sunholo 0.140.6__py3-none-any.whl → 0.140.8__py3-none-any.whl

sunholo/agents/flask/base.py

@@ -11,12 +11,38 @@
 #   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 #   See the License for the specific language governing permissions and
 #   limitations under the License.
+
+from ...custom_logging import log
+import time
+
 def create_app(name):
-    from flask import Flask
+    from flask import Flask, request
     
     app = Flask(name)
 
     app.config['TRAP_HTTP_EXCEPTIONS'] = True
     app.config['PROPAGATE_EXCEPTIONS'] = True
    
+    @app.before_request
+    def start_timer():
+        request.start_time = time.time()
+
+    @app.after_request  
+    def log_timing(response):
+        if hasattr(request, 'start_time'):
+            duration = time.time() - request.start_time
+            
+            # Log all VAC requests with different detail levels
+            if request.path.startswith('/vac/streaming/'):
+                log.info(f"🚀 STREAMING: {duration:.3f}s - {request.path}")
+            elif request.path.startswith('/vac/'):
+                log.info(f"⚡ VAC: {duration:.3f}s - {request.path}")
+            elif duration > 1.0:  # Log any slow requests
+                log.warning(f"🐌 SLOW REQUEST: {duration:.3f}s - {request.path}")
+            
+            # Add response headers with timing info for debugging
+            response.headers['X-Response-Time'] = f"{duration:.3f}s"
+            
+        return response
+    
     return app

sunholo/agents/flask/vac_routes.py

@@ -6,6 +6,10 @@ import random
 from functools import partial
 import inspect
 import asyncio
+import time
+import threading
+from functools import lru_cache
+from concurrent.futures import ThreadPoolExecutor
 
 from ..chat_history import extract_chat_history_with_cache, extract_chat_history_async_cached
 from ...qna.parsers import parse_output
@@ -32,6 +36,11 @@ except ImportError:
 # Cache dictionary to store validated API keys
 api_key_cache = {}
 cache_duration = timedelta(minutes=5)  # Cache duration
+# Global caches and thread pool 
+_config_cache = {}
+_config_lock = threading.Lock()
+_thread_pool = ThreadPoolExecutor(max_workers=4)
+
 
 class VACRoutes:
     """
@@ -69,8 +78,44 @@ if __name__ == "__main__":
         self.additional_routes = additional_routes if additional_routes is not None else []
         self.async_stream = async_stream
         self.add_langfuse_eval = add_langfuse_eval
+
+        # Pre-warm common configs
+        self._preload_common_configs()   
+
         self.register_routes()
+
+    def _preload_common_configs(self):
+        """Pre-load commonly used configurations to cache"""
+        common_vector_names = ["aitana3"]  # Add your common vector names 
+        for vector_name in common_vector_names:
+            try:
+                self._get_cached_config(vector_name)
+                log.info(f"Pre-loaded config for {vector_name}")
+            except Exception as e:
+                log.warning(f"Failed to pre-load config for {vector_name}: {e}")
+
+    def _get_cached_config(self, vector_name: str):
+        """Cached config loader with thread safety - CORRECTED VERSION"""
+        # Check cache first (without lock for read)
+        if vector_name in _config_cache:
+            log.debug(f"Using cached config for {vector_name}")
+            return _config_cache[vector_name]
         
+        # Need to load config
+        with _config_lock:
+            # Double-check inside lock (another thread might have loaded it)
+            if vector_name in _config_cache:
+                return _config_cache[vector_name]
+            
+            try:
+                log.info(f"Loading fresh config for {vector_name}")
+                config = ConfigManager(vector_name)
+                _config_cache[vector_name] = config
+                log.info(f"Cached config for {vector_name}")
+                return config
+            except Exception as e:
+                log.error(f"Error loading config for {vector_name}: {e}")
+                raise
 
     def vac_interpreter_default(self, question: str, vector_name: str, chat_history=[], **kwargs):
         # Create a callback that does nothing for streaming if you don't want intermediate outputs
@@ -228,22 +273,43 @@ if __name__ == "__main__":
 
