Merge branch 'develop' into develop

doc/FAQ.md

@@ -41,6 +41,10 @@
 
 **A:** 通过pip命令安装自己编译出的whl包，并设置SERVING_BIN环境变量为编译出的serving二进制文件路径。
 
+#### Q: 使用Java客户端，mvn compile过程出现"No compiler is provided in this environment. Perhaps you are running on a JRE rather than a JDK?"错误
+
+**A:** 没有安装JDK，或者JAVA_HOME路径配置错误（正确配置是JDK路径，常见错误配置成JRE路径，例如正确路径参考JAVA_HOME="/usr/lib/jvm/java-1.8.0-openjdk-1.8.0.262.b10-0.el7_8.x86_64/"）。Java JDK安装参考https://segmentfault.com/a/1190000015389941
+
 
 
 ## 部署问题

doc/GRPC_IMPL_CN.md

-# gRPC接口
+# gRPC接口使用介绍
+
+  - [1.与bRPC接口对比](#1与brpc接口对比)
+      - [1.1 服务端对比](#11-服务端对比)
+      - [1.2 客服端对比](#12-客服端对比)
+      - [1.3 其他](#13-其他)
+  - [2.示例：线性回归预测服务](#2示例线性回归预测服务)
+      - [获取数据](#获取数据)
+      - [开启 gRPC 服务端](#开启-grpc-服务端)
+    - [客户端预测](#客户端预测)
+      - [同步预测](#同步预测)
+      - [异步预测](#异步预测)
+      - [Batch 预测](#batch-预测)
+      - [通用 pb 预测](#通用-pb-预测)
+      - [预测超时](#预测超时)
+      - [List 输入](#list-输入)
+  - [3.更多示例](#3更多示例)
+  
+使用gRPC接口，Client端可以在Win/Linux/MacOS平台上调用不同语言。gRPC 接口实现结构如下：
+
+![](https://github.com/PaddlePaddle/Serving/blob/develop/doc/grpc_impl.png)
+
+## 1.与bRPC接口对比
+
+#### 1.1 服务端对比
+
+* gRPC Server 端 `load_model_config` 函数添加 `client_config_path` 参数：
 
-gRPC 接口实现形式类似 Web Service：
-
-![](grpc_impl.png)
-
-## 与bRPC接口对比
-
-1. gRPC Server 端 `load_model_config` 函数添加 `client_config_path` 参数：
-
-   ```python
+   ```
    def load_model_config(self, server_config_paths, client_config_path=None)
    ```
+    在一些例子中 bRPC Server 端与 bRPC Client 端的配置文件可能不同（如 在cube local 中，Client 端的数据先交给 cube，经过 cube 处理后再交给预测库），此时 gRPC Server 端需要手动设置 gRPC Client 端的配置`client_config_path`。
+    **`client_config_path` 默认为 `<server_config_path>/serving_server_conf.prototxt`。**
 
-   在一些例子中 bRPC Server 端与 bRPC Client 端的配置文件可能是不同的（如 cube local 例子中，Client 端的数据先交给 cube，经过 cube 处理后再交给预测库），所以 gRPC Server 端需要获取 gRPC Client 端的配置；同时为了取消 gRPC Client 端手动加载配置文件的过程，所以设计 gRPC Server 端同时加载两个配置文件。`client_config_path` 默认为 `<server_config_path>/serving_server_conf.prototxt`。
+#### 1.2 客服端对比
 
-2. gRPC Client 端取消 `load_client_config` 步骤：
+* gRPC Client 端取消 `load_client_config` 步骤：
 
    在 `connect` 步骤通过 RPC 获取相应的 prototxt（从任意一个 endpoint 获取即可）。
 
-3. gRPC Client 需要通过 RPC 方式设置 timeout 时间（调用形式与 bRPC Client保持一致）
+* gRPC Client 需要通过 RPC 方式设置 timeout 时间（调用形式与 bRPC Client保持一致）
 
    因为 bRPC Client 在 `connect` 后无法更改 timeout 时间，所以当 gRPC Server 收到变更 timeout 的调用请求时会重新创建 bRPC Client 实例以变更 bRPC Client timeout时间，同时 gRPC Client 会设置 gRPC 的 deadline 时间。
 
    **注意，设置 timeout 接口和 Inference 接口不能同时调用（非线程安全），出于性能考虑暂时不加锁。**
 
-4. gRPC Client 端 `predict` 函数添加 `asyn` 和 `is_python` 参数：
+* gRPC Client 端 `predict` 函数添加 `asyn` 和 `is_python` 参数：
 
-   ```python
+   ```
    def predict(self, feed, fetch, need_variant_tag=False, asyn=False, is_python=True)
    ```
 
-   其中，`asyn` 为异步调用选项。当 `asyn=True` 时为异步调用，返回 `MultiLangPredictFuture` 对象，通过 `MultiLangPredictFuture.result()` 阻塞获取预测值；当 `asyn=Fasle` 为同步调用。
+1.    `asyn` 为异步调用选项。当 `asyn=True` 时为异步调用，返回 `MultiLangPredictFuture` 对象，通过 `MultiLangPredictFuture.result()` 阻塞获取预测值；当 `asyn=Fasle` 为同步调用。
+
+2.    `is_python` 为 proto 格式选项。当 `is_python=True` 时，基于 numpy bytes 格式进行数据传输，目前只适用于 Python；当 `is_python=False` 时，以普通数据格式传输，更加通用。使用 numpy bytes 格式传输耗时比普通数据格式小很多（详见 [#654](https://github.com/PaddlePaddle/Serving/pull/654)）。
+
+#### 1.3 其他
+
+* 异常处理：当 gRPC Server 端的 bRPC Client 预测失败（返回 `None`）时，gRPC Client 端同样返回None。其他 gRPC 异常会在 Client 内部捕获，并在返回的 fetch_map 中添加一个 "status_code" 字段来区分是否预测正常（参考 timeout 样例）。
+
+* 由于 gRPC 只支持 pick_first 和 round_robin 负载均衡策略，ABTEST 特性还未打齐。
+
+* 系统兼容性：
+    * [x]  CentOS
+    * [x]  macOS
+    * [x]  Windows
+
+* 已经支持的客户端语言：
+
+   -  Python
+   -  Java
+   -  Go
+   
+   
+## 2.示例：线性回归预测服务
+
+以下是采用gRPC实现的关于线性回归预测的一个示例，具体代码详见此[链接](https://github.com/PaddlePaddle/Serving/tree/develop/python/examples/grpc_impl_example/fit_a_line)
+#### 获取数据
+
+```shell
+sh get_data.sh
+```
+
+#### 开启 gRPC 服务端
+
+``` shell
+python test_server.py uci_housing_model/
+```
+
+也可以通过下面的一行代码开启默认 gRPC 服务：
+
+```shell
+python -m paddle_serving_server.serve --model uci_housing_model --thread 10 --port 9393 --use_multilang
+```
+注：--use_multilang参数用来启用多语言客户端
+
+### 客户端预测
+
+#### 同步预测
+
+``` shell
+python test_sync_client.py
+```
+
+#### 异步预测
+
+``` shell
+python test_asyn_client.py
+```
+
+#### Batch 预测
+
+``` shell
+python test_batch_client.py
+```
 
-   `is_python` 为 proto 格式选项。当 `is_python=True` 时，基于 numpy bytes 格式进行数据传输，目前只适用于 Python；当 `is_python=False` 时，以普通数据格式传输，更加通用。使用 numpy bytes 格式传输耗时比普通数据格式小很多（详见 [#654](https://github.com/PaddlePaddle/Serving/pull/654)）。
+#### 通用 pb 预测
 
-5. 异常处理：当 gRPC Server 端的 bRPC Client 预测失败（返回 `None`）时，gRPC Client 端同样返回None。其他 gRPC 异常会在 Client 内部捕获，并在返回的 fetch_map 中添加一个 "status_code" 字段来区分是否预测正常（参考 timeout 样例）。
+``` shell
+python test_general_pb_client.py
+```
 
-6. 由于 gRPC 只支持 pick_first 和 round_robin 负载均衡策略，ABTEST 特性还未打齐。
+#### 预测超时
 
-7. 经测试，gRPC 版本可以在 Windows、macOS 平台使用。
+``` shell
+python test_timeout_client.py
+```
 
-8. 计划支持的客户端语言：
+#### List 输入
 
-   - [x] Python
-   - [ ] Java
-   - [ ] Go
-   - [ ] JavaScript
+``` shell
+python test_list_input_client.py
+```
 
