mirror of
https://github.com/BerriAI/litellm.git
synced 2026-10-04 02:31:27 +00:00
test(vertex_ai): split GCS URI MIME tests into test_vertex_gemini_gcs_uri_mime.py
This commit is contained in:
S0ngRu1
parent
ca448058ef
commit
527427e58d
2 changed files with 466 additions and 623 deletions
|
|
@ -1282,629 +1282,6 @@ def test_process_gemini_media():
|
|||
assert base64_result["inline_data"]["data"] == "/9j/4AAQSkZJRg..."
|
||||
|
||||
|
||||
def test_process_gemini_media_gcs_explicit_format_skips_litellm_registry():
|
||||
"""Explicit format for gs:// must not require litellm's FILE_MIME_TYPES entry."""
|
||||
from litellm.types.llms.vertex_ai import FileDataType
|
||||
|
||||
result = _process_gemini_media(
|
||||
"gs://bucket/object-no-ext",
|
||||
format="application/octet-stream",
|
||||
)
|
||||
assert result["file_data"] == FileDataType(
|
||||
mime_type="application/octet-stream",
|
||||
file_uri="gs://bucket/object-no-ext",
|
||||
)
|
||||
|
||||
|
||||
def test_process_gemini_media_gcs_explicit_format_still_applies_mime_alias():
|
||||
"""Known MIME aliases still apply when format is explicit for gs:// URIs."""
|
||||
from litellm.types.llms.vertex_ai import FileDataType
|
||||
|
||||
result = _process_gemini_media(
|
||||
"gs://bucket/object-no-ext",
|
||||
format="image/jpg",
|
||||
)
|
||||
assert result["file_data"] == FileDataType(
|
||||
mime_type="image/jpeg",
|
||||
file_uri="gs://bucket/object-no-ext",
|
||||
)
|
||||
|
||||
|
||||
def test_process_gemini_media_gcs_without_extension_raises_clear_error():
|
||||
# Mock the GCS metadata lookup to avoid real outbound HTTP in tests.
|
||||
with patch(
|
||||
"litellm.llms.vertex_ai.gemini.transformation._get_gcs_object_content_type",
|
||||
return_value=None,
|
||||
):
|
||||
with pytest.raises(litellm.BadRequestError) as exc_info:
|
||||
_process_gemini_media("gs://bucket/image-without-extension")
|
||||
|
||||
assert "Unable to determine mime type for gs URI" in str(exc_info.value)
|
||||
|
||||
|
||||
def test_process_gemini_media_gcs_without_extension_uses_gcs_metadata():
|
||||
from litellm.llms.vertex_ai.gemini.transformation import _process_gemini_media
|
||||
from litellm.types.llms.vertex_ai import FileDataType
|
||||
|
||||
with patch(
|
||||
"litellm.llms.vertex_ai.gemini.transformation._get_gcs_object_content_type",
|
||||
return_value="image/jpeg",
|
||||
):
|
||||
result = _process_gemini_media("gs://bucket/image-without-extension")
|
||||
|
||||
assert result["file_data"] == FileDataType(
|
||||
mime_type="image/jpeg", file_uri="gs://bucket/image-without-extension"
|
||||
)
|
||||
|
||||
|
||||
def test_file_block_uses_mime_type_alias_for_extensionless_gcs():
|
||||
from litellm.llms.vertex_ai.gemini.transformation import (
|
||||
_gemini_convert_messages_with_history,
|
||||
)
|
||||
from litellm.types.llms.vertex_ai import FileDataType
|
||||
|
||||
messages = [
|
||||
{
|
||||
"role": "user",
|
||||
"content": [
|
||||
{
|
||||
"type": "file",
|
||||
"file": {
|
||||
"file_id": "gs://bucket/no-extension-object",
|
||||
"mime_type": "application/pdf",
|
||||
},
|
||||
}
|
||||
],
|
||||
}
|
||||
]
|
||||
|
||||
converted = _gemini_convert_messages_with_history(
|
||||
messages=messages, model="gemini-2.5-flash"
|
||||
)
|
||||
assert converted[0]["parts"][0]["file_data"] == FileDataType(
|
||||
mime_type="application/pdf", file_uri="gs://bucket/no-extension-object"
|
||||
)
|
||||
|
||||
|
||||
def test_real_file_id_without_extension_resolves_to_metadata_mime_type():
|
||||
from litellm.llms.vertex_ai.gemini.transformation import (
|
||||
_gemini_convert_messages_with_history,
|
||||
)
|
||||
from litellm.types.llms.vertex_ai import FileDataType
|
||||
|
||||
messages = [
|
||||
{
|
||||
"role": "user",
|
||||
"content": [
|
||||
{
|
||||
"type": "file",
|
||||
"file": {
|
||||
"file_id": "gs://cdn.pagepop.top/gcs/input/source/18/1851d730e543cb91643835612508d7dd0e8d8c52015fe08b7429a7307c62076c",
|
||||
},
|
||||
}
|
||||
],
|
||||
}
|
||||
]
|
||||
|
||||
with patch(
|
||||
"litellm.llms.vertex_ai.gemini.transformation._get_gcs_object_content_type",
|
||||
return_value="image/jpeg",
|
||||
):
|
||||
converted = _gemini_convert_messages_with_history(
|
||||
messages=messages, model="gemini-2.5-flash"
|
||||
)
|
||||
|
||||
assert converted[0]["parts"][0]["file_data"] == FileDataType(
|
||||
mime_type="image/jpeg",
|
||||
file_uri="gs://cdn.pagepop.top/gcs/input/source/18/1851d730e543cb91643835612508d7dd0e8d8c52015fe08b7429a7307c62076c",
|
||||
)
|
||||
|
||||
|
||||
def test_dotted_bucket_name_up_to_222_chars_is_accepted():
|
||||
from litellm.llms.vertex_ai.gemini.transformation import _is_valid_gcs_bucket_name
|
||||
|
||||
dotted_bucket = ("a." * 110) + "aa" # total length = 222
|
||||
assert len(dotted_bucket) == 222
|
||||
assert _is_valid_gcs_bucket_name(dotted_bucket) is True
|
||||
|
||||
|
||||
def test_is_valid_gcs_bucket_name_rejects_too_short():
|
||||
from litellm.llms.vertex_ai.gemini.transformation import _is_valid_gcs_bucket_name
|
||||
|
||||
assert _is_valid_gcs_bucket_name("ab") is False
|
||||
|
||||
|
||||
def test_is_valid_gcs_bucket_name_rejects_too_long_plain():
|
||||
from litellm.llms.vertex_ai.gemini.transformation import _is_valid_gcs_bucket_name
|
||||
|
||||
bucket = "a" * 64
|
||||
assert len(bucket) == 64
|
||||
assert _is_valid_gcs_bucket_name(bucket) is False
|
||||
|
||||
|
||||
def test_is_valid_gcs_bucket_name_rejects_double_dot():
|
||||
from litellm.llms.vertex_ai.gemini.transformation import _is_valid_gcs_bucket_name
|
||||
|
||||
assert _is_valid_gcs_bucket_name("ab..cd") is False
|
||||
|
||||
|
||||
def test_is_valid_gcs_bucket_name_rejects_ip_address_format():
|
||||
"""Bucket names must not look like IPv4 addresses (GCS DNS constraint)."""