         log.info(f"OpenAI response: {openai_response}")
         return jsonify(openai_response)
-        
+
+    def _finalize_trace_background(self, trace, span, response, all_input):
+        """Finalize trace operations in background"""
+        try:
+            if span:
+                span.end(output=str(response))
+            if trace:
+                trace.update(output=str(response))
+                self.langfuse_eval_response(trace_id=trace.id, eval_percent=all_input.get('eval_percent'))
+        except Exception as e:
+            log.warning(f"Background trace finalization failed: {e}")
+
     def handle_stream_vac(self, vector_name):
+        request_start = time.time()
         observed_stream_interpreter = self.stream_interpreter
         is_async = inspect.iscoroutinefunction(self.stream_interpreter)
 
         if is_async:
             log.info(f"Stream interpreter is async: {observed_stream_interpreter}")
 
-        prep = self.prep_vac(request, vector_name)
-        log.info(f"Processing prep: {prep}")
-        trace = prep["trace"]
-        span = prep["span"]
+        # Call prep_vac and handle errors properly
+        try:
+            prep = self.prep_vac(request, vector_name)
+        except Exception as e:
+            log.error(f"prep_vac failed: {e}")
+            error_response = {'error': f'Prep error: {str(e)}'}
+            return jsonify(error_response), 500
+        
+        log.info(f"Processing prep completed in {time.time() - request_start:.3f}s")
+        
+        trace = prep.get("trace")
+        span = prep.get("span")
         vac_config = prep["vac_config"]
         all_input = prep["all_input"]
 
-        log.info(f'Streaming data with: {all_input}')
+        log.info(f'Starting stream with: {all_input["user_input"][:100]}...')
+        
         if span:
             span.update(
                 name="start_streaming_chat",
@@ -254,7 +320,7 @@ if __name__ == "__main__":
         def generate_response_content():
             try:
                 if is_async:
-                    from queue import Queue, Empty
+                    from queue import Queue
                     result_queue = Queue()
                     import threading
 
@@ -271,7 +337,7 @@ if __name__ == "__main__":
                                     trace_id=trace.id if trace else None,
                                     **all_input["kwargs"]
                                 )
-                                log.info(f"{async_gen=}")
+                                
                                 async for chunk in async_gen:
                                     if isinstance(chunk, dict) and 'answer' in chunk:
                                         if trace:
@@ -284,9 +350,12 @@ if __name__ == "__main__":
                                     else:
                                         result_queue.put(chunk)
                             except Exception as e:
-                                result_queue.put(f"Streaming Error: {str(e)} {traceback.format_exc()}")
+                                error_msg = f"Streaming Error: {str(e)} {traceback.format_exc()}"
+                                log.error(error_msg)
+                                result_queue.put(error_msg)
                             finally:
                                 result_queue.put(None)  # Sentinel
+                        
                         asyncio.run(process_async())
 
                     thread = threading.Thread(target=run_async)
@@ -301,7 +370,7 @@ if __name__ == "__main__":
 
                     thread.join()
                 else:
-                    log.info("sync streaming response")
+                    log.info("Starting sync streaming response")
                     for chunk in start_streaming_chat(
                         question=all_input["user_input"],
                         vector_name=vector_name,
@@ -325,17 +394,19 @@ if __name__ == "__main__":
                             yield chunk
 
             except Exception as e:
-                yield f"Streaming Error: {str(e)} {traceback.format_exc()}"
+                error_msg = f"Streaming Error: {str(e)} {traceback.format_exc()}"
+                log.error(error_msg)
+                yield error_msg
 
-        # Here, the generator function will handle streaming the content to the client.
+        # Create streaming response
         response = Response(generate_response_content(), content_type='text/plain; charset=utf-8')
         response.headers['Transfer-Encoding'] = 'chunked'
 