-## Python 端的一些例子 
+## 3.更多示例
 
-详见 `python/examples/grpc_impl_example` 下的示例文件。
+详见[`python/examples/grpc_impl_example`](https://github.com/PaddlePaddle/Serving/tree/develop/python/examples/grpc_impl_example)下的示例文件。

python/examples/bert/bert_client.py

@@ -23,7 +23,7 @@ args = benchmark_args()
 
 reader = ChineseBertReader({"max_seq_len": 128})
 fetch = ["pooled_output"]
-endpoint_list = ['127.0.0.1:8861']
+endpoint_list = ['127.0.0.1:9292']
 client = Client()
 client.load_client_config(args.model)
 client.connect(endpoint_list)
@@ -33,5 +33,5 @@ for line in sys.stdin:
     for key in feed_dict.keys():
         feed_dict[key] = np.array(feed_dict[key]).reshape((128, 1))
     #print(feed_dict)
-    result = client.predict(feed=feed_dict, fetch=fetch)
+    result = client.predict(feed=feed_dict, fetch=fetch, batch=True)
 print(result)

python/examples/bert/bert_web_service.py

@@ -29,13 +29,14 @@ class BertService(WebService):
 
     def preprocess(self, feed=[], fetch=[]):
         feed_res = []
+        is_batch = True
         for ins in feed:
             feed_dict = self.reader.process(ins["words"].encode("utf-8"))
             for key in feed_dict.keys():
                 feed_dict[key] = np.array(feed_dict[key]).reshape(
-                    (1, len(feed_dict[key]), 1))
+                    (len(feed_dict[key]), 1))
             feed_res.append(feed_dict)
-        return feed_res, fetch
+        return feed_res, fetch, is_batch
 
 
 bert_service = BertService(name="bert")

python/examples/imdb/benchmark.py

@@ -18,7 +18,7 @@ import sys
 import time
 import requests
 import numpy as np
-from paddle_serving_app.reader import IMDBDataset
+from paddle_serving_app.reader.imdb_reader import IMDBDataset
 from paddle_serving_client import Client
 from paddle_serving_client.utils import MultiThreadRunner
 from paddle_serving_client.utils import MultiThreadRunner, benchmark_args, show_latency

python/examples/imdb/test_client.py

@@ -13,7 +13,7 @@
 # limitations under the License.
 # pylint: disable=doc-string-missing
 from paddle_serving_client import Client
-from paddle_serving_app.reader import IMDBDataset
+from paddle_serving_app.reader.imdb_reader import IMDBDataset
 import sys
 import numpy as np
 

python/examples/imdb/text_classify_service.py

@@ -14,7 +14,7 @@
 # pylint: disable=doc-string-missing
 
 from paddle_serving_server.web_service import WebService
-from paddle_serving_app.reader import IMDBDataset
+from paddle_serving_app.reader.imdb_reader import IMDBDataset
 import sys
 import numpy as np
 

python/examples/pipeline/imdb_model_ensemble/config.yml

-rpc_port: 18080
+#rpc端口, rpc_port和http_port不允许同时为空。当rpc_port为空且http_port不为空时，会自动将rpc_port设置为http_port+1
+rpc_port: 18070
+
+#http端口, rpc_port和http_port不允许同时为空。当rpc_port可用且http_port为空时，不自动生成http_port
+http_port: 18071
+
+#worker_num, 最大并发数。当build_dag_each_worker=True时, 框架会创建worker_num个进程，每个进程内构建grpcSever和DAG
+#当build_dag_each_worker=False时，框架会设置主线程grpc线程池的max_workers=worker_num
 worker_num: 4
-build_dag_each_worker: false
+
+#build_dag_each_worker, False，框架在进程内创建一条DAG；True，框架会每个进程内创建多个独立的DAG
+build_dag_each_worker: False
+
 dag:
-    is_thread_op: true
+    #op资源类型, True, 为线程模型；False，为进程模型
+    is_thread_op: True
+
+    #重试次数
     retry: 1
-    use_profile: false
+
+    #使用性能分析, True，生成Timeline性能数据，对性能有一定影响；False为不使用
+    use_profile: False
+
+    #channel的最大长度，默认为0
+    channel_size: 0
+
+    #tracer, 跟踪框架吞吐，每个OP和channel的工作情况。无tracer时不生成数据
+    tracer:
+        #每次trace的时间间隔，单位秒/s
+        interval_s: 10
 op:
     bow:
-        concurrency: 2
-        remote_service_conf:
-            client_type: brpc
-            model_config: imdb_bow_model
-            devices: ""
-            rpc_port : 9393 
+        #并发数，is_thread_op=True时，为线程并发；否则为进程并发
+        concurrency: 1
+    
+        #client连接类型，brpc
+        client_type: brpc
+
+        #Serving交互重试次数，默认不重试
+        retry: 1
+
+        #Serving交互超时时间, 单位ms
+        timeout: 3000
+
+        #Serving IPs
+        server_endpoints: ["127.0.0.1:9393"]
+
+        #bow模型client端配置
+        client_config: "imdb_bow_client_conf/serving_client_conf.prototxt"
+
+        #Fetch结果列表，以client_config中fetch_var的alias_name为准
+        fetch_list: ["prediction"]
+
+        #批量查询Serving的数量, 默认1。batch_size>1要设置auto_batching_timeout，否则不足batch_size时会阻塞
+        batch_size: 1
+
+        #批量查询超时，与batch_size配合使用
+        auto_batching_timeout: 2000
     cnn:
-        concurrency: 2
-        remote_service_conf:
-            client_type: brpc
-            model_config: imdb_cnn_model
-            devices: ""
-            rpc_port : 9292
+        #并发数，is_thread_op=True时，为线程并发；否则为进程并发
+        concurrency: 1
+
+        #client连接类型，brpc
+        client_type: brpc
+
+        #Serving交互重试次数，默认不重试
+        retry: 1
+
+        #超时时间, 单位ms
+        timeout: 3000
+
+        #Serving IPs
+        server_endpoints: ["127.0.0.1:9292"]
+
+        #cnn模型client端配置
+        client_config: "imdb_cnn_client_conf/serving_client_conf.prototxt"
+
+        #Fetch结果列表，以client_config中fetch_var的alias_name为准
+        fetch_list: ["prediction"]
+        
+        #批量查询Serving的数量, 默认1。batch_size>1要设置auto_batching_timeout，否则不足batch_size时会阻塞
+        batch_size: 1
+
+        #批量查询超时，与batch_size配合使用
+        auto_batching_timeout: 2000
+    combine:
+        #并发数，is_thread_op=True时，为线程并发；否则为进程并发
+        concurrency: 1
+
+        #Serving交互重试次数，默认不重试
+        retry: 1
+
+        #超时时间, 单位ms
+        timeout: 3000
+        
+        #批量查询Serving的数量, 默认1。batch_size>1要设置auto_batching_timeout，否则不足batch_size时会阻塞
+        batch_size: 1
+
+        #批量查询超时，与batch_size配合使用
+        auto_batching_timeout: 2000

python/examples/pipeline/imdb_model_ensemble/test_pipeline_client.py

@@ -15,21 +15,22 @@ from paddle_serving_server.pipeline import PipelineClient
 import numpy as np
 
 client = PipelineClient()
-client.connect(['127.0.0.1:18080'])
+client.connect(['127.0.0.1:18070'])
 
 words = 'i am very sad | 0'
 
 futures = []
-for i in range(4):
+for i in range(100):
     futures.append(
         client.predict(
-            feed_dict={"words": words},
+            feed_dict={"words": words,
+                       "logid": 10000 + i},
             fetch=["prediction"],
             asyn=True,
             profile=False))
 
 for f in futures:
     res = f.result()
-    if res["ecode"] != 0:
+    if res.err_no != 0:
         print("predict failed: {}".format(res))
     print(res)

python/examples/pipeline/imdb_model_ensemble/test_pipeline_server.py

@@ -15,10 +15,14 @@
 from paddle_serving_server.pipeline import Op, RequestOp, ResponseOp
 from paddle_serving_server.pipeline import PipelineServer
 from paddle_serving_server.pipeline.proto import pipeline_service_pb2
-from paddle_serving_server.pipeline.channel import ChannelDataEcode
+from paddle_serving_server.pipeline.channel import ChannelDataErrcode
 import numpy as np
-from paddle_serving_app.reader import IMDBDataset
+from paddle_serving_app.reader.imdb_reader import IMDBDataset
 import logging
+try:
+    from paddle_serving_server.web_service import WebService
+except ImportError:
+    from paddle_serving_server_gpu.web_service import WebService
 
 _LOGGER = logging.getLogger()
 user_handler = logging.StreamHandler()
@@ -43,76 +47,66 @@ class ImdbRequestOp(RequestOp):
             word_ids, _ = self.imdb_dataset.get_words_and_label(words)
             word_len = len(word_ids)
             dictdata[key] = np.array(word_ids).reshape(word_len, 1)
-            dictdata["{}.lod".format(key)] = [0, word_len]
-        return dictdata
+            dictdata["{}.lod".format(key)] = np.array([0, word_len])
+
+        log_id = None
+        if request.logid is not None:
+            log_id = request.logid
+        return dictdata, log_id, None, ""
 
 
 class CombineOp(Op):
-    def preprocess(self, input_data):
+    def preprocess(self, input_data, data_id, log_id):
+        #_LOGGER.info("Enter CombineOp::preprocess")
         combined_prediction = 0
         for op_name, data in input_data.items():
             _LOGGER.info("{}: {}".format(op_name, data["prediction"]))
             combined_prediction += data["prediction"]
         data = {"prediction": combined_prediction / 2}
-        return data
+        return data, False, None, ""
 
 
 class ImdbResponseOp(ResponseOp):
     # Here ImdbResponseOp is consistent with the default ResponseOp implementation
     def pack_response_package(self, channeldata):
         resp = pipeline_service_pb2.Response()
-        resp.ecode = channeldata.ecode
-        if resp.ecode == ChannelDataEcode.OK.value:
+        resp.err_no = channeldata.error_code
+        if resp.err_no == ChannelDataErrcode.OK.value:
             feed = channeldata.parse()
             # ndarray to string
             for name, var in feed.items():
                 resp.value.append(var.__repr__())
                 resp.key.append(name)
         else:
-            resp.error_info = channeldata.error_info
+            resp.err_msg = channeldata.error_info
         return resp
 
 
 read_op = ImdbRequestOp()
-bow_op = Op(name="bow",
-            input_ops=[read_op],
-            server_endpoints=["127.0.0.1:9393"],
-            fetch_list=["prediction"],
-            client_config="imdb_bow_client_conf/serving_client_conf.prototxt",
-            client_type='brpc',
-            concurrency=1,
-            timeout=-1,
-            retry=1,
-            batch_size=1,
-            auto_batching_timeout=None)
-cnn_op = Op(name="cnn",
-            input_ops=[read_op],
-            server_endpoints=["127.0.0.1:9292"],
-            fetch_list=["prediction"],
-            client_config="imdb_cnn_client_conf/serving_client_conf.prototxt",
-            client_type='brpc',
-            concurrency=1,
-            timeout=-1,
-            retry=1,
-            batch_size=1,
-            auto_batching_timeout=None)
-combine_op = CombineOp(
-    name="combine",
-    input_ops=[bow_op, cnn_op],
-    concurrency=1,
-    timeout=-1,
-    retry=1,
-    batch_size=2,
-    auto_batching_timeout=None)
+
+
+class BowOp(Op):
+    def init_op(self):
+        pass
+
+
+class CnnOp(Op):
+    def init_op(self):
+        pass
+
+
+bow_op = BowOp("bow", input_ops=[read_op])
+cnn_op = CnnOp("cnn", input_ops=[read_op])
+combine_op = CombineOp("combine", input_ops=[bow_op, cnn_op])
 