|
||||
from litellm.llms.vertex_ai.gemini.transformation import _is_valid_gcs_bucket_name
|
||||
|
||||
assert _is_valid_gcs_bucket_name("1.2.3.4") is False
|
||||
assert _is_valid_gcs_bucket_name("192.168.0.1") is False
|
||||
|
||||
|
||||
def test_is_valid_gcs_bucket_name_rejects_invalid_chars():
|
||||
from litellm.llms.vertex_ai.gemini.transformation import _is_valid_gcs_bucket_name
|
||||
|
||||
assert _is_valid_gcs_bucket_name("Bucket-Upper") is False
|
||||
assert _is_valid_gcs_bucket_name("bucket@name") is False
|
||||
assert _is_valid_gcs_bucket_name("bucket name") is False
|
||||
|
||||
|
||||
def test_is_valid_gcs_bucket_name_rejects_leading_trailing_special():
|
||||
from litellm.llms.vertex_ai.gemini.transformation import _is_valid_gcs_bucket_name
|
||||
|
||||
assert _is_valid_gcs_bucket_name("-mybucket") is False
|
||||
assert _is_valid_gcs_bucket_name("mybucket-") is False
|
||||
assert _is_valid_gcs_bucket_name(".mybucket") is False
|
||||
assert _is_valid_gcs_bucket_name("mybucket.") is False
|
||||
|
||||
|
||||
def test_get_gcs_object_content_type_uses_shared_vertex_base_instance():
|
||||
from litellm.llms.vertex_ai.gemini import transformation as gemini_transformation
|
||||
|
||||
mock_vertex_base = MagicMock()
|
||||
mock_vertex_base.get_access_token.return_value = ("test-token", "test-project")
|
||||
mock_http_response = MagicMock()
|
||||
mock_http_response.is_error = False
|
||||
mock_http_response.status_code = 200
|
||||
mock_http_response.json.return_value = {"contentType": "image/png"}
|
||||
|
||||
mock_http_handler = MagicMock()
|
||||
mock_http_handler.get.return_value = mock_http_response
|
||||
|
||||
with (
|
||||
patch.object(
|
||||
gemini_transformation, "_GCS_METADATA_VERTEX_BASE", mock_vertex_base
|
||||
),
|
||||
patch(
|
||||
"litellm.llms.vertex_ai.gemini.transformation._get_gcs_metadata_http_handler",
|
||||
return_value=mock_http_handler,
|
||||
),
|
||||
):
|
||||
content_type = gemini_transformation._get_gcs_object_content_type(
|
||||
image_url="gs://my-bucket/path/to/image-without-extension",
|
||||
vertex_project="project-123",
|
||||
vertex_credentials="credential-json",
|
||||
)
|
||||
|
||||
assert content_type == "image/png"
|
||||
mock_vertex_base.get_access_token.assert_called_once_with(
|
||||
credentials="credential-json",
|
||||
project_id="project-123",
|
||||
)
|
||||
|
||||
|
||||
def test_process_gemini_media_normalizes_metadata_mime_alias():
|
||||
from litellm.llms.vertex_ai.gemini.transformation import _process_gemini_media
|
||||
from litellm.types.llms.vertex_ai import FileDataType
|
||||
|
||||
with patch(
|
||||
"litellm.llms.vertex_ai.gemini.transformation._get_gcs_object_content_type",
|
||||
return_value="image/jpg",
|
||||
):
|
||||
result = _process_gemini_media("gs://bucket/image-without-extension")
|
||||
|
||||
assert result["file_data"] == FileDataType(
|
||||
mime_type="image/jpeg", file_uri="gs://bucket/image-without-extension"
|
||||
)
|
||||
|
||||
|
||||
def test_process_gemini_media_rejects_unsupported_metadata_mime_type():
|
||||
from litellm.llms.vertex_ai.gemini.transformation import _process_gemini_media
|
||||
|
||||
with patch(
|
||||
"litellm.llms.vertex_ai.gemini.transformation._get_gcs_object_content_type",
|
||||
return_value="application/octet-stream",
|
||||
):
|
||||
with pytest.raises(
|
||||
litellm.BadRequestError, match="File type not supported by gemini"
|
||||
):
|
||||
_process_gemini_media("gs://bucket/image-without-extension")
|
||||
|
||||
|
||||
def test_get_gcs_object_content_type_fails_fast_with_explicit_credentials():
|
||||
from litellm.llms.vertex_ai.gemini import transformation as gemini_transformation
|
||||
|
||||
mock_vertex_base = MagicMock()
|
||||
mock_vertex_base.get_access_token.side_effect = Exception("token failure")
|
||||
with patch.object(
|
||||
gemini_transformation, "_GCS_METADATA_VERTEX_BASE", mock_vertex_base
|
||||
):
|
||||
with pytest.raises(
|
||||
litellm.BadRequestError,
|
||||
match="Unable to fetch GCS metadata with provided Vertex credentials/project",
|
||||
):
|
||||
gemini_transformation._get_gcs_object_content_type(
|
||||
image_url="gs://my-bucket/path/to/image-without-extension",
|
||||
vertex_project="project-123",
|
||||
vertex_credentials="credential-json",
|
||||
)
|
||||
|
||||
|
||||
def test_get_gcs_object_content_type_raises_with_http_details_when_explicit_creds():
|
||||
"""HTTP failure after successful token fetch must not be swallowed as None."""
