write logs for inference and loader to file

Azaion.Inference/azaion-inference.spec

@@ -26,6 +26,8 @@ tmp_ret = collect_all('pynvml')
 datas += tmp_ret[0]; binaries += tmp_ret[1]; hiddenimports += tmp_ret[2]
 tmp_ret = collect_all('jwt')
 datas += tmp_ret[0]; binaries += tmp_ret[1]; hiddenimports += tmp_ret[2]
+tmp_ret = collect_all('loguru')
+datas += tmp_ret[0]; binaries += tmp_ret[1]; hiddenimports += tmp_ret[2]
 
 
 a = Analysis(

Azaion.Inference/build_inference.cmd

@@ -33,6 +33,7 @@ venv\Scripts\pyinstaller --name=azaion-inference ^
 --collect-all pycuda ^
 --collect-all pynvml ^
 --collect-all jwt ^
+--collect-all loguru ^
 --hidden-import constants ^
 --hidden-import file_data ^
 --hidden-import remote_command ^

Azaion.Inference/constants.pxd

@@ -13,5 +13,5 @@ cdef str MODELS_FOLDER
 
 cdef int SMALL_SIZE_KB
 
-
-cdef log(str log_message, bytes client_id=*)
+cdef log(str log_message)
+cdef logerror(str error)

Azaion.Inference/constants.pyx

@@ -1,4 +1,6 @@
-import time
+import sys
+
+from loguru import logger
 
 cdef str CONFIG_FILE = "config.yaml"  # Port for the zmq
 
@@ -10,7 +12,32 @@ cdef str MODELS_FOLDER = "models"
 
 cdef int SMALL_SIZE_KB = 3
 
-cdef log(str log_message, bytes client_id=None):
-    local_time = time.strftime('%Y-%m-%d %H:%M:%S', time.localtime())
-    client_str = '' if client_id is None else f' {client_id}'
-    print(f'[{local_time}{client_str}]: {log_message}')
+logger.remove()
+log_format = "[{time:HH:mm:ss} {level}] {message}"
+logger.add(
+    sink="Logs/log_inference_{time:YYYYMMDD}.txt",
+    level="INFO",
+    format=log_format,
+    enqueue=True,
+    rotation="1 day",
+    retention="30 days",
+)
+logger.add(
+    sys.stdout,
+    level="DEBUG",
+    format=log_format,
+    filter=lambda record: record["level"].name in ("INFO", "DEBUG", "SUCCESS"),
+    colorize=True
+)
+logger.add(
+    sys.stderr,
+    level="WARNING",
+    format=log_format,
+    colorize=True
+)
+
+cdef log(str log_message):
+    logger.info(log_message)
+
+cdef logerror(str error):
+    logger.error(error)

Azaion.Inference/inference.pyx

@@ -16,7 +16,7 @@ cdef int check_tensor_gpu_index():
         deviceCount = pynvml.nvmlDeviceGetCount()
 
         if deviceCount == 0:
-            print('No NVIDIA GPUs found.')
+            constants.logerror('No NVIDIA GPUs found.')
             return -1
 
         for i in range(deviceCount):
@@ -24,10 +24,10 @@ cdef int check_tensor_gpu_index():
             major, minor = pynvml.nvmlDeviceGetCudaComputeCapability(handle)
 
             if major > 6 or (major == 6 and minor >= 1):
-                print('found NVIDIA GPU!')
+                constants.log('found NVIDIA GPU!')
                 return i
 
-        print('NVIDIA GPU doesnt support TensorRT!')
+        constants.logerror('NVIDIA GPU doesnt support TensorRT!')
         return -1
 
     except pynvml.NVMLError:
@@ -36,7 +36,7 @@ cdef int check_tensor_gpu_index():
         try:
             pynvml.nvmlShutdown()
         except:
-            print('Failed to shutdown pynvml cause probably no NVidia GPU')
+            constants.logerror('Failed to shutdown pynvml cause probably no NVIDIA GPU')
             pass
 
 tensor_gpu_index = check_tensor_gpu_index()
@@ -70,15 +70,15 @@ cdef class Inference:
 
             res = self.loader_client.load_big_small_resource(engine_filename, models_dir)
             if res.err is None:
-                print('tensor rt engine is here, no need to build')
+                constants.log('tensor rt engine is here, no need to build')
                 self.is_building_engine = False
                 updater_callback('enabled')
                 return
 
-            print(res.err)
+            constants.logerror(res.err)
             # time.sleep(8) # prevent simultaneously loading dll and models
             updater_callback('converting')
-            print('try to load onnx')
+            constants.log('try to load onnx')
             res = self.loader_client.load_big_small_resource(constants.AI_ONNX_MODEL_FILE, models_dir)
             if res.err is not None:
                 updater_callback(f'Error. {res.err}')
@@ -87,7 +87,7 @@ cdef class Inference:
             res = self.loader_client.upload_big_small_resource(model_bytes, <str> engine_filename, models_dir)
             if res.err is not None:
                 updater_callback(f'Error. {res.err}')
-            print(f'uploaded {engine_filename} to CDN and API')
+            constants.log(f'uploaded {engine_filename} to CDN and API')
             self.is_building_engine = False
             updater_callback('enabled')
         except Exception as e:
@@ -212,11 +212,11 @@ cdef class Inference:
         # images first, it's faster
         if len(images) > 0:
             for chunk in self.split_list_extend(images, self.engine.get_batch_size()):
-                print(f'run inference on {" ".join(chunk)}...')
+                constants.log(f'run inference on {" ".join(chunk)}...')
                 self._process_images(cmd, ai_config, chunk)
         if len(videos) > 0:
             for v in videos:
-                print(f'run inference on {v}...')
+                constants.log(f'run inference on {v}...')
                 self._process_video(cmd, ai_config, v)
 