 # fetch output of bow_op
-# response_op = ImdbResponseOp(input_ops=[bow_op])
+#response_op = ImdbResponseOp(input_ops=[bow_op])
 
 # fetch output of combine_op
 response_op = ImdbResponseOp(input_ops=[combine_op])
 
 # use default ResponseOp implementation
-# response_op = ResponseOp(input_ops=[combine_op])
+#response_op = ResponseOp(input_ops=[combine_op])
 
 server = PipelineServer()
 server.set_response_op(response_op)

python/examples/pipeline/ocr/README.md

@@ -28,31 +28,9 @@ python web_service.py &>log.txt &
 python pipeline_http_client.py
 ```
 
-
-
 <!--
 ## More (PipelineServing)
 
-You can choose one of the following versions to start Service.
-
-### Remote Service Version
-```
-python -m paddle_serving_server_gpu.serve --model ocr_det_model --port 12000 --gpu_id 0 &> det.log &
-python -m paddle_serving_server_gpu.serve --model ocr_rec_model --port 12001 --gpu_id 0 &> rec.log &
-python remote_service_pipeline_server.py &>pipeline.log &
-```
-
-### Local Service Version
-```
-python local_service_pipeline_server.py &>pipeline.log &
-```
-
-### Hybrid Service Version
-```
-python -m paddle_serving_server_gpu.serve --model ocr_rec_model --port 12001 --gpu_id 0 &> rec.log &
-python hybrid_service_pipeline_server.py &>pipeline.log &
-```
-
 ## Client Prediction
 
 ### RPC

python/examples/pipeline/ocr/README_CN.md

@@ -31,26 +31,6 @@ python pipeline_http_client.py
 <!--
 ## 其他 (PipelineServing)
 
-你可以选择下面任意一种版本启动服务。
-
-### 远程服务版本
-```
-python -m paddle_serving_server.serve --model ocr_det_model --port 12000 --gpu_id 0 &> det.log &
-python -m paddle_serving_server.serve --model ocr_rec_model --port 12001 --gpu_id 0 &> rec.log &
-python remote_service_pipeline_server.py &>pipeline.log &
-```
-
-### 本地服务版本
-```
-python local_service_pipeline_server.py &>pipeline.log &
-```
-
-### 混合服务版本
-```
-python -m paddle_serving_server_gpu.serve --model ocr_rec_model --port 12001 --gpu_id 0 &> rec.log &
-python hybrid_service_pipeline_server.py &>pipeline.log &
-```
-
 ## 启动客户端
 
 ### RPC

python/examples/pipeline/ocr/config.yml

-rpc_port: 18080
-worker_num: 4
-build_dag_each_worker: false
+#rpc端口, rpc_port和http_port不允许同时为空。当rpc_port为空且http_port不为空时，会自动将rpc_port设置为http_port+1
+rpc_port: 18090
+
+#http端口, rpc_port和http_port不允许同时为空。当rpc_port可用且http_port为空时，不自动生成http_port
 http_port: 9999
+
+#worker_num, 最大并发数。当build_dag_each_worker=True时, 框架会创建worker_num个进程，每个进程内构建grpcSever和DAG
+##当build_dag_each_worker=False时，框架会设置主线程grpc线程池的max_workers=worker_num
+worker_num: 1
+
+#build_dag_each_worker, False，框架在进程内创建一条DAG；True，框架会每个进程内创建多个独立的DAG
+build_dag_each_worker: false
+
 dag:
-    is_thread_op: false
+    #op资源类型, True, 为线程模型；False，为进程模型
+    is_thread_op: False
+
+    #重试次数
     retry: 1
+
+    #使用性能分析, True，生成Timeline性能数据，对性能有一定影响；False为不使用
     use_profile: false
 op:
     det:
+        #并发数，is_thread_op=True时，为线程并发；否则为进程并发
         concurrency: 2
+
+        #当op配置没有server_endpoints时，从local_service_conf读取本地服务配置
         local_service_conf:
+            #client类型，包括brpc, grpc和local_predictor.local_predictor不启动Serving服务，进程内预测
             client_type: local_predictor
+
+            #det模型路径
             model_config: ocr_det_model
-            devices: ""
+
+            #Fetch结果列表，以client_config中fetch_var的alias_name为准
+            fetch_list: ["concat_1.tmp_0"]
+
+            #计算硬件ID，当devices为""或不写时为CPU预测；当devices为"0", "0,1,2"时为GPU预测，表示使用的GPU卡
+            devices: "0"
     rec:
-        concurrency: 1
+        #并发数，is_thread_op=True时，为线程并发；否则为进程并发
+        concurrency: 2
+
+        #超时时间, 单位ms
         timeout: -1
+ 
+        #Serving交互重试次数，默认不重试
         retry: 1
+
+        #当op配置没有server_endpoints时，从local_service_conf读取本地服务配置
         local_service_conf:
+
+            #client类型，包括brpc, grpc和local_predictor。local_predictor不启动Serving服务，进程内预测
             client_type: local_predictor
+
+            #rec模型路径
             model_config: ocr_rec_model
-            devices: ""
+
+            #Fetch结果列表，以client_config中fetch_var的alias_name为准
+            fetch_list: ["ctc_greedy_decoder_0.tmp_0", "softmax_0.tmp_0"] 
+
+            #计算硬件ID，当devices为""或不写时为CPU预测；当devices为"0", "0,1,2"时为GPU预测，表示使用的GPU卡
+            devices: "0"

python/examples/pipeline/ocr/hybrid_service_pipeline_server.py

-# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# 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.
-# pylint: disable=doc-string-missing
-from paddle_serving_server_gpu.pipeline import Op, RequestOp, ResponseOp
-from paddle_serving_server_gpu.pipeline import PipelineServer
-from paddle_serving_server_gpu.pipeline.proto import pipeline_service_pb2
-from paddle_serving_server_gpu.pipeline.channel import ChannelDataEcode
-from paddle_serving_server_gpu.pipeline import LocalRpcServiceHandler
-import numpy as np
-import cv2
-import time
-import base64
-import json
-from paddle_serving_app.reader import OCRReader
-from paddle_serving_app.reader import Sequential, ResizeByFactor
-from paddle_serving_app.reader import Div, Normalize, Transpose
-from paddle_serving_app.reader import DBPostProcess, FilterBoxes, GetRotateCropImage, SortedBoxes
-import time
-import re
-import base64
-import logging
-
-_LOGGER = logging.getLogger()
-
-
-class DetOp(Op):
-    def init_op(self):
-        self.det_preprocess = Sequential([
-            ResizeByFactor(32, 960), Div(255),
-            Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]), Transpose(
-                (2, 0, 1))
-        ])
-        self.filter_func = FilterBoxes(10, 10)
-        self.post_func = DBPostProcess({
-            "thresh": 0.3,
-            "box_thresh": 0.5,
-            "max_candidates": 1000,
-            "unclip_ratio": 1.5,
-            "min_size": 3
-        })
-
-    def preprocess(self, input_dicts):
-        (_, input_dict), = input_dicts.items()
-        data = base64.b64decode(input_dict["image"].encode('utf8'))
-        data = np.fromstring(data, np.uint8)
-        # Note: class variables(self.var) can only be used in process op mode
-        self.im = cv2.imdecode(data, cv2.IMREAD_COLOR)
-        self.ori_h, self.ori_w, _ = self.im.shape
-        det_img = self.det_preprocess(self.im)
-        _, self.new_h, self.new_w = det_img.shape
-        return {"image": det_img}
-
-    def postprocess(self, input_dicts, fetch_dict):
-        det_out = fetch_dict["concat_1.tmp_0"]
-        ratio_list = [
-            float(self.new_h) / self.ori_h, float(self.new_w) / self.ori_w
-        ]
-        dt_boxes_list = self.post_func(det_out, [ratio_list])
-        dt_boxes = self.filter_func(dt_boxes_list[0], [self.ori_h, self.ori_w])
-        out_dict = {"dt_boxes": dt_boxes, "image": self.im}
-        return out_dict
-
-
-class RecOp(Op):
-    def init_op(self):
-        self.ocr_reader = OCRReader()
-        self.get_rotate_crop_image = GetRotateCropImage()
-        self.sorted_boxes = SortedBoxes()
-
-    def preprocess(self, input_dicts):
-        (_, input_dict), = input_dicts.items()
-        im = input_dict["image"]
-        dt_boxes = input_dict["dt_boxes"]
-        dt_boxes = self.sorted_boxes(dt_boxes)
-        feed_list = []
-        img_list = []
-        max_wh_ratio = 0
-        for i, dtbox in enumerate(dt_boxes):
-            boximg = self.get_rotate_crop_image(im, dt_boxes[i])
-            img_list.append(boximg)
-            h, w = boximg.shape[0:2]
-            wh_ratio = w * 1.0 / h
-            max_wh_ratio = max(max_wh_ratio, wh_ratio)
-        for img in img_list:
-            norm_img = self.ocr_reader.resize_norm_img(img, max_wh_ratio)
-            feed = {"image": norm_img}
-            feed_list.append(feed)
-        return feed_list
-
-    def postprocess(self, input_dicts, fetch_dict):
-        rec_res = self.ocr_reader.postprocess(fetch_dict, with_score=True)
-        res_lst = []
-        for res in rec_res:
-            res_lst.append(res[0])
-        res = {"res": str(res_lst)}
-        return res
-
-
-read_op = RequestOp()
-det_op = DetOp(
-    name="det",
-    input_ops=[read_op],
-    local_rpc_service_handler=LocalRpcServiceHandler(
-        model_config="ocr_det_model",
-        workdir="det_workdir",  # defalut: "workdir"
-        thread_num=2,  # defalut: 2
-        devices="0",  # gpu0. defalut: "" (cpu)
-        mem_optim=True,  # defalut: True
-        ir_optim=False,  # defalut: False
-        available_port_generator=None),  # defalut: None
-    concurrency=1)
-rec_op = RecOp(
-    name="rec",
-    input_ops=[det_op],
-    server_endpoints=["127.0.0.1:12001"],
-    fetch_list=["ctc_greedy_decoder_0.tmp_0", "softmax_0.tmp_0"],
-    client_config="ocr_rec_client/serving_client_conf.prototxt",
-    concurrency=1)
-response_op = ResponseOp(input_ops=[rec_op])
-
-server = PipelineServer("ocr")
-server.set_response_op(response_op)
-server.prepare_server('config.yml')
-server.run_server()

python/examples/pipeline/ocr/local_service_pipeline_server.py

-# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# 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.
-# pylint: disable=doc-string-missing
-from paddle_serving_server.pipeline import Op, RequestOp, ResponseOp
-from paddle_serving_server.pipeline import PipelineServer
-from paddle_serving_server.pipeline.proto import pipeline_service_pb2
-from paddle_serving_server.pipeline.channel import ChannelDataEcode