|
||||
from litellm.llms.vertex_ai.gemini import transformation as gemini_transformation
|
||||
|
||||
mock_vertex_base = MagicMock()
|
||||
mock_vertex_base.get_access_token.return_value = ("test-token", "test-project")
|
||||
|
||||
mock_http_response = MagicMock()
|
||||
mock_http_response.is_error = True
|
||||
mock_http_response.status_code = 403
|
||||
mock_http_response.text = '{"error":{"message":"Permission denied"}}'
|
||||
|
||||
mock_http_handler = MagicMock()
|
||||
mock_http_handler.get.return_value = mock_http_response
|
||||
|
||||
with (
|
||||
patch.object(
|
||||
gemini_transformation, "_GCS_METADATA_VERTEX_BASE", mock_vertex_base
|
||||
),
|
||||
patch(
|
||||
"litellm.llms.vertex_ai.gemini.transformation._get_gcs_metadata_http_handler",
|
||||
return_value=mock_http_handler,
|
||||
),
|
||||
):
|
||||
with pytest.raises(litellm.BadRequestError, match="HTTP 403") as exc_info:
|
||||
gemini_transformation._get_gcs_object_content_type(
|
||||
image_url="gs://my-bucket/path/to/obj",
|
||||
vertex_project="project-123",
|
||||
vertex_credentials="credential-json",
|
||||
)
|
||||
assert "Permission denied" in str(exc_info.value)
|
||||
|
||||
|
||||
def test_get_gcs_object_content_type_returns_none_on_http_error_without_creds():
|
||||
"""Anonymous metadata: HTTP errors stay soft (no surfaced oracle for private objects)."""
|
||||
from litellm.llms.vertex_ai.gemini import transformation as gemini_transformation
|
||||
|
||||
mock_vertex_base = MagicMock()
|
||||
mock_http_response = MagicMock()
|
||||
mock_http_response.is_error = True
|
||||
mock_http_response.status_code = 403
|
||||
mock_http_response.text = "Forbidden"
|
||||
|
||||
mock_http_handler = MagicMock()
|
||||
mock_http_handler.get.return_value = mock_http_response
|
||||
|
||||
with (
|
||||
patch.object(
|
||||
gemini_transformation, "_GCS_METADATA_VERTEX_BASE", mock_vertex_base
|
||||
),
|
||||
patch(
|
||||
"litellm.llms.vertex_ai.gemini.transformation._get_gcs_metadata_http_handler",
|
||||
return_value=mock_http_handler,
|
||||
),
|
||||
):
|
||||
content_type = gemini_transformation._get_gcs_object_content_type(
|
||||
image_url="gs://public-bucket/public-object",
|
||||
)
|
||||
|
||||
assert content_type is None
|
||||
mock_vertex_base.get_access_token.assert_not_called()
|
||||
|
||||
|
||||
def test_get_gcs_object_content_type_without_credentials_skips_auth():
|
||||
"""Without explicit Vertex credentials, must not use the server's default
|
||||
Google credentials to access GCS.
|
||||
|
||||
Prevents the Gemini API-key (Google AI Studio) path from being used as an
|
||||
oracle for private GCS objects (per veria-ai review feedback).
|
||||
"""
|
||||
from litellm.llms.vertex_ai.gemini import transformation as gemini_transformation
|
||||
|
||||
mock_vertex_base = MagicMock()
|
||||
mock_http_response = MagicMock()
|
||||
mock_http_response.is_error = False
|
||||
mock_http_response.status_code = 200
|
||||
mock_http_response.json.return_value = {"contentType": "image/jpeg"}
|
||||
|
||||
mock_http_handler = MagicMock()
|
||||
mock_http_handler.get.return_value = mock_http_response
|
||||
|
||||
with (
|
||||
patch.object(
|
||||
gemini_transformation, "_GCS_METADATA_VERTEX_BASE", mock_vertex_base
|
||||
),
|
||||
patch(
|
||||
"litellm.llms.vertex_ai.gemini.transformation._get_gcs_metadata_http_handler",
|
||||
return_value=mock_http_handler,
|
||||
),
|
||||
):
|
||||
content_type = gemini_transformation._get_gcs_object_content_type(
|
||||
image_url="gs://public-bucket/public-object"
|
||||
)
|
||||
|
||||
# Must not call get_access_token (so default server credentials are not used)
|
||||
mock_vertex_base.get_access_token.assert_not_called()
|
||||
# An anonymous request is still sent, covering publicly-readable objects.
|
||||
mock_http_handler.get.assert_called_once()
|
||||
call_kwargs = mock_http_handler.get.call_args.kwargs
|
||||
headers = call_kwargs.get("headers")
|
||||
assert headers is None or "Authorization" not in headers
|
||||
assert content_type == "image/jpeg"
|
||||
|
||||
|
||||
def test_async_transform_request_body_does_not_block_event_loop():
|
||||
"""When the sync GCS metadata lookup blocks, async_transform_request_body must not block the event loop."""