-        log.debug(f"streaming response: {response}")
+        log.info(f"Streaming response created in {time.time() - request_start:.3f}s")
+        
+        # Do final trace operations in background (don't block the response)
         if trace:
-            span.end(output=response)
-            trace.update(output=response)
-            self.langfuse_eval_response(trace_id=trace.id, eval_percent=all_input.get('eval_percent'))
+            _thread_pool.submit(self._finalize_trace_background, trace, span, response, all_input)
 
         return response
 
@@ -654,78 +725,127 @@ if __name__ == "__main__":
             tags = tags,
             release = package_version
         )
+    
+    def _create_langfuse_trace_background(self, request, vector_name, trace_id):
+        """Create Langfuse trace in background"""
+        try:
+            return self.create_langfuse_trace(request, vector_name, trace_id)
+        except Exception as e:
+            log.warning(f"Background trace creation failed: {e}")
+            return None
 
+    def _handle_file_upload_background(self, file, vector_name):
+        """Handle file upload in background thread"""
+        try:
+            # Save with timestamp to avoid conflicts
+            temp_filename = f"temp_{int(time.time() * 1000)}_{file.filename}"
+            file.save(temp_filename)
+            
+            # Upload to GCS
+            image_uri = add_file_to_gcs(temp_filename, vector_name)
+            
+            # Clean up
+            os.remove(temp_filename)
+            
+            return {"image_uri": image_uri, "mime": file.mimetype}
+        except Exception as e:
+            log.error(f"Background file upload failed: {e}")
+            return {}
+    
     def prep_vac(self, request, vector_name):
-
+        start_time = time.time()
+        
+        # Fast request parsing - KEEP ORIGINAL ERROR HANDLING STYLE
         if request.content_type.startswith('application/json'):
             data = request.get_json()
         elif request.content_type.startswith('multipart/form-data'):
             data = request.form.to_dict()
+            # Handle file upload in background if present
             if 'file' in request.files:
                 file = request.files['file']
                 if file.filename != '':
-                    log.info(f"Found file: {file.filename} to upload to GCS")
-                    try:
-                        image_uri, mime_type = self.handle_file_upload(file, vector_name)
-                        data["image_uri"] = image_uri
-                        data["mime"] = mime_type
-                    except Exception as e:
-                        log.error(traceback.format_exc())
-                        return jsonify({'error': str(e), 'traceback': traceback.format_exc()}), 500
-                else:
-                    log.error("No file selected")
-                    return jsonify({"error": "No file selected"}), 400
+                    log.info(f"Found file: {file.filename} - uploading in background")
+                    # Start file upload in background, don't block
+                    upload_future = _thread_pool.submit(self._handle_file_upload_background, file, vector_name)
+                    data["_upload_future"] = upload_future
         else:
-            return jsonify({"error": "Unsupported content type"}), 400
+            # KEEP ORIGINAL STYLE - return the error response directly
+            raise ValueError("Unsupported content type")
 
-        log.info(f"vac/{vector_name} got data: {data}")
+        log.info(f"vac/{vector_name} got data keys: {list(data.keys())}")
 
-        trace = None
-        span = None
-        if self.add_langfuse_eval:
-            trace_id = data.get('trace_id')
-            trace = self.create_langfuse_trace(request, vector_name, trace_id)
-            log.info(f"Using existing langfuse trace: {trace_id}")
-        
-        #config, _ = load_config("config/llm_config.yaml")
+        # Get config from cache first (before processing other data)
         try:
-            vac_config = ConfigManager(vector_name)
+            vac_config = self._get_cached_config(vector_name)
         except Exception as e:
             raise ValueError(f"Unable to find vac_config for {vector_name} - {str(e)}")
 
-        if trace:
-            this_vac_config = vac_config.configs_by_kind.get("vacConfig")
-            metadata_config=None
-            if this_vac_config:
-                metadata_config = this_vac_config.get(vector_name)
-
-            trace.update(input=data, metadata=metadata_config)
+        # Initialize trace variables
+        trace = None
+        span = None
+        if self.add_langfuse_eval:
+            trace_id = data.get('trace_id')
+            # Create trace in background - don't block
+            trace_future = _thread_pool.submit(self._create_langfuse_trace_background, request, vector_name, trace_id)
 