-from paddle_serving_server.pipeline import LocalServiceHandler
-import numpy as np
-import cv2
-import time
-import base64
-import json
-from paddle_serving_app.reader import OCRReader
-from paddle_serving_app.reader import Sequential, ResizeByFactor
-from paddle_serving_app.reader import Div, Normalize, Transpose
-from paddle_serving_app.reader import DBPostProcess, FilterBoxes, GetRotateCropImage, SortedBoxes
-import time
-import re
-import base64
-import logging
-
-_LOGGER = logging.getLogger()
-
-
-class DetOp(Op):
-    def init_op(self):
-        self.det_preprocess = Sequential([
-            ResizeByFactor(32, 960), Div(255),
-            Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]), Transpose(
-                (2, 0, 1))
-        ])
-        self.filter_func = FilterBoxes(10, 10)
-        self.post_func = DBPostProcess({
-            "thresh": 0.3,
-            "box_thresh": 0.5,
-            "max_candidates": 1000,
-            "unclip_ratio": 1.5,
-            "min_size": 3
-        })
-
-    def preprocess(self, input_dicts):
-        (_, input_dict), = input_dicts.items()
-        data = base64.b64decode(input_dict["image"].encode('utf8'))
-        data = np.fromstring(data, np.uint8)
-        # Note: class variables(self.var) can only be used in process op mode
-        self.im = cv2.imdecode(data, cv2.IMREAD_COLOR)
-        print(self.im)
-        self.ori_h, self.ori_w, _ = self.im.shape
-        det_img = self.det_preprocess(self.im)
-        _, self.new_h, self.new_w = det_img.shape
-        print("image", det_img)
-        return {"image": det_img}
-
-    def postprocess(self, input_dicts, fetch_dict):
-        det_out = fetch_dict["concat_1.tmp_0"]
-        ratio_list = [
-            float(self.new_h) / self.ori_h, float(self.new_w) / self.ori_w
-        ]
-        dt_boxes_list = self.post_func(det_out, [ratio_list])
-        dt_boxes = self.filter_func(dt_boxes_list[0], [self.ori_h, self.ori_w])
-        out_dict = {"dt_boxes": dt_boxes, "image": self.im}
-        return out_dict
-
-
-class RecOp(Op):
-    def init_op(self):
-        self.ocr_reader = OCRReader()
-        self.get_rotate_crop_image = GetRotateCropImage()
-        self.sorted_boxes = SortedBoxes()
-
-    def preprocess(self, input_dicts):
-        (_, input_dict), = input_dicts.items()
-        im = input_dict["image"]
-        dt_boxes = input_dict["dt_boxes"]
-        dt_boxes = self.sorted_boxes(dt_boxes)
-        feed_list = []
-        img_list = []
-        max_wh_ratio = 0
-        for i, dtbox in enumerate(dt_boxes):
-            boximg = self.get_rotate_crop_image(im, dt_boxes[i])
-            img_list.append(boximg)
-            h, w = boximg.shape[0:2]
-            wh_ratio = w * 1.0 / h
-            max_wh_ratio = max(max_wh_ratio, wh_ratio)
-        for img in img_list:
-            norm_img = self.ocr_reader.resize_norm_img(img, max_wh_ratio)
-            feed = {"image": norm_img}
-            feed_list.append(feed)
-        return feed_list
-
-    def postprocess(self, input_dicts, fetch_dict):
-        rec_res = self.ocr_reader.postprocess(fetch_dict, with_score=True)
-        res_lst = []
-        for res in rec_res:
-            res_lst.append(res[0])
-        res = {"res": str(res_lst)}
-        return res
-
-
-read_op = RequestOp()
-det_op = DetOp(
-    name="det",
-    input_ops=[read_op],
-    client_type="local_predictor",
-    local_service_handler=LocalServiceHandler(
-        model_config="ocr_det_model",
-        workdir="det_workdir",  # defalut: "workdir"
-        thread_num=2,  # defalut: 2
-        mem_optim=True,  # defalut: True
-        ir_optim=False,  # defalut: False
-        available_port_generator=None),  # defalut: None
-    concurrency=1)
-rec_op = RecOp(
-    name="rec",
-    input_ops=[det_op],
-    client_type="local_predictor",
-    local_service_handler=LocalServiceHandler(model_config="ocr_rec_model"),
-    concurrency=1)
-response_op = ResponseOp(input_ops=[rec_op])
-
-server = PipelineServer("ocr")
-server.set_response_op(response_op)
-server.prepare_server('config.yml')
-server.run_server()

python/examples/pipeline/ocr/pipeline_rpc_client.py

@@ -11,7 +11,7 @@
 # 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 paddle_serving_server.pipeline import PipelineClient
+from paddle_serving_server_gpu.pipeline import PipelineClient
 import numpy as np
 import requests
 import json
@@ -20,7 +20,7 @@ import base64
 import os
 
 client = PipelineClient()
-client.connect(['127.0.0.1:18080'])
+client.connect(['127.0.0.1:18090'])
 
 
 def cv2_to_base64(image):

python/examples/pipeline/ocr/remote_service_pipeline_server.py

-# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# 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.
-# pylint: disable=doc-string-missing
-from paddle_serving_server_gpu.pipeline import Op, RequestOp, ResponseOp
-from paddle_serving_server_gpu.pipeline import PipelineServer
-from paddle_serving_server_gpu.pipeline.proto import pipeline_service_pb2
-from paddle_serving_server_gpu.pipeline.channel import ChannelDataEcode
-import numpy as np
-import cv2
-import time
-import base64
-import json
-from paddle_serving_app.reader import OCRReader
-from paddle_serving_app.reader import Sequential, ResizeByFactor
-from paddle_serving_app.reader import Div, Normalize, Transpose
-from paddle_serving_app.reader import DBPostProcess, FilterBoxes, GetRotateCropImage, SortedBoxes
-import time
-import re
-import base64
-import logging
-
-_LOGGER = logging.getLogger()
-
-
-class DetOp(Op):
-    def init_op(self):
-        self.det_preprocess = Sequential([
-            ResizeByFactor(32, 960), Div(255),
-            Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]), Transpose(
-                (2, 0, 1))
-        ])
-        self.filter_func = FilterBoxes(10, 10)
-        self.post_func = DBPostProcess({
-            "thresh": 0.3,
-            "box_thresh": 0.5,
-            "max_candidates": 1000,
-            "unclip_ratio": 1.5,
-            "min_size": 3
-        })
-
-    def preprocess(self, input_dicts):
-        (_, input_dict), = input_dicts.items()
-        data = base64.b64decode(input_dict["image"].encode('utf8'))
-        data = np.fromstring(data, np.uint8)
-        # Note: class variables(self.var) can only be used in process op mode
-        self.im = cv2.imdecode(data, cv2.IMREAD_COLOR)
-        self.ori_h, self.ori_w, _ = self.im.shape
-        det_img = self.det_preprocess(self.im)
-        _, self.new_h, self.new_w = det_img.shape
-        return {"image": det_img}
-
-    def postprocess(self, input_dicts, fetch_dict):
-        det_out = fetch_dict["concat_1.tmp_0"]
-        ratio_list = [
-            float(self.new_h) / self.ori_h, float(self.new_w) / self.ori_w
-        ]
-        dt_boxes_list = self.post_func(det_out, [ratio_list])
-        dt_boxes = self.filter_func(dt_boxes_list[0], [self.ori_h, self.ori_w])
-        out_dict = {"dt_boxes": dt_boxes, "image": self.im}
-        return out_dict
-
-
-class RecOp(Op):
-    def init_op(self):
-        self.ocr_reader = OCRReader()
-        self.get_rotate_crop_image = GetRotateCropImage()
-        self.sorted_boxes = SortedBoxes()
-
-    def preprocess(self, input_dicts):
-        (_, input_dict), = input_dicts.items()
-        im = input_dict["image"]
-        dt_boxes = input_dict["dt_boxes"]
-        dt_boxes = self.sorted_boxes(dt_boxes)
-        feed_list = []
-        img_list = []
-        max_wh_ratio = 0
-        for i, dtbox in enumerate(dt_boxes):
-            boximg = self.get_rotate_crop_image(im, dt_boxes[i])
-            img_list.append(boximg)
-            h, w = boximg.shape[0:2]
-            wh_ratio = w * 1.0 / h
-            max_wh_ratio = max(max_wh_ratio, wh_ratio)
-        for img in img_list:
-            norm_img = self.ocr_reader.resize_norm_img(img, max_wh_ratio)
-            feed = {"image": norm_img}
-            feed_list.append(feed)
-        return feed_list
-
-    def postprocess(self, input_dicts, fetch_dict):
-        rec_res = self.ocr_reader.postprocess(fetch_dict, with_score=True)
-        res_lst = []
-        for res in rec_res:
-            res_lst.append(res[0])
-        res = {"res": str(res_lst)}
-        return res
-
-
-read_op = RequestOp()
-det_op = DetOp(
-    name="det",
-    input_ops=[read_op],
-    server_endpoints=["127.0.0.1:12000"],
-    fetch_list=["concat_1.tmp_0"],
-    client_config="ocr_det_client/serving_client_conf.prototxt",
-    concurrency=1)
-rec_op = RecOp(
-    name="rec",
-    input_ops=[det_op],
-    server_endpoints=["127.0.0.1:12001"],
-    fetch_list=["ctc_greedy_decoder_0.tmp_0", "softmax_0.tmp_0"],
-    client_config="ocr_rec_client/serving_client_conf.prototxt",
-    concurrency=1)
-response_op = ResponseOp(input_ops=[rec_op])
-
-server = PipelineServer("ocr")
-server.set_response_op(response_op)
-server.prepare_server('config.yml')
-server.run_server()

python/examples/pipeline/ocr/web_service.py

@@ -14,7 +14,7 @@
 try:
     from paddle_serving_server.web_service import WebService, Op
 except ImportError:
-    from paddle_serving_server.web_service import WebService, Op
+    from paddle_serving_server_gpu.web_service import WebService, Op
 import logging
 import numpy as np
 import cv2
@@ -43,7 +43,7 @@ class DetOp(Op):
