test(e2e): cover /chat/completions on Azure OpenAI, streaming and non-streaming

tests/e2e/coverage_registry/llm_conversational.yaml

@@ -29,6 +29,7 @@
 - {id: llm.chat_completions.vertex.vision.nonstream.works, module: llm, tier: P0, subject_endpoint: chat_completions, route: vertex, capability: vision, streaming: nonstream, assertions: [works], source: "model_prices json", rationale: "Gemini vision"}
 - {id: llm.chat_completions.vertex.prompt_cache_5m.nonstream.works, module: llm, tier: P1, subject_endpoint: chat_completions, route: vertex, capability: prompt_cache_5m, streaming: nonstream, assertions: [works], source: "model_prices json", rationale: "Vertex Gemini prompt caching"}
 - {id: llm.chat_completions.azure_openai.basic.nonstream.works, module: llm, tier: P0, subject_endpoint: chat_completions, route: azure_openai, capability: basic, streaming: nonstream, assertions: [works], source: "proxy_server.py:8455", rationale: "P0 route; Azure OpenAI deployments"}
+- {id: llm.chat_completions.azure_openai.basic.stream.works, module: llm, tier: P0, subject_endpoint: chat_completions, route: azure_openai, capability: basic, streaming: stream, assertions: [works], source: "proxy_server.py:8455", rationale: "Streaming over Azure OpenAI"}
 - {id: llm.chat_completions.azure_openai.tool_use.nonstream.works, module: llm, tier: P0, subject_endpoint: chat_completions, route: azure_openai, capability: tool_use, streaming: nonstream, assertions: [works], source: "model_prices json", rationale: "Azure OpenAI function_calling"}
 - {id: llm.chat_completions.azure_foundry.basic.nonstream.works, module: llm, tier: P1, subject_endpoint: chat_completions, route: azure_foundry, capability: basic, streaming: nonstream, assertions: [works], source: "proxy_server.py:8455", rationale: "Azure Foundry (azure_ai); newer, smoke"}
 - {id: llm.messages.anthropic.basic.nonstream.works, module: llm, tier: P0, subject_endpoint: messages, route: anthropic, capability: basic, streaming: nonstream, assertions: [works], source: "anthropic_endpoints/endpoints.py:64", rationale: "Core endpoint; Anthropic Messages native"}

tests/e2e/docker-compose.yml

@@ -66,6 +66,12 @@ configs:
             model: openai/text-embedding-3-small
             api_key: os.environ/OPENAI_API_KEY
 
+        - model_name: azure-gpt-5.4-mini
+          litellm_params:
+            model: azure/gpt-5.4-mini
+            api_base: os.environ/AZURE_API_BASE
+            api_key: os.environ/AZURE_API_KEY
+
 services:
   litellm:
     image: ghcr.io/berriai/litellm:main-latest

tests/e2e/e2e_http.py

@@ -121,6 +121,7 @@ class StreamingResponse(BaseModel):
     headers: dict[str, str] = {}
     body: str
     chunks: int = 0  # streamed events (0 for non-streaming)
+    events: tuple[str, ...] = ()  # SSE "data:" payloads, prefix stripped (streaming only)
     # First in-stream error event, if any. A streamed call commits its HTTP 200
     # before the upstream completes, so upstream failures (e.g. insufficient
     # quota) arrive as SSE error events inside an otherwise-successful response;
@@ -293,28 +294,35 @@ def _streaming_outcome(resp: requests.Response, stream: bool) -> StreamingRespon
             headers=headers,
             body=resp.text,
         )
-    lines = cast("Iterator[bytes]", resp.iter_lines())
-    chunks = 0
-    stream_error: str | None = None
-    for line in lines:
-        if not line:
-            continue
-        chunks += 1
-        if stream_error is None and (
-            line.startswith(b"event: error")
-            or b'"type":"error"' in line
-            or b'"type": "error"' in line
-            or line.startswith(b'data: {"error"')
-        ):
-            stream_error = line.decode(errors="replace")[:300]
+    raw_lines = tuple(
+        line.decode(errors="replace")
+        for line in cast("Iterator[bytes]", resp.iter_lines())
+        if line
+    )
+    stream_error = next(
+        (
+            line[:300]
+            for line in raw_lines
+            if line.startswith("event: error")
+            or '"type":"error"' in line
+            or '"type": "error"' in line
+            or line.startswith('data: {"error"')
+        ),
+        None,
+    )
     return StreamingResponse(
         status_code=resp.status_code,
         call_id=call_id,
         response_cost=response_cost,
         content_type=content_type,
         headers=headers,
         body="<streamed>",
-        chunks=chunks,
+        chunks=len(raw_lines),
+        events=tuple(
+            line.removeprefix("data:").strip()
+            for line in raw_lines
+            if line.startswith("data:")
+        ),
         stream_error=stream_error,
     )
 