|
||||
import asyncio
|
||||
import time
|
||||
|
||||
from litellm.llms.vertex_ai.gemini import transformation as gemini_transformation
|
||||
|
||||
messages = [
|
||||
{
|
||||
"role": "user",
|
||||
"content": [
|
||||
{
|
||||
"type": "image_url",
|
||||
"image_url": {"url": "gs://bucket/image-without-extension"},
|
||||
}
|
||||
],
|
||||
}
|
||||
]
|
||||
|
||||
# Simulate a blocking sync httpx.get (as if the real GCS metadata lookup
|
||||
# timed out). If async_transform_request_body does not offload the sync
|
||||
# portion to a worker thread, this blocks the event loop and a concurrent
|
||||
# sleep cannot make progress.
|
||||
def slow_http_get(*args, **kwargs):
|
||||
time.sleep(0.5)
|
||||
response = MagicMock()
|
||||
response.is_error = False
|
||||
response.status_code = 200
|
||||
response.raise_for_status.return_value = None
|
||||
response.json.return_value = {"contentType": "image/png"}
|
||||
return response
|
||||
|
||||
# Stub out the context-caching lookup so it does not make a real HTTP call.
|
||||
async def fake_check_and_create_cache(self, **kwargs):
|
||||
return kwargs["messages"], kwargs["optional_params"], None
|
||||
|
||||
mock_vertex_base = MagicMock()
|
||||
mock_vertex_base.get_access_token.return_value = ("token", "project")
|
||||
|
||||
async def run_scenario() -> float:
|
||||
async def concurrent_sleep() -> float:
|
||||
start = time.monotonic()
|
||||
await asyncio.sleep(0.05)
|
||||
return time.monotonic() - start
|
||||
|
||||
transform_task = asyncio.create_task(
|
||||
gemini_transformation.async_transform_request_body(
|
||||
gemini_api_key=None,
|
||||
messages=messages,
|
||||
api_base=None,
|
||||
model="gemini-2.5-flash",
|
||||
client=None,
|
||||
timeout=None,
|
||||
extra_headers=None,
|
||||
optional_params={},
|
||||
logging_obj=MagicMock(),
|
||||
custom_llm_provider="vertex_ai",
|
||||
litellm_params={},
|
||||
vertex_project=None,
|
||||
vertex_location=None,
|
||||
vertex_auth_header=None,
|
||||
)
|
||||
)
|
||||
sleep_elapsed = await concurrent_sleep()
|
||||
await transform_task
|
||||
return sleep_elapsed
|
||||
|
||||
mock_http_handler = MagicMock()
|
||||
mock_http_handler.get.side_effect = slow_http_get
|
||||
|
||||
with (
|
||||
patch.object(
|
||||
gemini_transformation, "_GCS_METADATA_VERTEX_BASE", mock_vertex_base
|
||||
),
|
||||
patch(
|
||||
"litellm.llms.vertex_ai.gemini.transformation._get_gcs_metadata_http_handler",
|
||||
return_value=mock_http_handler,
|
||||
),
|
||||
patch(
|
||||
"litellm.llms.vertex_ai.context_caching.vertex_ai_context_caching."
|
||||
"ContextCachingEndpoints.async_check_and_create_cache",
|
||||
new=fake_check_and_create_cache,
|
||||
),
|
||||
):
|
||||
sleep_elapsed = asyncio.run(run_scenario())
|
||||
|
||||
# If the loop is not blocked, a 0.05s sleep must not be stretched toward
|
||||
# the sync block duration (0.5s).
|
||||
assert sleep_elapsed < 0.4, (
|
||||
f"Event loop blocked for {sleep_elapsed:.3f}s; "
|
||||
"async_transform_request_body did not offload the sync GCS metadata "
|
||||
"lookup to a worker thread"
|
||||
)
|
||||
|
||||
|
||||
def test_openai_messages_may_need_sync_gcs_metadata_fetch_false_for_plain_text():
|
||||
from litellm.llms.vertex_ai.gemini.transformation import (
|
||||
_openai_messages_may_need_sync_gcs_metadata_fetch,
|
||||
)
|
||||
|
||||
assert (
|
||||
_openai_messages_may_need_sync_gcs_metadata_fetch(
|
||||
[{"role": "user", "content": "hello"}]
|
||||
)
|
||||
is False
|
||||
)
|
||||
|
||||
|
||||
def test_openai_messages_may_need_sync_gcs_metadata_fetch_true_for_extensionless_gs_image_url():
|
||||
from litellm.llms.vertex_ai.gemini.transformation import (
|
||||
_openai_messages_may_need_sync_gcs_metadata_fetch,
|
||||
)
|
||||
|
||||
messages = [
|
||||
{
|
||||
"role": "user",
|
||||
"content": [
|
||||
{
|
||||
"type": "image_url",
|
||||
"image_url": {"url": "gs://bucket/image-without-extension"},
|
||||
}
|
||||
],
|
||||
}
|
||||
]
|
||||
assert _openai_messages_may_need_sync_gcs_metadata_fetch(messages) is True
|
||||
|
||||
|
||||
def test_openai_messages_may_need_sync_gcs_metadata_fetch_false_when_gs_has_file_extension():
|
||||
from litellm.llms.vertex_ai.gemini.transformation import (
|
||||
_openai_messages_may_need_sync_gcs_metadata_fetch,
|
||||
)
|
||||
|
||||
messages = [
|
||||
{
|
||||
"role": "user",
|
||||
"content": [
|
||||
{
|
||||
"type": "image_url",
|
||||
"image_url": {"url": "gs://bucket/image.png"},
|
||||
}
|
||||
],
|
||||
}
|
||||
]
|
||||
assert _openai_messages_may_need_sync_gcs_metadata_fetch(messages) is False
|
||||
|
||||
|
||||
def test_openai_messages_may_need_sync_gcs_metadata_fetch_false_when_extensionless_gs_has_mime_hint():