             "min_size": 3
         })
 
-    def preprocess(self, input_dicts):
+    def preprocess(self, input_dicts, data_id, log_id):
         (_, input_dict), = input_dicts.items()
         data = base64.b64decode(input_dict["image"].encode('utf8'))
         data = np.fromstring(data, np.uint8)
@@ -52,9 +52,9 @@ class DetOp(Op):
         self.ori_h, self.ori_w, _ = self.im.shape
         det_img = self.det_preprocess(self.im)
         _, self.new_h, self.new_w = det_img.shape
-        return {"image": det_img[np.newaxis, :]}
+        return {"image": det_img[np.newaxis, :].copy()}, False, None, ""
 
-    def postprocess(self, input_dicts, fetch_dict):
+    def postprocess(self, input_dicts, fetch_dict, log_id):
         det_out = fetch_dict["concat_1.tmp_0"]
         ratio_list = [
             float(self.new_h) / self.ori_h, float(self.new_w) / self.ori_w
@@ -63,7 +63,7 @@ class DetOp(Op):
         dt_boxes = self.filter_func(dt_boxes_list[0], [self.ori_h, self.ori_w])
         out_dict = {"dt_boxes": dt_boxes, "image": self.im}
         print("out dict", out_dict)
-        return out_dict
+        return out_dict, None, ""
 
 
 class RecOp(Op):
@@ -72,7 +72,7 @@ class RecOp(Op):
         self.get_rotate_crop_image = GetRotateCropImage()
         self.sorted_boxes = SortedBoxes()
 
-    def preprocess(self, input_dicts):
+    def preprocess(self, input_dicts, data_id, log_id):
         (_, input_dict), = input_dicts.items()
         im = input_dict["image"]
         dt_boxes = input_dict["dt_boxes"]
@@ -93,15 +93,15 @@ class RecOp(Op):
             norm_img = self.ocr_reader.resize_norm_img(img, max_wh_ratio)
             imgs[id] = norm_img
         feed = {"image": imgs.copy()}
-        return feed
+        return feed, False, None, ""
 
-    def postprocess(self, input_dicts, fetch_dict):
+    def postprocess(self, input_dicts, fetch_dict, log_id):
         rec_res = self.ocr_reader.postprocess(fetch_dict, with_score=True)
         res_lst = []
         for res in rec_res:
             res_lst.append(res[0])
         res = {"res": str(res_lst)}
-        return res
+        return res, None, ""
 
 
 class OcrService(WebService):
@@ -112,5 +112,5 @@ class OcrService(WebService):
 
 
 uci_service = OcrService(name="ocr")
-uci_service.prepare_pipeline_config("brpc_config.yml")
+uci_service.prepare_pipeline_config("config.yml")
 uci_service.run_service()

python/examples/pipeline/simple_web_service/README.md

@@ -15,5 +15,5 @@ python web_service.py &>log.txt &
 
 ## Http test
 ```
-curl -X POST -k http://localhost:18080/uci/prediction -d '{"key": ["x"], "value": ["0.0137, -0.1136, 0.2553, -0.0692, 0.0582, -0.0727, -0.1583, -0.0584, 0.6283, 0.4919, 0.1856, 0.0795, -0.0332"]}'
+curl -X POST -k http://localhost:18082/uci/prediction -d '{"key": ["x"], "value": ["0.0137, -0.1136, 0.2553, -0.0692, 0.0582, -0.0727, -0.1583, -0.0584, 0.6283, 0.4919, 0.1856, 0.0795, -0.0332"]}'
 ```

python/examples/pipeline/simple_web_service/README_CN.md

@@ -15,5 +15,5 @@ python web_service.py &>log.txt &
 
 ## 测试
 ```
-curl -X POST -k http://localhost:18080/uci/prediction -d '{"key": ["x"], "value": ["0.0137, -0.1136, 0.2553, -0.0692, 0.0582, -0.0727, -0.1583, -0.0584, 0.6283, 0.4919, 0.1856, 0.0795, -0.0332"]}'
+curl -X POST -k http://localhost:18082/uci/prediction -d '{"key": ["x"], "value": ["0.0137, -0.1136, 0.2553, -0.0692, 0.0582, -0.0727, -0.1583, -0.0584, 0.6283, 0.4919, 0.1856, 0.0795, -0.0332"]}'
 ```

python/examples/pipeline/simple_web_service/config.yml

-worker_num: 4
-http_port: 18080
+#worker_num, 最大并发数。当build_dag_each_worker=True时, 框架会创建worker_num个进程，每个进程内构建grpcSever和DAG
+##当build_dag_each_worker=False时，框架会设置主线程grpc线程池的max_workers=worker_num
+worker_num: 1
+
+#http端口, rpc_port和http_port不允许同时为空。当rpc_port可用且http_port为空时，不自动生成http_port
+http_port: 18082
+
 dag:
-    is_thread_op: false
+    #op资源类型, True, 为线程模型；False，为进程模型
+    is_thread_op: False
 op:
     uci:
+        #当op配置没有server_endpoints时，从local_service_conf读取本地服务配置
         local_service_conf:
+
+            #并发数，is_thread_op=True时，为线程并发；否则为进程并发
+            concurrency: 2
+
+            #uci模型路径
             model_config: uci_housing_model
-            devices: "" # "0,1"
-            client_type: brpc
+
+            #计算硬件ID，当devices为""或不写时为CPU预测；当devices为"0", "0,1,2"时为GPU预测，表示使用的GPU卡
+            devices: "0" # "0,1"
+
+            #client类型，包括brpc, grpc和local_predictor.local_predictor不启动Serving服务，进程内预测
+            client_type: local_predictor
+
+            #Fetch结果列表，以client_config中fetch_var的alias_name为准
+            fetch_list: ["price"] 

python/examples/pipeline/simple_web_service/web_service.py

@@ -25,20 +25,25 @@ class UciOp(Op):
     def init_op(self):
         self.separator = ","
 
-    def preprocess(self, input_dicts):
+    def preprocess(self, input_dicts, data_id, log_id):
         (_, input_dict), = input_dicts.items()
-        _LOGGER.info(input_dict)
+        _LOGGER.error("UciOp::preprocess >>> log_id:{}, input:{}".format(
+            log_id, input_dict))
         x_value = input_dict["x"]
+        proc_dict = {}
         if isinstance(x_value, (str, unicode)):
             input_dict["x"] = np.array(
                 [float(x.strip())
                  for x in x_value.split(self.separator)]).reshape(1, 13)
-        return input_dict
+            _LOGGER.error("input_dict:{}".format(input_dict))
 
-    def postprocess(self, input_dicts, fetch_dict):
-        # _LOGGER.info(fetch_dict)
+        return input_dict, False, None, ""
+
+    def postprocess(self, input_dicts, fetch_dict, log_id):
+        _LOGGER.info("UciOp::postprocess >>> log_id:{}, fetch_dict:{}".format(
+            log_id, fetch_dict))
         fetch_dict["price"] = str(fetch_dict["price"][0][0])
-        return fetch_dict
+        return fetch_dict, None, ""
 
 
 class UciService(WebService):

python/paddle_serving_app/local_predict.py

@@ -32,6 +32,12 @@ logger.setLevel(logging.INFO)
 
 
 class LocalPredictor(object):
+    """
+    Prediction in the current process of the local environment, in process
+    call, Compared with RPC/HTTP, LocalPredictor has better performance, 
+    because of no network and packaging load.
+    """
+
     def __init__(self):
         self.feed_names_ = []
         self.fetch_names_ = []
@@ -42,13 +48,41 @@ class LocalPredictor(object):
         self.fetch_names_to_idx_ = {}
         self.fetch_names_to_type_ = {}
 
-    def load_model_config(self, model_path, gpu=False, profile=True, cpu_num=1):
+    def load_model_config(self,
+                          model_path,
+                          use_gpu=False,
+                          gpu_id=0,
+                          use_profile=False,
+                          thread_num=1,
+                          mem_optim=True,
+                          ir_optim=False,
+                          use_trt=False,
+                          use_feed_fetch_ops=False):
+        """
+        Load model config and set the engine config for the paddle predictor
+   
+        Args:
+            model_path: model config path.
+            use_gpu: calculating with gpu, False default.
+            gpu_id: gpu id, 0 default.
+            use_profile: use predictor profiles, False default.
+            thread_num: thread nums, default 1. 
+            mem_optim: memory optimization, True default.
+            ir_optim: open calculation chart optimization, False default.
+            use_trt: use nvidia TensorRT optimization, False default
+            use_feed_fetch_ops: use feed/fetch ops, False default.
+        """
         client_config = "{}/serving_server_conf.prototxt".format(model_path)
         model_conf = m_config.GeneralModelConfig()
         f = open(client_config, 'r')
         model_conf = google.protobuf.text_format.Merge(
             str(f.read()), model_conf)
         config = AnalysisConfig(model_path)
+        logger.info("load_model_config params: model_path:{}, use_gpu:{},\
+            gpu_id:{}, use_profile:{}, thread_num:{}, mem_optim:{}, ir_optim:{},\
+            use_trt:{}, use_feed_fetch_ops:{}".format(
+            model_path, use_gpu, gpu_id, use_profile, thread_num, mem_optim,
+            ir_optim, use_trt, use_feed_fetch_ops))
 
         self.feed_names_ = [var.alias_name for var in model_conf.feed_var]
         self.fetch_names_ = [var.alias_name for var in model_conf.fetch_var]
@@ -64,19 +98,43 @@ class LocalPredictor(object):
             self.fetch_names_to_idx_[var.alias_name] = i
             self.fetch_names_to_type_[var.alias_name] = var.fetch_type
 
-        if not gpu:
-            config.disable_gpu()
-        else:
-            config.enable_use_gpu(100, 0)
-        if profile:
+        if use_profile:
             config.enable_profile()
+        if mem_optim:
+            config.enable_memory_optim()
+        config.switch_ir_optim(ir_optim)
+        config.set_cpu_math_library_num_threads(thread_num)
+        config.switch_use_feed_fetch_ops(use_feed_fetch_ops)
         config.delete_pass("conv_transpose_eltwiseadd_bn_fuse_pass")
-        config.set_cpu_math_library_num_threads(cpu_num)
-        config.switch_ir_optim(False)
-        config.switch_use_feed_fetch_ops(False)
+
+        if not use_gpu:
+            config.disable_gpu()
+        else:
+            config.enable_use_gpu(100, gpu_id)