+        # Extract data (keep original logic)
         user_input = data.pop('user_input').strip()
         stream_wait_time = data.pop('stream_wait_time', 7)
         stream_timeout = data.pop('stream_timeout', 120)
         chat_history = data.pop('chat_history', None)
         eval_percent = data.pop('eval_percent', 0.01)
-        vector_name = data.pop('vector_name', vector_name)
-        data.pop('trace_id', None) # to ensure not in kwargs
+        vector_name_param = data.pop('vector_name', vector_name)
+        data.pop('trace_id', None)  # to ensure not in kwargs
 
+        # Process chat history with caching
         paired_messages = extract_chat_history_with_cache(chat_history)
 
-        all_input = {'user_input': user_input, 
-                     'vector_name': vector_name, 
-                     'chat_history': paired_messages, 
-                     'stream_wait_time': stream_wait_time,
-                     'stream_timeout': stream_timeout,
-                     'eval_percent': eval_percent,
-                     'kwargs': data}
+        # Wait for file upload if it was started (with timeout)
+        if "_upload_future" in data:
+            try:
+                upload_result = data["_upload_future"].result(timeout=3.0)  # 3 sec max wait
+                data.update(upload_result)
+                log.info(f"File upload completed: {upload_result.get('image_uri', 'no uri')}")
+            except Exception as e:
+                log.warning(f"File upload failed or timed out: {e}")
+            finally:
+                data.pop("_upload_future", None)
+
+        # Build final input
+        all_input = {
+            'user_input': user_input, 
+            'vector_name': vector_name_param, 
+            'chat_history': paired_messages, 
+            'stream_wait_time': stream_wait_time,
+            'stream_timeout': stream_timeout,
+            'eval_percent': eval_percent,
+            'kwargs': data
+        }
+
+        # Try to get trace result if available (don't block long)
+        if self.add_langfuse_eval:
+            try:
+                trace = trace_future.result(timeout=0.1)  # Very short timeout
+                if trace:
+                    this_vac_config = vac_config.configs_by_kind.get("vacConfig")
+                    metadata_config = None
+                    if this_vac_config:
+                        metadata_config = this_vac_config.get(vector_name)
+                    trace.update(input=data, metadata=metadata_config)
+                    
+                    span = trace.span(
+                        name="VAC",
+                        metadata=vac_config.configs_by_kind,
+                        input=all_input
+                    )
+            except Exception as e:
+                log.warning(f"Langfuse trace creation timed out or failed: {e}")
+                trace = None
+                span = None
+
+        prep_time = time.time() - start_time
+        log.info(f"prep_vac completed in {prep_time:.3f}s")
 
-        if trace:
-            span = trace.span(
-                name="VAC",
-                metadata=vac_config.configs_by_kind,
-                input = all_input
-            )
-    
         return {
             "trace": trace,
             "span": span,

{sunholo-0.140.6.dist-info → sunholo-0.140.8.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: sunholo
-Version: 0.140.6
+Version: 0.140.8
 Summary: AI DevOps - a package to help deploy GenAI to the Cloud.
 Author-email: Holosun ApS <multivac@sunholo.com>
 License: Apache License, Version 2.0