|
||||
from litellm.llms.vertex_ai.gemini.transformation import (
|
||||
_openai_messages_may_need_sync_gcs_metadata_fetch,
|
||||
)
|
||||
|
||||
messages = [
|
||||
{
|
||||
"role": "user",
|
||||
"content": [
|
||||
{
|
||||
"type": "image_url",
|
||||
"image_url": {
|
||||
"url": "gs://bucket/image-without-extension",
|
||||
"mime_type": "image/png",
|
||||
},
|
||||
}
|
||||
],
|
||||
}
|
||||
]
|
||||
assert _openai_messages_may_need_sync_gcs_metadata_fetch(messages) is False
|
||||
|
||||
|
||||
def test_openai_messages_may_need_sync_gcs_metadata_fetch_true_for_assistant_images_extensionless_gs():
|
||||
from litellm.llms.vertex_ai.gemini.transformation import (
|
||||
_openai_messages_may_need_sync_gcs_metadata_fetch,
|
||||
)
|
||||
|
||||
messages = [
|
||||
{
|
||||
"role": "assistant",
|
||||
"content": [],
|
||||
"images": [
|
||||
{"image_url": {"url": "gs://bucket/gen-without-extension"}},
|
||||
],
|
||||
}
|
||||
]
|
||||
assert _openai_messages_may_need_sync_gcs_metadata_fetch(messages) is True
|
||||
|
||||
|
||||
def test_openai_messages_may_need_sync_gcs_metadata_fetch_false_for_assistant_images_gs_with_extension():
|
||||
from litellm.llms.vertex_ai.gemini.transformation import (
|
||||
_openai_messages_may_need_sync_gcs_metadata_fetch,
|
||||
)
|
||||
|
||||
messages = [
|
||||
{
|
||||
"role": "assistant",
|
||||
"content": [],
|
||||
"images": [{"image_url": {"url": "gs://bucket/gen.png"}}],
|
||||
}
|
||||
]
|
||||
assert _openai_messages_may_need_sync_gcs_metadata_fetch(messages) is False
|
||||
|
||||
|
||||
def test_openai_messages_may_need_sync_gcs_metadata_fetch_false_for_assistant_images_with_mime_hint():
|
||||
from litellm.llms.vertex_ai.gemini.transformation import (
|
||||
_openai_messages_may_need_sync_gcs_metadata_fetch,
|
||||
)
|
||||
|
||||
messages = [
|
||||
{
|
||||
"role": "assistant",
|
||||
"content": [],
|
||||
"images": [
|
||||
{
|
||||
"image_url": {
|
||||
"url": "gs://bucket/gen-no-ext",
|
||||
"mime_type": "image/png",
|
||||
},
|
||||
}
|
||||
],
|
||||
}
|
||||
]
|
||||
assert _openai_messages_may_need_sync_gcs_metadata_fetch(messages) is False
|
||||
|
||||
|
||||
def test_async_transform_request_body_skips_asyncify_for_text_only_requests():
|
||||
"""Hot path: no extension-less gs:// metadata fetch → run sync transform in-loop."""
|
||||
import asyncio
|
||||
from unittest.mock import MagicMock, patch
|
||||
|
||||
from litellm.llms.vertex_ai.gemini import transformation as gemini_transformation
|
||||
|
||||
async def fake_check_and_create_cache(self, **kwargs):
|
||||
return kwargs["messages"], kwargs["optional_params"], None
|
||||
|
||||
async def run():
|
||||
with patch(
|
||||
"litellm.llms.vertex_ai.gemini.transformation.asyncify",
|
||||
side_effect=AssertionError(
|
||||
"asyncify must not run when no extensionless GCS metadata fetch is needed"
|
||||
),
|
||||
):
|
||||
return await gemini_transformation.async_transform_request_body(
|
||||
gemini_api_key=None,
|
||||
messages=[{"role": "user", "content": "hello"}],
|
||||
api_base=None,
|
||||
model="gemini-2.5-flash",
|
||||
client=None,
|
||||
timeout=None,
|
||||
extra_headers=None,
|
||||
optional_params={},
|
||||
logging_obj=MagicMock(),
|
||||
custom_llm_provider="vertex_ai",
|
||||
litellm_params={},
|
||||
vertex_project=None,
|
||||
vertex_location=None,
|
||||
vertex_auth_header=None,
|
||||
)
|
||||
|
||||
with patch(
|
||||
"litellm.llms.vertex_ai.context_caching.vertex_ai_context_caching."
|
||||
"ContextCachingEndpoints.async_check_and_create_cache",
|
||||
new=fake_check_and_create_cache,
|
||||
):
|
||||
body = asyncio.run(run())
|
||||
|
||||
assert body is not None
|
||||
assert "contents" in body
|
||||
|
||||
|
||||
def test_get_image_mime_type_from_url():
|
||||
"""Test the _get_image_mime_type_from_url function for different image URLs"""
|
||||
|
|
|
|||
|
|
@ -0,0 +1,466 @@
|
|||
"""Vertex Gemini: extensionless gs:// MIME + GCS metadata tests.
|
||||
|
||||
Split from test_vertex.py to satisfy CI per-file size limits.
|
||||
"""
|
||||
import asyncio
|
||||
import os
|
||||
import sys
|
||||
import time
|
||||
|
||||
from dotenv import load_dotenv
|
||||
|
||||
load_dotenv()
|
||||
|
||||
import pytest
|
||||
|
||||
import litellm
|
||||
from unittest.mock import MagicMock, patch
|
||||
|
||||
sys.path.insert(0, os.path.abspath("../.."))
|
||||
|
||||
from litellm.llms.vertex_ai.gemini.transformation import _process_gemini_media
|
||||
|
||||
|
||||
def test_process_gemini_media_gcs_explicit_format_octet_stream_and_alias():
|
||||
"""Explicit format bypasses registry; image/jpg alias still applies."""
