fix: Do not throw error when in a uvloop context (#142)

PDF page splitting uses asyncio but the SDK is not async. Therefore, we
had to manage our own event loop, which can lead to issues in other
event loop contexts. Uvloop is one context that does not allow us to use
nested event loops. When we find ourselves in a uvloop.Loop, we have to
fallback to non splitting mode. #135 will make the whole SDK async so we
don't have to hack this.

Closes #133

Author: awalker4

Diff for: Makefile

@@ -9,7 +9,7 @@ DOCKER_IMAGE ?= downloads.unstructured.io/unstructured-io/unstructured-api:lates
 
 .PHONY: install-test
 install-test:
-	pip install pytest pytest-asyncio pytest-mock requests_mock pypdf deepdiff requests-toolbelt
+	pip install pytest pytest-asyncio pytest-mock requests_mock pypdf deepdiff requests-toolbelt uvloop
 
 .PHONY: install-dev
 install-dev:

Diff for: _test_unstructured_client/integration/test_integration_freemium.py

@@ -91,3 +91,32 @@ def test_partition_handling_server_error(error, split_pdf, monkeypatch, doc_path
 
     with pytest.raises(sdk_raises):
         response = client.general.partition(req)
+
+
+def test_uvloop_partitions_without_errors(client, doc_path):
+    async def call_api():
+        filename = "layout-parser-paper-fast.pdf"
+        with open(doc_path / filename, "rb") as f:
+            files = shared.Files(
+                content=f.read(),
+                file_name=filename,
+            )
+
+        req = shared.PartitionParameters(
+            files=files,
+            strategy="fast",
+            languages=["eng"],
+            split_pdf_page=True,
+        )
+
+        resp = client.general.partition(req)
+
+        if resp is not None:
+            return resp.elements
+        else:
+            return []
+
+    import uvloop
+    uvloop.install()
+    elements = asyncio.run(call_api())
+    assert len(elements) > 0

Diff for: src/unstructured_client/_hooks/custom/split_pdf_hook.py

@@ -73,6 +73,15 @@ async def run_tasks(coroutines: list[Awaitable], allow_failed: bool = False) ->
     return sorted(results, key=lambda x: x[0])
 
 
+def context_is_uvloop():
+    """Return true if uvloop is installed and we're currently in a uvloop context. Our asyncio splitting code currently doesn't work under uvloop."""
+    try:
+        import uvloop  # pylint: disable=import-outside-toplevel
+        loop = asyncio.get_event_loop()
+        return isinstance(loop, uvloop.Loop)
+    except ImportError:
+        return False
+
 def get_optimal_split_size(num_pages: int, concurrency_level: int) -> int:
     """Distributes pages to workers evenly based on the number of pages and desired concurrency level."""
     if num_pages < MAX_PAGES_PER_SPLIT * concurrency_level:
@@ -94,10 +103,6 @@ class SplitPdfHook(SDKInitHook, BeforeRequestHook, AfterSuccessHook, AfterErrorH
     """
 
     def __init__(self) -> None:
-        # This allows us to use an event loop in an env with an existing loop
-        # Temporary fix until we can improve the async splitting behavior
-        nest_asyncio.apply()
-
         self.client: Optional[requests.Session] = None
         self.coroutines_to_execute: dict[
             str, list[Coroutine[Any, Any, requests.Response]]
@@ -121,6 +126,8 @@ def sdk_init(
         self.client = client
         return base_url, client
 
+
+    # pylint: disable=too-many-return-statements
     def before_request(
             self, hook_ctx: BeforeRequestContext, request: requests.PreparedRequest
     ) -> Union[requests.PreparedRequest, Exception]:
@@ -143,6 +150,13 @@ def before_request(
             logger.warning("HTTP client not accessible! Continuing without splitting.")
             return request
 
+        if context_is_uvloop():
+            logger.warning("Splitting is currently incompatible with uvloop. Continuing without splitting.")
+            return request
+
+        # This allows us to use an event loop in an env with an existing loop
+        # Temporary fix until we can improve the async splitting behavior
+        nest_asyncio.apply()
         operation_id = hook_ctx.operation_id
         content_type = request.headers.get("Content-Type")
         body = request.body