{sunholo-0.140.6.dist-info → sunholo-0.140.8.dist-info}/RECORD

@@ -13,8 +13,8 @@ sunholo/agents/fastapi/__init__.py,sha256=S_pj4_bTUmDGoq_exaREHlOKThi0zTuGT0VZY0
 sunholo/agents/fastapi/base.py,sha256=W-cyF8ZDUH40rc-c-Apw3-_8IIi2e4Y9qRtnoVnsc1Q,2521
 sunholo/agents/fastapi/qna_routes.py,sha256=lKHkXPmwltu9EH3RMwmD153-J6pE7kWQ4BhBlV3to-s,3864
 sunholo/agents/flask/__init__.py,sha256=dEoByI3gDNUOjpX1uVKP7uPjhfFHJubbiaAv3xLopnk,63
-sunholo/agents/flask/base.py,sha256=HLz3Z5efWaewTwSFEM6JH48NA9otoJBoVFJlARGk9L8,788
-sunholo/agents/flask/vac_routes.py,sha256=al4-k-QNKH5bX9Ai8FP7DC1R7yomSO3Lnq_cugnUHcw,32622
+sunholo/agents/flask/base.py,sha256=vnpxFEOnCmt9humqj-jYPLfJcdwzsop9NorgkJ-tSaU,1756
+sunholo/agents/flask/vac_routes.py,sha256=YOW64HaRYa0MfMnzwbx2s9IrU6lz-CeqpcfmIo_L3ho,37664
 sunholo/archive/__init__.py,sha256=qNHWm5rGPVOlxZBZCpA1wTYPbalizRT7f8X4rs2t290,31
 sunholo/archive/archive.py,sha256=PxVfDtO2_2ZEEbnhXSCbXLdeoHoQVImo4y3Jr2XkCFY,1204
 sunholo/auth/__init__.py,sha256=TeP-OY0XGxYV_8AQcVGoh35bvyWhNUcMRfhuD5l44Sk,91
@@ -168,9 +168,9 @@ sunholo/vertex/init.py,sha256=1OQwcPBKZYBTDPdyU7IM4X4OmiXLdsNV30C-fee2scQ,2875
 sunholo/vertex/memory_tools.py,sha256=tBZxqVZ4InTmdBvLlOYwoSEWu4-kGquc-gxDwZCC4FA,7667
 sunholo/vertex/safety.py,sha256=S9PgQT1O_BQAkcqauWncRJaydiP8Q_Jzmu9gxYfy1VA,2482
 sunholo/vertex/type_dict_to_json.py,sha256=uTzL4o9tJRao4u-gJOFcACgWGkBOtqACmb6ihvCErL8,4694
-sunholo-0.140.6.dist-info/licenses/LICENSE.txt,sha256=SdE3QjnD3GEmqqg9EX3TM9f7WmtOzqS1KJve8rhbYmU,11345
-sunholo-0.140.6.dist-info/METADATA,sha256=J62v0HZ3NpRqt-zt0jpcA-KgGXyb5aEQyPGt6D4W-B8,10067
-sunholo-0.140.6.dist-info/WHEEL,sha256=Nw36Djuh_5VDukK0H78QzOX-_FQEo6V37m3nkm96gtU,91
-sunholo-0.140.6.dist-info/entry_points.txt,sha256=bZuN5AIHingMPt4Ro1b_T-FnQvZ3teBes-3OyO0asl4,49
-sunholo-0.140.6.dist-info/top_level.txt,sha256=wt5tadn5--5JrZsjJz2LceoUvcrIvxjHJe-RxuudxAk,8
-sunholo-0.140.6.dist-info/RECORD,,
+sunholo-0.140.8.dist-info/licenses/LICENSE.txt,sha256=SdE3QjnD3GEmqqg9EX3TM9f7WmtOzqS1KJve8rhbYmU,11345
+sunholo-0.140.8.dist-info/METADATA,sha256=30zPLVCgeU87lsGIdFxyaAOFvYDuC-EOayXWNgophxI,10067
+sunholo-0.140.8.dist-info/WHEEL,sha256=Nw36Djuh_5VDukK0H78QzOX-_FQEo6V37m3nkm96gtU,91
+sunholo-0.140.8.dist-info/entry_points.txt,sha256=bZuN5AIHingMPt4Ro1b_T-FnQvZ3teBes-3OyO0asl4,49
+sunholo-0.140.8.dist-info/top_level.txt,sha256=wt5tadn5--5JrZsjJz2LceoUvcrIvxjHJe-RxuudxAk,8
+sunholo-0.140.8.dist-info/RECORD,,