+            if use_trt:
+                config.enable_tensorrt_engine(
+                    workspace_size=1 << 20,
+                    max_batch_size=32,
+                    min_subgraph_size=3,
+                    use_static=False,
+                    use_calib_mode=False)
+
         self.predictor = create_paddle_predictor(config)
 
     def predict(self, feed=None, fetch=None, batch=False, log_id=0):
+        """
+        Predict locally
+
+        Args:
+            feed: feed var
+            fetch: fetch var
+            batch: batch data or not, False default.If batch is False, a new
+                   dimension is added to header of the shape[np.newaxis].
+            log_id: for logging
+
+        Returns:
+            fetch_map: dict 
+        """
         if feed is None or fetch is None:
             raise ValueError("You should specify feed and fetch for prediction")
         fetch_list = []

python/paddle_serving_app/reader/__init__.py

@@ -18,5 +18,5 @@ from .image_reader import RCNNPostprocess, SegPostprocess, PadStride, BlazeFaceP
 from .image_reader import DBPostProcess, FilterBoxes, GetRotateCropImage, SortedBoxes
 from .lac_reader import LACReader
 from .senta_reader import SentaReader
-from .imdb_reader import IMDBDataset
+#from .imdb_reader import IMDBDataset
 from .ocr_reader import OCRReader

python/paddle_serving_app/reader/pddet/image_tool.py

@@ -22,18 +22,17 @@ import yaml
 import copy
 import argparse
 import logging
-import paddle.fluid as fluid
 import json
 
 FORMAT = '%(asctime)s-%(levelname)s: %(message)s'
 logging.basicConfig(level=logging.INFO, format=FORMAT)
 logger = logging.getLogger(__name__)
 
-precision_map = {
-    'trt_int8': fluid.core.AnalysisConfig.Precision.Int8,
-    'trt_fp32': fluid.core.AnalysisConfig.Precision.Float32,
-    'trt_fp16': fluid.core.AnalysisConfig.Precision.Half
-}
+#precision_map = {
+#    'trt_int8': fluid.core.AnalysisConfig.Precision.Int8,
+#    'trt_fp32': fluid.core.AnalysisConfig.Precision.Float32,
+#    'trt_fp16': fluid.core.AnalysisConfig.Precision.Half
+#}
 
 
 class Resize(object):

python/paddle_serving_server/web_service.py

@@ -112,13 +112,14 @@ class WebService(object):
         if "fetch" not in request.json:
             abort(400)
         try:
-            feed, fetch = self.preprocess(request.json["feed"],
-                                          request.json["fetch"])
+            feed, fetch, is_batch = self.preprocess(request.json["feed"],
+                                                    request.json["fetch"])
             if isinstance(feed, dict) and "fetch" in feed:
                 del feed["fetch"]
             if len(feed) == 0:
                 raise ValueError("empty input")
-            fetch_map = self.client.predict(feed=feed, fetch=fetch, batch=True)
+            fetch_map = self.client.predict(
+                feed=feed, fetch=fetch, batch=is_batch)
             result = self.postprocess(
                 feed=request.json["feed"], fetch=fetch, fetch_map=fetch_map)
             result = {"result": result}
@@ -188,7 +189,8 @@ class WebService(object):
 
     def preprocess(self, feed=[], fetch=[]):
         print("This API will be deprecated later. Please do not use it")
-        return feed, fetch
+        is_batch = True
+        return feed, fetch, is_batch
 
     def postprocess(self, feed=[], fetch=[], fetch_map=None):
         print("This API will be deprecated later. Please do not use it")

python/paddle_serving_server_gpu/web_service.py

@@ -167,13 +167,14 @@ class WebService(object):
         if "fetch" not in request.json:
             abort(400)
         try:
-            feed, fetch = self.preprocess(request.json["feed"],
-                                          request.json["fetch"])
+            feed, fetch, is_batch = self.preprocess(request.json["feed"],
+                                                    request.json["fetch"])
             if isinstance(feed, dict) and "fetch" in feed:
                 del feed["fetch"]
             if len(feed) == 0:
                 raise ValueError("empty input")
-            fetch_map = self.client.predict(feed=feed, fetch=fetch)
+            fetch_map = self.client.predict(
+                feed=feed, fetch=fetch, batch=is_batch)
             result = self.postprocess(
                 feed=request.json["feed"], fetch=fetch, fetch_map=fetch_map)
             result = {"result": result}
@@ -249,7 +250,8 @@ class WebService(object):
 
     def preprocess(self, feed=[], fetch=[]):
         print("This API will be deprecated later. Please do not use it")
-        return feed, fetch
+        is_batch = True
+        return feed, fetch, is_batch
 
     def postprocess(self, feed=[], fetch=[], fetch_map=None):
         print("This API will be deprecated later. Please do not use it")

python/pipeline/channel.py

@@ -32,7 +32,10 @@ import copy
 _LOGGER = logging.getLogger(__name__)
 
 
-class ChannelDataEcode(enum.Enum):
+class ChannelDataErrcode(enum.Enum):
+    """
+    ChannelData error code
+    """
     OK = 0
     TIMEOUT = 1
     NOT_IMPLEMENTED = 2
@@ -42,9 +45,21 @@ class ChannelDataEcode(enum.Enum):
     CLOSED_ERROR = 6
     NO_SERVICE = 7
     UNKNOW = 8
+    PRODUCT_ERROR = 9
+
+
+class ProductErrCode(enum.Enum):
+    """
+    ProductErrCode is a base class for recording business error code. 
+    product developers inherit this class and extend more error codes. 
+    """
+    pass
 
 
 class ChannelDataType(enum.Enum):
+    """
+    Channel data type
+    """
     DICT = 0
     CHANNEL_NPDATA = 1
     ERROR = 2
@@ -56,20 +71,23 @@ class ChannelData(object):
                  npdata=None,
                  dictdata=None,
                  data_id=None,
-                 ecode=None,
+                 log_id=None,
+                 error_code=None,
                  error_info=None,
+                 prod_error_code=None,
+                 prod_error_info=None,
                  client_need_profile=False):
         '''
         There are several ways to use it:
         
-        1. ChannelData(ChannelDataType.CHANNEL_NPDATA.value, npdata, data_id)
-        2. ChannelData(ChannelDataType.DICT.value, dictdata, data_id)
-        3. ChannelData(ecode, error_info, data_id)
+        1. ChannelData(ChannelDataType.CHANNEL_NPDATA.value, npdata, data_id, log_id)
+        2. ChannelData(ChannelDataType.DICT.value, dictdata, data_id, log_id)
+        3. ChannelData(error_code, error_info, prod_error_code, prod_error_info, data_id, log_id)
 
         Protobufs are not pickle-able:
         https://stackoverflow.com/questions/55344376/how-to-import-protobuf-module
         '''
-        if ecode is not None:
+        if error_code is not None or prod_error_code is not None:
             if data_id is None or error_info is None:
                 _LOGGER.critical("Failed to generate ChannelData: data_id"
                                  " and error_info cannot be None")
@@ -77,25 +95,30 @@ class ChannelData(object):
             datatype = ChannelDataType.ERROR.value
         else:
             if datatype == ChannelDataType.CHANNEL_NPDATA.value:
-                ecode, error_info = ChannelData.check_npdata(npdata)
-                if ecode != ChannelDataEcode.OK.value:
+                error_code, error_info = ChannelData.check_npdata(npdata)
+                if error_code != ChannelDataErrcode.OK.value:
                     datatype = ChannelDataType.ERROR.value
-                    _LOGGER.error("(logid={}) {}".format(data_id, error_info))
+                    _LOGGER.error("(data_id={} log_id={}) {}".format(
+                        data_id, log_id, error_info))
             elif datatype == ChannelDataType.DICT.value:
-                ecode, error_info = ChannelData.check_dictdata(dictdata)
-                if ecode != ChannelDataEcode.OK.value:
+                error_code, error_info = ChannelData.check_dictdata(dictdata)
+                if error_code != ChannelDataErrcode.OK.value:
                     datatype = ChannelDataType.ERROR.value
-                    _LOGGER.error("(logid={}) {}".format(data_id, error_info))
+                    _LOGGER.error("(data_id={} log_id={}) {}".format(
+                        data_id, log_id, error_info))
             else:
-                _LOGGER.critical("(logid={}) datatype not match".format(
-                    data_id))
+                _LOGGER.critical("(data_id={} log_id={}) datatype not match".
+                                 format(data_id, log_id))
                 os._exit(-1)
         self.datatype = datatype
         self.npdata = npdata
         self.dictdata = dictdata
         self.id = data_id
-        self.ecode = ecode
+        self.log_id = log_id
+        self.error_code = error_code
         self.error_info = error_info
+        self.prod_error_code = prod_error_code
+        self.prod_error_info = prod_error_info
         self.client_need_profile = client_need_profile
         self.profile_data_set = set()
 
@@ -106,67 +129,67 @@ class ChannelData(object):
 
     @staticmethod
     def check_dictdata(dictdata):
-        ecode = ChannelDataEcode.OK.value
+        error_code = ChannelDataErrcode.OK.value
         error_info = None
         if isinstance(dictdata, list):
             # batch data
