/declarative/decoders/test_composite_decoder.py

@@ -332,6 +332,83 @@ def test_composite_raw_decoder_csv_parser_without_mocked_response():
         thread.join(timeout=5)  # ensure thread is cleaned up
 
 
+@pytest.mark.parametrize(
+    "content,inner_parser,stream,use_by_headers,expected_count",
+    [
+        pytest.param(
+            generate_csv(encoding="utf-8", delimiter="\t", should_compress=True),
+            CsvParser(encoding="utf-8", delimiter="\\t"),
+            True,
+            False,
+            3,
+            id="gzip_csv_no_header_streamed",
+        ),
+        pytest.param(
+            generate_csv(encoding="iso-8859-1", delimiter="\t", should_compress=True),
+            CsvParser(encoding="iso-8859-1", delimiter="\\t"),
+            True,
+            False,
+            3,
+            id="gzip_csv_no_header_iso_encoding",
+        ),
+        pytest.param(
+            generate_compressed_jsonlines(),
+            JsonLineParser(),
+            True,
+            True,
+            3,
+            id="gzip_jsonl_no_header_by_headers_fallback",
+        ),
+        pytest.param(
+            "".join(generate_jsonlines()).encode("utf-8"),
+            JsonLineParser(),
+            True,
+            False,
+            3,
+            id="non_gzip_passthrough",
+        ),
+        pytest.param(
+            generate_compressed_jsonlines(),
+            JsonLineParser(),
+            False,
+            False,
+            3,
+            id="gzip_no_header_non_streamed",
+        ),
+        pytest.param(
+            b"",
+            JsonLineParser(),
+            True,
+            False,
+            0,
+            id="empty_data",
+        ),
+    ],
+)
+def test_gzip_parser_auto_detection(
+    requests_mock, content, inner_parser, stream, use_by_headers, expected_count
+):
+    requests_mock.register_uri(
+        "GET",
+        "https://airbyte.io/",
+        content=content,
+    )
+    response = requests.get("https://airbyte.io/", stream=stream)
+
+    parser = GzipParser(inner_parser=inner_parser)
+    if use_by_headers:
+        composite_raw_decoder = CompositeRawDecoder.by_headers(
+            [({"Content-Encoding"}, {"gzip"}, parser)],
+            stream_response=stream,
+            fallback_parser=parser,
+        )
+    else:
+        composite_raw_decoder = CompositeRawDecoder(parser=parser, stream_response=stream)
+
+    parsed_records = list(composite_raw_decoder.decode(response))
+    assert len(parsed_records) == expected_count
+
+
 def test_given_response_already_consumed_when_decode_then_no_data_is_returned(requests_mock):
     requests_mock.register_uri(
         "GET", "https://airbyte.io/", content=json.dumps({"test": "test"}).encode()