|
||||
from litellm.types.llms.vertex_ai import FileDataType
|
||||
|
||||
r1 = _process_gemini_media(
|
||||
"gs://bucket/object-no-ext",
|
||||
format="application/octet-stream",
|
||||
)
|
||||
assert r1["file_data"] == FileDataType(
|
||||
mime_type="application/octet-stream",
|
||||
file_uri="gs://bucket/object-no-ext",
|
||||
)
|
||||
r2 = _process_gemini_media("gs://bucket/object-no-ext", format="image/jpg")
|
||||
assert r2["file_data"] == FileDataType(
|
||||
mime_type="image/jpeg",
|
||||
file_uri="gs://bucket/object-no-ext",
|
||||
)
|
||||
|
||||
|
||||
def test_process_gemini_media_gcs_without_extension_errors_and_metadata_mock():
|
||||
with patch(
|
||||
"litellm.llms.vertex_ai.gemini.transformation._get_gcs_object_content_type",
|
||||
return_value=None,
|
||||
):
|
||||
with pytest.raises(litellm.BadRequestError) as exc:
|
||||
_process_gemini_media("gs://bucket/image-without-extension")
|
||||
assert "Unable to determine mime type for gs URI" in str(exc.value)
|
||||
|
||||
from litellm.types.llms.vertex_ai import FileDataType
|
||||
|
||||
with patch(
|
||||
"litellm.llms.vertex_ai.gemini.transformation._get_gcs_object_content_type",
|
||||
return_value="image/jpeg",
|
||||
) as m:
|
||||
r = _process_gemini_media("gs://bucket/image-without-extension")
|
||||
assert r["file_data"] == FileDataType(
|
||||
mime_type="image/jpeg", file_uri="gs://bucket/image-without-extension"
|
||||
)
|
||||
m.assert_called()
|
||||
|
||||
with patch(
|
||||
"litellm.llms.vertex_ai.gemini.transformation._get_gcs_object_content_type",
|
||||
return_value="image/jpg",
|
||||
):
|
||||
r_alias = _process_gemini_media("gs://bucket/image-without-extension")
|
||||
assert r_alias["file_data"]["mime_type"] == "image/jpeg"
|
||||
|
||||
|
||||
def test_process_gemini_media_rejects_gcs_metadata_mime_not_supported_by_gemini():
|
||||
"""Non-empty GCS contentType that fails _normalize_and_validate_gemini_mime_type."""
|
||||
with patch(
|
||||
"litellm.llms.vertex_ai.gemini.transformation._get_gcs_object_content_type",
|
||||
return_value="application/x-litellm-unit-test-unknown-mime",
|
||||
):
|
||||
with pytest.raises(
|
||||
litellm.BadRequestError,
|
||||
match="File type not supported by gemini",
|
||||
):
|
||||
_process_gemini_media("gs://bucket/object-without-extension")
|
||||
|
||||
|
||||
def test_file_block_uses_mime_type_alias_for_extensionless_gcs():
|
||||
from litellm.llms.vertex_ai.gemini.transformation import (
|
||||
_gemini_convert_messages_with_history,
|
||||
)
|
||||
from litellm.types.llms.vertex_ai import FileDataType
|
||||
|
||||
messages = [
|
||||
{
|
||||
"role": "user",
|
||||
"content": [
|
||||
{
|
||||
"type": "file",
|
||||
"file": {
|
||||
"file_id": "gs://bucket/no-extension-object",
|
||||
"mime_type": "application/pdf",
|
||||
},
|
||||
}
|
||||
],
|
||||
}
|
||||
]
|
||||
converted = _gemini_convert_messages_with_history(
|
||||
messages=messages, model="gemini-2.5-flash"
|
||||
)
|
||||
assert converted[0]["parts"][0]["file_data"] == FileDataType(
|
||||
mime_type="application/pdf", file_uri="gs://bucket/no-extension-object"
|
||||
)
|
||||
|
||||
|
||||
@pytest.mark.parametrize(
|
||||
"bucket,expected",
|
||||
[
|
||||
(("a." * 110) + "aa", True),
|
||||
("ab", False),
|
||||
("a" * 64, False),
|
||||
("ab..cd", False),
|
||||
("1.2.3.4", False),
|
||||
("192.168.0.1", False),
|
||||
("Bucket-Upper", False),
|
||||
("bucket@name", False),
|
||||
("bucket name", False),
|
||||
("-mybucket", False),
|
||||
("mybucket-", False),
|
||||
(".mybucket", False),
|
||||
("mybucket.", False),
|
||||
],
|
||||
)
|
||||
def test_is_valid_gcs_bucket_name_matrix(bucket, expected):
|
||||
from litellm.llms.vertex_ai.gemini.transformation import _is_valid_gcs_bucket_name
|
||||
|
||||
assert _is_valid_gcs_bucket_name(bucket) is expected
|
||||
|
||||
|
||||
def test_get_gcs_object_content_type_explicit_vertex_success_and_token_failure():
|
||||
from litellm.llms.vertex_ai.gemini import transformation as gt
|
||||
|
||||
mock_v = MagicMock()
|
||||
mock_v.get_access_token.return_value = ("test-token", "test-project")
|
||||
resp = MagicMock()
|
||||
resp.is_error = False
|
||||
resp.status_code = 200
|
||||
resp.json.return_value = {"contentType": "image/png"}
|
||||
http = MagicMock()
|
||||
http.get.return_value = resp
|
||||
|
||||
with (
|
||||
patch.object(gt, "_GCS_METADATA_VERTEX_BASE", mock_v),
|
||||
patch(
|
||||
"litellm.llms.vertex_ai.gemini.transformation._get_gcs_metadata_http_handler",
|
||||
return_value=http,
|
||||
),
|
||||
):
|
||||
assert (
|
||||
gt._get_gcs_object_content_type(
|
||||
image_url="gs://my-bucket/path/to/image-without-extension",
|
||||
vertex_project="project-123",
|
||||
vertex_credentials="credential-json",
|
||||
)
|
||||
== "image/png"
|
||||
)
|
||||
mock_v.get_access_token.assert_called_once_with(
|
||||
credentials="credential-json",
|
||||
project_id="project-123",
|
||||
)
|
||||
|
||||
mock_v2 = MagicMock()
|
||||
mock_v2.get_access_token.side_effect = Exception("token failure")
|
||||
with patch.object(gt, "_GCS_METADATA_VERTEX_BASE", mock_v2):