             for sample in dictdata:
                 if not isinstance(sample, dict):
-                    ecode = ChannelDataEcode.TYPE_ERROR.value
+                    error_code = ChannelDataErrcode.TYPE_ERROR.value
                     error_info = "Failed to check data: the type of " \
                             "data must be dict, but get {}.".format(type(sample))
                     break
         elif not isinstance(dictdata, dict):
             # batch size = 1
-            ecode = ChannelDataEcode.TYPE_ERROR.value
+            error_code = ChannelDataErrcode.TYPE_ERROR.value
             error_info = "Failed to check data: the type of data must " \
                     "be dict, but get {}.".format(type(dictdata))
-        return ecode, error_info
+        return error_code, error_info
 
     @staticmethod
     def check_batch_npdata(batch):
-        ecode = ChannelDataEcode.OK.value
+        error_code = ChannelDataErrcode.OK.value
         error_info = None
         for npdata in batch:
-            ecode, error_info = ChannelData.check_npdata(npdata)
-            if ecode != ChannelDataEcode.OK.value:
+            error_code, error_info = ChannelData.check_npdata(npdata)
+            if error_code != ChannelDataErrcode.OK.value:
                 break
-        return ecode, error_info
+        return error_code, error_info
 
     @staticmethod
     def check_npdata(npdata):
-        ecode = ChannelDataEcode.OK.value
+        error_code = ChannelDataErrcode.OK.value
         error_info = None
         if isinstance(npdata, list):
             # batch data
             for sample in npdata:
                 if not isinstance(sample, dict):
-                    ecode = ChannelDataEcode.TYPE_ERROR.value
+                    error_code = ChannelDataErrcode.TYPE_ERROR.value
                     error_info = "Failed to check data: the " \
                             "value of data must be dict, but get {}.".format(
                                     type(sample))
                     break
                 for _, value in sample.items():
                     if not isinstance(value, np.ndarray):
-                        ecode = ChannelDataEcode.TYPE_ERROR.value
+                        error_code = ChannelDataErrcode.TYPE_ERROR.value
                         error_info = "Failed to check data: the" \
                                 " value of data must be np.ndarray, but get {}.".format(
                                         type(value))
-                        return ecode, error_info
+                        return error_code, error_info
         elif isinstance(npdata, dict):
             # batch_size = 1
             for _, value in npdata.items():
                 if not isinstance(value, np.ndarray):
-                    ecode = ChannelDataEcode.TYPE_ERROR.value
+                    error_code = ChannelDataErrcode.TYPE_ERROR.value
                     error_info = "Failed to check data: the value " \
                             "of data must be np.ndarray, but get {}.".format(
                                     type(value))
                     break
         else:
-            ecode = ChannelDataEcode.TYPE_ERROR.value
+            error_code = ChannelDataErrcode.TYPE_ERROR.value
             error_info = "Failed to check data: the value of data " \
                     "must be dict, but get {}.".format(type(npdata))
-        return ecode, error_info
+        return error_code, error_info
 
     def parse(self):
         feed = None
@@ -191,8 +214,9 @@ class ChannelData(object):
             return 1
 
     def __str__(self):
-        return "type[{}], ecode[{}], id[{}]".format(
-            ChannelDataType(self.datatype).name, self.ecode, self.id)
+        return "type[{}], error_code[{}], data_id[{}], log_id[{}], dict_data[{}]".format(
+            ChannelDataType(self.datatype).name, self.error_code, self.id,
+            self.log_id, str(self.dictdata))
 
 
 class ProcessChannel(object):
@@ -289,14 +313,14 @@ class ProcessChannel(object):
 
     def push(self, channeldata, op_name=None):
         _LOGGER.debug(
-            self._log("(logid={}) Op({}) Pushing data".format(channeldata.id,
-                                                              op_name)))
+            self._log("(data_id={} log_id={}) Op({}) Enter channel::push".
+                      format(channeldata.id, channeldata.log_id, op_name)))
         if len(self._producers) == 0:
             _LOGGER.critical(
                 self._log(
-                    "(logid={}) Op({}) Failed to push data: expected number"
+                    "(data_id={} log_id={}) Op({}) Failed to push data: expected number"
                     " of producers to be greater than 0, but the it is 0.".
-                    format(channeldata.id, op_name)))
+                    format(channeldata.id, channeldata.log_id, op_name)))
             os._exit(-1)
         elif len(self._producers) == 1:
             with self._cv:
@@ -310,19 +334,21 @@ class ProcessChannel(object):
                     raise ChannelStopError()
                 self._cv.notify_all()
             _LOGGER.debug(
-                self._log("(logid={}) Op({}) Pushed data into internal queue.".
-                          format(channeldata.id, op_name)))
+                self._log(
+                    "(data_id={} log_id={}) Op({}) Pushed data into internal queue.".
+                    format(channeldata.id, channeldata.log_id, op_name)))
             return True
         elif op_name is None:
             _LOGGER.critical(
                 self._log(
-                    "(logid={}) Op({}) Failed to push data: there are multiple "
+                    "(data_id={} log_id={}) Op({}) Failed to push data: there are multiple "
                     "producers, so op_name cannot be None.".format(
-                        channeldata.id, op_name)))
+                        channeldata.id, channeldata.log_id, op_name)))
             os._exit(-1)
 
         producer_num = len(self._producers)
         data_id = channeldata.id
+        log_id = channeldata.log_id
         put_data = None
         with self._cv:
             if data_id not in self._input_buf:
@@ -347,8 +373,8 @@ class ProcessChannel(object):
             if put_data is None:
                 _LOGGER.debug(
                     self._log(
-                        "(logid={}) Op({}) Pushed data into input_buffer.".
-                        format(data_id, op_name)))
+                        "(data_id={} log_id={}) Op({}) Pushed data into input_buffer.".
+                        format(data_id, log_id, op_name)))
             else:
                 while self._stop.value == 0:
                     try:
@@ -361,8 +387,8 @@ class ProcessChannel(object):
 
                 _LOGGER.debug(
                     self._log(
-                        "(logid={}) Op({}) Pushed data into internal_queue.".
-                        format(data_id, op_name)))
+                        "(data_id={} log_id={}) Op({}) Pushed data into internal_queue.".
+                        format(data_id, log_id, op_name)))
             self._cv.notify_all()
         return True
 
@@ -404,8 +430,8 @@ class ProcessChannel(object):
                 if self._stop.value == 1:
                     raise ChannelStopError()
             _LOGGER.debug(
-                self._log("(logid={}) Op({}) Got data".format(resp.values()[0]
-                                                              .id, op_name)))
+                self._log("(data_id={} log_id={}) Op({}) Got data".format(
+                    resp.values()[0].id, resp.values()[0].log_id, op_name)))
             return resp
         elif op_name is None:
             _LOGGER.critical(
@@ -434,8 +460,9 @@ class ProcessChannel(object):
                     self._output_buf.append(channeldata)
                     _LOGGER.debug(
                         self._log(
-                            "(logid={}) Op({}) Pop ready item into output_buffer".
-                            format(channeldata.values()[0].id, op_name)))
+                            "(data_id={} log_id={}) Op({}) Pop ready item into output_buffer".
+                            format(channeldata.values()[0].id,
+                                   channeldata.values()[0].log_id, op_name)))
                     break
                 except Queue.Empty:
                     if timeout is not None:
@@ -487,8 +514,9 @@ class ProcessChannel(object):
             self._cv.notify_all()
 
         _LOGGER.debug(
-            self._log("(logid={}) Op({}) Got data from output_buffer".format(
-                resp.values()[0].id, op_name)))
+            self._log(
+                "(data_id={} log_id={}) Op({}) Got data from output_buffer".
+                format(resp.values()[0].id, resp.values()[0].log_id, op_name)))
         return resp
 
     def stop(self):
@@ -586,14 +614,14 @@ class ThreadChannel(Queue.PriorityQueue):
 
     def push(self, channeldata, op_name=None):
         _LOGGER.debug(
-            self._log("(logid={}) Op({}) Pushing data".format(channeldata.id,
-                                                              op_name)))
+            self._log("(data_id={} log_id={}) Op({}) Pushing data".format(
+                channeldata.id, channeldata.log_id, op_name)))