tests/e2e/llm_translation/test_chat_completions_regression_e2e.py

@@ -4,9 +4,11 @@
 releases: a clean 200 came back but with no real completion. A status check
 alone would not have caught it, so each case here asserts the product promise -
 a non-empty assistant message and a real model name in the body - across the
-three providers wired into the gateway config (OpenAI, Anthropic, Gemini). A
-regression that empties the completion for any provider fails that provider's
-row here.
+providers wired into the gateway config (OpenAI, Anthropic, Gemini, Azure
+OpenAI). A regression that empties the completion for any provider fails that
+provider's row here. The Azure OpenAI streaming case applies the same standard
+to the SSE path: every data event must parse as a chat.completion.chunk and the
+deltas must reassemble into real text, not just count as a 200 with chunks.
 """
 
 from __future__ import annotations
@@ -15,15 +17,18 @@
 
 from e2e_config import unique_marker
 from e2e_http import unwrap
-from models import ChatBody, ChatMessage
+from models import ChatBody, ChatMessage, ChatStreamChunk
 from passthrough_client import PassthroughClient
 
 pytestmark = pytest.mark.e2e
 
+AZURE_CHAT_MODEL = "azure-gpt-5.4-mini"
+
 CHAT_MODELS: tuple[tuple[str, str], ...] = (
     ("gpt-5.5", "openai"),
     ("claude-haiku-4-5", "anthropic"),
     ("gemini-2.5-flash", "gemini"),
+    (AZURE_CHAT_MODEL, "azure_openai"),
 )
 
 
@@ -37,6 +42,7 @@ class TestChatCompletionsRegression:
         "llm.chat_completions.openai.basic.nonstream.works",
         "llm.chat_completions.anthropic.basic.nonstream.works",
         "llm.chat_completions.vertex.basic.nonstream.works",
+        "llm.chat_completions.azure_openai.basic.nonstream.works",
         exercised_on=[],
     )
     def test_chat_returns_real_completion(
@@ -68,3 +74,62 @@ def test_chat_returns_real_completion(
         assert (
             message is not None and message.content and message.content.strip()
         ), f"{model} ({route}): 200 with an empty completion (#28991): {response}"
+
+    @pytest.mark.covers("llm.chat_completions.azure_openai.basic.stream.works")
+    def test_azure_openai_stream_returns_real_completion(
+        self, client: PassthroughClient, scoped_key: str
+    ) -> None:
+        result = client.gateway.chat_stream(
+            scoped_key,
+            ChatBody(
+                model=AZURE_CHAT_MODEL,
+                messages=[
+                    ChatMessage(
+                        role="user",
+                        content=f"reply with one word {unique_marker()}",
+                    )
+                ],
+                max_tokens=512,
+                stream=True,
+            ),
+        )
+
+        assert result.ok, (
+            f"{AZURE_CHAT_MODEL}: stream failed with status "
+            f"{result.status_code}: {result.body[:300]}"
+        )
+        assert result.is_streaming, (
+            f"{AZURE_CHAT_MODEL}: expected text/event-stream, got "
+            f"{result.content_type}: {result.body[:300]}"
+        )
+        assert result.stream_error is None, (
+            f"{AZURE_CHAT_MODEL}: 200 stream carried an error event: "
+            f"{result.stream_error}"
+        )
+        assert result.events, f"{AZURE_CHAT_MODEL}: stream carried no SSE data events"
+        assert result.events[-1] == "[DONE]", (
+            f"{AZURE_CHAT_MODEL}: stream did not terminate with [DONE]: "
+            f"{result.events[-1][:200]}"
+        )
+
+        chunks = [
+            ChatStreamChunk.model_validate_json(event) for event in result.events[:-1]
+        ]
+        assert chunks, f"{AZURE_CHAT_MODEL}: stream held only the [DONE] sentinel"
+        assert all(
+            chunk.object == "chat.completion.chunk" for chunk in chunks
+        ), f"{AZURE_CHAT_MODEL}: malformed chunk object types: {result.events[:5]}"
+        assert any(
+            chunk.model for chunk in chunks
+        ), f"{AZURE_CHAT_MODEL}: no chunk carried a model name: {result.events[:5]}"
+
+        content = "".join(
+            choice.delta.content or ""
+            for chunk in chunks
+            for choice in chunk.choices
+            if choice.delta is not None
+        )
+        assert content.strip(), (
+            f"{AZURE_CHAT_MODEL}: stream chunks reassembled to an empty "
+            f"completion (#28991): {result.events[:5]}"
+        )

tests/e2e/models.py

@@ -191,6 +191,21 @@ class ChatResponse(BaseModel):
     service_tier: str | None = None
 
 
+class ChatStreamDelta(BaseModel):
+    content: str | None = None
+
+
+class ChatStreamChoice(BaseModel):
+    delta: ChatStreamDelta | None = None
+
+
+class ChatStreamChunk(BaseModel):
+    id: str | None = None
+    object: str | None = None
+    model: str | None = None
+    choices: list[ChatStreamChoice] = []
+
+
 class EmbedBody(BaseModel):
     model: str
     input: str