|
||||
with pytest.raises(
|
||||
litellm.BadRequestError,
|
||||
match="Unable to fetch GCS metadata with provided Vertex credentials/project",
|
||||
):
|
||||
gt._get_gcs_object_content_type(
|
||||
image_url="gs://my-bucket/path/to/image-without-extension",
|
||||
vertex_project="project-123",
|
||||
vertex_credentials="credential-json",
|
||||
)
|
||||
|
||||
|
||||
def test_get_gcs_object_content_type_http_error_explicit_vs_anonymous():
|
||||
from litellm.llms.vertex_ai.gemini import transformation as gt
|
||||
|
||||
mock_v = MagicMock()
|
||||
mock_v.get_access_token.return_value = ("t", "p")
|
||||
err_resp = MagicMock()
|
||||
err_resp.is_error = True
|
||||
err_resp.status_code = 403
|
||||
err_resp.text = '{"error":{"message":"Permission denied"}}'
|
||||
http = MagicMock()
|
||||
http.get.return_value = err_resp
|
||||
|
||||
with (
|
||||
patch.object(gt, "_GCS_METADATA_VERTEX_BASE", mock_v),
|
||||
patch(
|
||||
"litellm.llms.vertex_ai.gemini.transformation._get_gcs_metadata_http_handler",
|
||||
return_value=http,
|
||||
),
|
||||
):
|
||||
with pytest.raises(litellm.BadRequestError, match="HTTP 403") as ei:
|
||||
gt._get_gcs_object_content_type(
|
||||
image_url="gs://my-bucket/path/to/obj",
|
||||
vertex_project="project-123",
|
||||
vertex_credentials="credential-json",
|
||||
)
|
||||
assert "Permission denied" in str(ei.value)
|
||||
|
||||
mock_v2 = MagicMock()
|
||||
anon_err = MagicMock()
|
||||
anon_err.is_error = True
|
||||
anon_err.status_code = 403
|
||||
anon_err.text = "Forbidden"
|
||||
http2 = MagicMock()
|
||||
http2.get.return_value = anon_err
|
||||
with (
|
||||
patch.object(gt, "_GCS_METADATA_VERTEX_BASE", mock_v2),
|
||||
patch(
|
||||
"litellm.llms.vertex_ai.gemini.transformation._get_gcs_metadata_http_handler",
|
||||
return_value=http2,
|
||||
),
|
||||
):
|
||||
assert (
|
||||
gt._get_gcs_object_content_type(image_url="gs://public-bucket/public-object")
|
||||
is None
|
||||
)
|
||||
mock_v2.get_access_token.assert_not_called()
|
||||
|
||||
|
||||
def test_get_gcs_object_content_type_anonymous_success_no_auth_header():
|
||||
from litellm.llms.vertex_ai.gemini import transformation as gt
|
||||
|
||||
mock_v = MagicMock()
|
||||
ok = MagicMock()
|
||||
ok.is_error = False
|
||||
ok.status_code = 200
|
||||
ok.json.return_value = {"contentType": "image/jpeg"}
|
||||
http = MagicMock()
|
||||
http.get.return_value = ok
|
||||
|
||||
with (
|
||||
patch.object(gt, "_GCS_METADATA_VERTEX_BASE", mock_v),
|
||||
patch(
|
||||
"litellm.llms.vertex_ai.gemini.transformation._get_gcs_metadata_http_handler",
|
||||
return_value=http,
|
||||
),
|
||||
):
|
||||
assert (
|
||||
gt._get_gcs_object_content_type(image_url="gs://public-bucket/public-object")
|
||||
== "image/jpeg"
|
||||
)
|
||||
mock_v.get_access_token.assert_not_called()
|
||||
hdrs = http.get.call_args.kwargs.get("headers")
|
||||
assert hdrs is None or "Authorization" not in hdrs
|
||||
|
||||
|
||||
def test_async_transform_request_body_offloads_extensionless_gs_not_plain_text():
|
||||
from litellm.llms.vertex_ai.gemini import transformation as gemini_transformation
|
||||
|
||||
messages = [
|
||||
{
|
||||
"role": "user",
|
||||
"content": [
|
||||
{
|
||||
"type": "image_url",
|
||||
"image_url": {"url": "gs://bucket/image-without-extension"},
|
||||
}
|
||||
],
|
||||
}
|
||||
]
|
||||
|
||||
def slow_http_get(*args, **kwargs):
|
||||
time.sleep(0.5)
|
||||
response = MagicMock()
|
||||
response.is_error = False
|
||||
response.status_code = 200
|
||||
response.raise_for_status.return_value = None
|
||||
response.json.return_value = {"contentType": "image/png"}
|
||||
return response
|
||||
|
||||
async def fake_check_and_create_cache(self, **kwargs):
|
||||
return kwargs["messages"], kwargs["optional_params"], None
|
||||
|
||||
mock_v = MagicMock()
|
||||
mock_v.get_access_token.return_value = ("token", "project")
|
||||
mock_http = MagicMock()
|
||||
mock_http.get.side_effect = slow_http_get
|
||||
|
||||
async def run_scenario() -> float:
|
||||
async def concurrent_sleep() -> float:
|
||||
start = time.monotonic()
|
||||
await asyncio.sleep(0.05)
|
||||
return time.monotonic() - start
|
||||
|
||||
task = asyncio.create_task(
|
||||
gemini_transformation.async_transform_request_body(
|
||||
gemini_api_key=None,
|
||||
messages=messages,
|
||||
api_base=None,
|
||||
model="gemini-2.5-flash",
|
||||
client=None,
|
||||
timeout=None,
|
||||
extra_headers=None,
|
||||
optional_params={},
|
||||
logging_obj=MagicMock(),
|
||||
custom_llm_provider="vertex_ai",
|
||||
litellm_params={},
|
||||
vertex_project=None,
|
||||
vertex_location=None,
|
||||
vertex_auth_header=None,
|
||||
)
|
||||
)
|
||||
elapsed = await concurrent_sleep()
|
||||
await task
|
||||
return elapsed
|
||||
|
||||
with (
|
||||
patch.object(gemini_transformation, "_GCS_METADATA_VERTEX_BASE", mock_v),
|
||||
patch(
|
||||
"litellm.llms.vertex_ai.gemini.transformation._get_gcs_metadata_http_handler",
|
||||
return_value=mock_http,
|
||||
),
|
||||
patch(
|
||||
"litellm.llms.vertex_ai.context_caching.vertex_ai_context_caching."