 

Azaion.Inference/main_inference.pyx

@@ -37,7 +37,7 @@ cdef class CommandProcessor:
                 continue
             except Exception as e:
                 traceback.print_exc()
-        print('EXIT!')
+        constants.log('EXIT!')
 
     cdef on_command(self, RemoteCommand command):
         try:
@@ -54,7 +54,7 @@ cdef class CommandProcessor:
             else:
                 pass
         except Exception as e:
-            print(f"Error handling client: {e}")
+            constants.logerror(f"Error handling client: {e}")
 
     cdef on_annotation(self, RemoteCommand cmd, Annotation annotation):
         cdef RemoteCommand response = RemoteCommand(CommandType.INFERENCE_DATA, annotation.serialize())

Azaion.Inference/onnx_engine.pyx

@@ -1,5 +1,6 @@
 from inference_engine cimport InferenceEngine
 import onnxruntime as onnx
+cimport constants
 
 cdef class OnnxEngine(InferenceEngine):
     def __init__(self, model_bytes: bytes, batch_size: int = 1, **kwargs):
@@ -10,9 +11,9 @@ cdef class OnnxEngine(InferenceEngine):
         self.input_name = self.model_inputs[0].name
         self.input_shape = self.model_inputs[0].shape
         self.batch_size = self.input_shape[0] if self.input_shape[0] != -1 else batch_size
-        print(f'AI detection model input: {self.model_inputs} {self.input_shape}')
+        constants.log(f'AI detection model input: {self.model_inputs} {self.input_shape}')
         model_meta = self.session.get_modelmeta()
-        print("Metadata:", model_meta.custom_metadata_map)
+        constants.log(f"Metadata: {model_meta.custom_metadata_map}")
 
     cpdef tuple get_input_shape(self):
         shape = self.input_shape

Azaion.Inference/remote_command_handler.pyx

@@ -3,7 +3,6 @@ import zmq
 from threading import Thread, Event
 from remote_command cimport RemoteCommand
 cimport constants
-import yaml
 
 cdef class RemoteCommandHandler:
     def __init__(self, int zmq_port, object on_command):
@@ -28,7 +27,7 @@ cdef class RemoteCommandHandler:
         for _ in range(4):  # 4 worker threads
             worker = Thread(target=self._worker_loop, daemon=True)
             self._workers.append(worker)
-        print(f'Listening to commands on port {zmq_port}...')
+        constants.log(f'Listening to commands on port {zmq_port}...')
 
     cdef start(self):
         self._proxy_thread.start()
@@ -40,7 +39,7 @@ cdef class RemoteCommandHandler:
             zmq.proxy_steerable(self._router, self._dealer, control=self._control)
         except zmq.error.ZMQError as e:
             if self._shutdown_event.is_set():
-                print("Shutdown, exit proxy loop.")
+                constants.log("Shutdown, exit proxy loop.")
             else:
                 raise
 
@@ -59,7 +58,7 @@ cdef class RemoteCommandHandler:
                         client_id, message = worker_socket.recv_multipart()
                         cmd = RemoteCommand.from_msgpack(<bytes> message)
                         cmd.client_id = client_id
-                        constants.log(<str>f'{cmd}', client_id)
+                        constants.log(cmd)
                         self._on_command(cmd)
                 except Exception as e:
                     if not self._shutdown_event.is_set():

Azaion.Inference/requirements.txt

@@ -14,3 +14,4 @@ pycuda
 tensorrt
 pynvml
 boto3
+loguru

Azaion.Inference/tensorrt_engine.pyx

@@ -103,11 +103,11 @@ cdef class TensorRTEngine(InferenceEngine):
                 constants.log('Converting to supported fp16')
                 config.set_flag(trt.BuilderFlag.FP16)
             else:
-                print('Converting to supported fp32. (fp16 is not supported)')
+                constants.log('Converting to supported fp32. (fp16 is not supported)')
             plan = builder.build_serialized_network(network, config)
 
             if plan is None:
-                print('Conversion failed.')
+                constants.logerror('Conversion failed.')
                 return None
             constants.log('conversion done!')
             return bytes(plan)

Azaion.Inference/test/test_download_large_file.py

@@ -215,7 +215,7 @@ class Api:
             response = requests.post(url, data=payload, headers=headers, stream=True)
 
         if response.status_code == HTTPStatus.INTERNAL_SERVER_ERROR:
-            print('500!')
+            print('500!') #test
 
         key = self.get_encryption_key(hardware.hash)