Merge branch 'main' into feat/2941-elasticsearch-sparse-embedding-retriever

Author: davidsbatista

integrations/ollama/CHANGELOG.md

@@ -1,5 +1,23 @@
 # Changelog
 
+## [integrations/ollama-v6.3.0] - 2026-04-10
+
+### 🐛 Bug Fixes
+
+- Replace in-place dataclass mutations with dataclasses.replace() (#3112)
+- Ollama - better reasoning streaming support (#3131)
+
+### 🧪 Testing
+
+- Track test coverage for all integrations (#3065)
+
+### 🧹 Chores
+
+- Add ANN ruff ruleset to llama_cpp, llama_stack, mcp, meta_llama, mistral, mongodb_atlas, nvidia, ollama, openrouter, opensearch (#2991)
+- Enforce ruff docstring rules (D102/D103/D205/D209/D213/D417/D419) in integrations 21-30 (#3010)
+- Increase lower pins for 3.14 support in some integrations + test with 3.14 (#3033)
+
+
 ## [integrations/ollama-v6.2.0] - 2026-03-09
 
 ### 🚀 Features

integrations/ollama/src/haystack_integrations/components/generators/ollama/chat/chat_generator.py

@@ -376,14 +376,25 @@ def _handle_streaming_response(
         id_order: list[str] = []
         tool_call_index: int = 0
 
+        # track reasoning and content blocks to correctly set start=True on the first chunk of each block
+        reasoning_started = False
+        content_started = False
+
         # Stream
         for index, raw in enumerate(response_iter):
             if raw.message.tool_calls:
                 tool_call_index += 1
             chunk = _build_chunk(
                 chunk_response=raw, component_info=component_info, index=index, tool_call_index=tool_call_index
             )
-            start = index == 0 or bool(chunk.tool_calls)
+
+            if chunk.reasoning:
+                start = not reasoning_started or bool(chunk.tool_calls)
+                reasoning_started = True
+            else:
+                start = (not content_started) or bool(chunk.tool_calls)
+                content_started = True
+
             chunk = replace(chunk, start=start)
             chunks.append(chunk)
 
@@ -458,6 +469,10 @@ async def _handle_streaming_response_async(
         id_order: list[str] = []
         tool_call_index: int = 0
 
+        # track reasoning and content blocks to correctly set start=True on the first chunk of each block
+        reasoning_started = False
+        content_started = False
+
         # Stream
         index = 0
         async for raw in response_iter:
@@ -466,7 +481,14 @@ async def _handle_streaming_response_async(
             chunk = _build_chunk(
                 chunk_response=raw, component_info=component_info, index=index, tool_call_index=tool_call_index
             )
-            start = index == 0 or bool(chunk.tool_calls)
+
+            if chunk.reasoning:
+                start = not reasoning_started or bool(chunk.tool_calls)
+                reasoning_started = True
+            else:
+                start = (not content_started) or bool(chunk.tool_calls)
+                content_started = True
+
             chunk = replace(chunk, start=start)
             chunks.append(chunk)
 

integrations/ollama/tests/test_chat_generator.py

@@ -267,6 +267,70 @@ def test_callback(chunk: StreamingChunk):
         assert streaming_chunks[1].start is False
         assert streaming_chunks[2].start is False
 
+    def test_handle_streaming_response_with_thinking(self):
+        ollama_chunks = [
+            ChatResponse(
+                model="qwen3:0.6b",
+                created_at="2025-07-31T15:27:09.265818Z",
+                done=False,
+                message=Message(role="assistant", content="", thinking="Let me think"),
+            ),
+            ChatResponse(
+                model="qwen3:0.6b",
+                created_at="2025-07-31T15:27:09.265818Z",
+                done=False,
+                message=Message(role="assistant", content="", thinking=" about this."),
+            ),
+            ChatResponse(
+                model="qwen3:0.6b",
+                created_at="2025-07-31T15:27:09.265818Z",
+                done=False,
+                message=Message(role="assistant", content="The capital"),
+            ),
+            ChatResponse(
+                model="qwen3:0.6b",
+                created_at="2025-07-31T15:27:09.265818Z",
+                done=False,
+                message=Message(role="assistant", content=" is Amman."),
+            ),
+            ChatResponse(
+                model="qwen3:0.6b",
+                created_at="2025-07-31T15:27:09.355211Z",
+                done=True,
+                done_reason="stop",
+                total_duration=1303416458,
+                load_duration=953922333,
+                prompt_eval_count=22,
+                prompt_eval_duration=254166208,
+                eval_count=3,
+                eval_duration=92965792,
+                message=Message(role="assistant", content=""),
+            ),
+        ]
+
+        generator = OllamaChatGenerator()
+        streaming_chunks = []
+
+        def test_callback(chunk: StreamingChunk):
+            streaming_chunks.append(chunk)
+
+        response = generator._handle_streaming_response(ollama_chunks, test_callback)
+        assert response["replies"][0].text == "The capital is Amman."
+        assert response["replies"][0].reasoning.reasoning_text == "Let me think about this."
+
+        assert len(streaming_chunks) == 5
+        # First reasoning chunk: start=True
+        assert streaming_chunks[0].start is True
+        assert streaming_chunks[0].reasoning.reasoning_text == "Let me think"
+        # Second reasoning chunk: start=False
+        assert streaming_chunks[1].start is False
+        # First content chunk after reasoning: start=True
+        assert streaming_chunks[2].start is True
+        assert streaming_chunks[2].content == "The capital"
+        # Remaining content chunks: start=False
+        assert streaming_chunks[3].start is False
+        assert streaming_chunks[4].start is False
+
     def test_handle_streaming_response_tool_calls(self):
         ollama_chunks = [
             ChatResponse(

integrations/opensearch/CHANGELOG.md

@@ -1,5 +1,20 @@
 # Changelog
 
+## [integrations/opensearch-v7.2.0] - 2026-04-10
+
+### 🚀 Features
+
+- *(opensearch)* Add support for nested fields in OpenSearchDocumentStore (#3117)
+
+### 🐛 Bug Fixes
+
+- Replace in-place dataclass mutations in document stores (#3114)
+
+### 🧪 Testing
+
+- Better categorize some Document Stores tests (#3085)
+
+
 ## [integrations/opensearch-v7.1.0] - 2026-04-01
 
 ### 🚀 Features