|
||||
"ContextCachingEndpoints.async_check_and_create_cache",
|
||||
new=fake_check_and_create_cache,
|
||||
),
|
||||
):
|
||||
sleep_elapsed = asyncio.run(run_scenario())
|
||||
|
||||
assert sleep_elapsed < 0.4, (
|
||||
f"Event loop blocked for {sleep_elapsed:.3f}s; "
|
||||
"async_transform_request_body did not offload sync GCS metadata"
|
||||
)
|
||||
|
||||
async def fake_cache2(self, **kwargs):
|
||||
return kwargs["messages"], kwargs["optional_params"], None
|
||||
|
||||
async def run_plain():
|
||||
with patch(
|
||||
"litellm.llms.vertex_ai.gemini.transformation.asyncify",
|
||||
side_effect=AssertionError("asyncify must not run without extensionless gs://"),
|
||||
):
|
||||
return await gemini_transformation.async_transform_request_body(
|
||||
gemini_api_key=None,
|
||||
messages=[{"role": "user", "content": "hello"}],
|
||||
api_base=None,
|
||||
model="gemini-2.5-flash",
|
||||
client=None,
|
||||
timeout=None,
|
||||
extra_headers=None,
|
||||
optional_params={},
|
||||
logging_obj=MagicMock(),
|
||||
custom_llm_provider="vertex_ai",
|
||||
litellm_params={},
|
||||
vertex_project=None,
|
||||
vertex_location=None,
|
||||
vertex_auth_header=None,
|
||||
)
|
||||
|
||||
with patch(
|
||||
"litellm.llms.vertex_ai.context_caching.vertex_ai_context_caching."
|
||||
"ContextCachingEndpoints.async_check_and_create_cache",
|
||||
new=fake_cache2,
|
||||
):
|
||||
body = asyncio.run(run_plain())
|
||||
assert body is not None and "contents" in body
|
||||
|
||||
|
||||
@pytest.mark.parametrize(
|
||||
"messages,expected",
|
||||
[
|
||||
([{"role": "user", "content": "hello"}], False),
|
||||
(
|
||||
[
|
||||
{
|
||||
"role": "user",
|
||||
"content": [
|
||||
{
|
||||
"type": "image_url",
|
||||
"image_url": {"url": "gs://bucket/image-without-extension"},
|
||||
}
|
||||
],
|
||||
}
|
||||
],
|
||||
True,
|
||||
),
|
||||
(
|
||||
[
|
||||
{
|
||||
"role": "user",
|
||||
"content": [
|
||||
{
|
||||
"type": "image_url",
|
||||
"image_url": {"url": "gs://bucket/image.png"},
|
||||
}
|
||||
],
|
||||
}
|
||||
],
|
||||
False,
|
||||
),
|
||||
(
|
||||
[
|
||||
{
|
||||
"role": "user",
|
||||
"content": [
|
||||
{
|
||||
"type": "image_url",
|
||||
"image_url": {
|
||||
"url": "gs://bucket/image-without-extension",
|
||||
"mime_type": "image/png",
|
||||
},
|
||||
}
|
||||
],
|
||||
}
|
||||
],
|
||||
False,
|
||||
),
|
||||
(
|
||||
[
|
||||
{
|
||||
"role": "assistant",
|
||||
"content": [],
|
||||
"images": [
|
||||
{"image_url": {"url": "gs://bucket/gen-without-extension"}},
|
||||
],
|
||||
}
|
||||
],
|
||||
True,
|
||||
),
|
||||
(
|
||||
[
|
||||
{
|
||||
"role": "assistant",
|
||||
"content": [],
|
||||
"images": [{"image_url": {"url": "gs://bucket/gen.png"}}],
|
||||
}
|
||||
],
|
||||
False,
|
||||
),
|
||||
(
|
||||
[
|
||||
{
|
||||
"role": "assistant",
|
||||
"content": [],
|
||||
"images": [
|
||||
{
|
||||
"image_url": {
|
||||
"url": "gs://bucket/gen-no-ext",
|
||||
"mime_type": "image/png",
|
||||
},
|
||||
}
|
||||
],
|
||||
}
|
||||
],
|
||||
False,
|
||||
),
|
||||
],
|
||||
)
|
||||
def test_openai_messages_may_need_sync_gcs_metadata_fetch_matrix(messages, expected):
|
||||
from litellm.llms.vertex_ai.gemini.transformation import (
|
||||
_openai_messages_may_need_sync_gcs_metadata_fetch,
|
||||
)
|
||||
|
||||
assert _openai_messages_may_need_sync_gcs_metadata_fetch(messages) is expected
|
||||
Loading…
Add table
Reference in a new issue