         if len(self._producers) == 0:
             _LOGGER.critical(
                 self._log(
-                    "(logid={}) Op({}) Failed to push data: expected number of "
+                    "(data_id={} log_id={}) Op({}) Failed to push data: expected number of "
                     "producers to be greater than 0, but the it is 0.".format(
-                        channeldata.id, op_name)))
+                        channeldata.id, channeldata.log_id, op_name)))
             os._exit(-1)
         elif len(self._producers) == 1:
             with self._cv:
@@ -607,19 +635,21 @@ class ThreadChannel(Queue.PriorityQueue):
                     raise ChannelStopError()
                 self._cv.notify_all()
             _LOGGER.debug(
-                self._log("(logid={}) Op({}) Pushed data into internal_queue.".
-                          format(channeldata.id, op_name)))
+                self._log(
+                    "(data_id={} log_id={}) Op({}) Pushed data into internal_queue.".
+                    format(channeldata.id, channeldata.log_id, op_name)))
             return True
         elif op_name is None:
             _LOGGER.critical(
                 self._log(
-                    "(logid={}) Op({}) Failed to push data: there are multiple"
+                    "(data_id={} log_id={}) Op({}) Failed to push data: there are multiple"
                     " producers, so op_name cannot be None.".format(
-                        channeldata.id, op_name)))
+                        channeldata.id, channeldata.log_id, op_name)))
             os._exit(-1)
 
         producer_num = len(self._producers)
         data_id = channeldata.id
+        log_id = channeldata.log_id
         put_data = None
         with self._cv:
             if data_id not in self._input_buf:
@@ -639,8 +669,8 @@ class ThreadChannel(Queue.PriorityQueue):
             if put_data is None:
                 _LOGGER.debug(
                     self._log(
-                        "(logid={}) Op({}) Pushed data into input_buffer.".
-                        format(data_id, op_name)))
+                        "(data_id={} log_id={}) Op({}) Pushed data into input_buffer.".
+                        format(data_id, log_id, op_name)))
             else:
                 while self._stop is False:
                     try:
@@ -653,8 +683,8 @@ class ThreadChannel(Queue.PriorityQueue):
 
                 _LOGGER.debug(
                     self._log(
-                        "(logid={}) Op({}) Pushed data into internal_queue.".
-                        format(data_id, op_name)))
+                        "(data_id={} log_id={}) Op({}) Pushed data into internal_queue.".
+                        format(data_id, log_id, op_name)))
             self._cv.notify_all()
         return True
 
@@ -697,8 +727,8 @@ class ThreadChannel(Queue.PriorityQueue):
                 if self._stop:
                     raise ChannelStopError()
             _LOGGER.debug(
-                self._log("(logid={}) Op({}) Got data".format(resp.values()[0]
-                                                              .id, op_name)))
+                self._log("(data_id={} log_id={}) Op({}) Got data".format(
+                    resp.values()[0].id, resp.values()[0].log_id, op_name)))
             return resp
         elif op_name is None:
             _LOGGER.critical(
@@ -727,8 +757,9 @@ class ThreadChannel(Queue.PriorityQueue):
                     self._output_buf.append(channeldata)
                     _LOGGER.debug(
                         self._log(
-                            "(logid={}) Op({}) Pop ready item into output_buffer".
-                            format(channeldata.values()[0].id, op_name)))
+                            "(data_id={} log_id={}) Op({}) Pop ready item into output_buffer".
+                            format(channeldata.values()[0].id,
+                                   channeldata.values()[0].log_id, op_name)))
                     break
                 except Queue.Empty:
                     if timeout is not None:
@@ -780,8 +811,9 @@ class ThreadChannel(Queue.PriorityQueue):
             self._cv.notify_all()
 
         _LOGGER.debug(
-            self._log("(logid={}) Op({}) Got data from output_buffer".format(
-                resp.values()[0].id, op_name)))
+            self._log(
+                "(data_id={} log_id={}) Op({}) Got data from output_buffer".
+                format(resp.values()[0].id, resp.values()[0].log_id, op_name)))
         return resp
 
     def stop(self):

python/pipeline/dag.py

@@ -25,10 +25,12 @@ else:
 import os
 import logging
 import collections
+import json
 
 from .operator import Op, RequestOp, ResponseOp, VirtualOp
 from .channel import (ThreadChannel, ProcessChannel, ChannelData,
-                      ChannelDataEcode, ChannelDataType, ChannelStopError)
+                      ChannelDataErrcode, ChannelDataType, ChannelStopError,
+                      ProductErrCode)
 from .profiler import TimeProfiler, PerformanceTracer
 from .util import NameGenerator, ThreadIdGenerator, PipelineProcSyncManager
 from .proto import pipeline_service_pb2
@@ -37,7 +39,23 @@ _LOGGER = logging.getLogger(__name__)
 
 
 class DAGExecutor(object):
+    """
+    DAG Executor, the service entrance of DAG.
+    """
+
     def __init__(self, response_op, server_conf, worker_idx):
+        """
+        Initialize DAGExecutor.
+
+        Args:
+            response_op: Response OP
+            server_conf: server conf. config.yaml
+            worker_idx: DAGExecutor index, PipelineServer creates many
+                DAGExecutors when _build_dag_each_worker is true.
+
+        Returns:
+            None.
+        """
         build_dag_each_worker = server_conf["build_dag_each_worker"]
         server_worker_num = server_conf["worker_num"]
         dag_conf = server_conf["dag"]
@@ -74,7 +92,9 @@ class DAGExecutor(object):
         if self._tracer is not None:
             self._tracer.start()
 
-        # generate id: data_id == request_id == log_id
+        # generate id 
+        # data_id: Server Unique ID, automatically generated by the framework
+        # log_id: Trace one product request, can be empty, not unique.
         base_counter = 0
         gen_id_step = 1
         if build_dag_each_worker:
@@ -94,6 +114,15 @@ class DAGExecutor(object):
         self._client_profile_value = "1"
 
     def start(self):
+        """
+        Starting one thread for receiving data from the last channel background.
+
+        Args:
+            None
+
+        Returns:
+            None
+        """
         self._recive_func = threading.Thread(
             target=DAGExecutor._recive_out_channel_func, args=(self, ))
         self._recive_func.daemon = True
@@ -101,11 +130,30 @@ class DAGExecutor(object):
         _LOGGER.debug("[DAG Executor] Start recive thread")
 
     def stop(self):
+        """
+        Stopping DAG
+
+        Args:
+            None
+
+        Returns:
+            None
+        """
         self._dag.stop()
         self._dag.join()
         _LOGGER.info("[DAG Executor] Stop")
 
     def _get_next_data_id(self):
+        """
+        Generate data_id incrementally and Uniquely
+   
+        Args:
+            None
+
+        Returns:
+            data_id: uniq id
+            cond_v: condition variable
+        """
         data_id = self._id_generator.next()
         cond_v = threading.Condition()
         with self._cv_for_cv_pool:
@@ -114,6 +162,15 @@ class DAGExecutor(object):
         return data_id, cond_v
 
     def _set_in_channel(self, in_channel):
+        """
+        Set in_channel of DAG
+
+        Args:
+            in_channel: input channel of DAG
+
+        Returns:
+            None 
+        """
         if not isinstance(in_channel, (ThreadChannel, ProcessChannel)):
             _LOGGER.critical("[DAG Executor] Failed to set in_channel: "
                              "in_channel must be Channel type, but get {}".
@@ -121,8 +178,18 @@ class DAGExecutor(object):
             os._exit(-1)
         in_channel.add_producer(self.name)
         self._in_channel = in_channel
+        _LOGGER.info("[DAG] set in channel succ, name [{}]".format(self.name))
 
     def _set_out_channel(self, out_channel):
+        """
+        Set out_channel of DAG
+
+        Args:
+            out_channel: output channel of DAG
+
+        Returns:
+            None 
+        """
         if not isinstance(out_channel, (ThreadChannel, ProcessChannel)):
             _LOGGER.critical("[DAG Executor] Failed to set out_channel: "
                              "must be Channel type, but get {}".format(
@@ -132,6 +199,17 @@ class DAGExecutor(object):
         self._out_channel = out_channel
 
     def _recive_out_channel_func(self):
+        """
+        Receiving data from the output channel, and pushing data into 
+        _fetch_buffer. Function _get_channeldata_from_fetch_buffer gets 
+        data by retry time.
+
+        Args:
+            None
+
+        Returns:
+            None
+        """
         cv = None
         while True:
             try:
@@ -141,14 +219,13 @@ class DAGExecutor(object):
                 with self._cv_for_cv_pool:
                     for data_id, cv in self._cv_pool.items():
                         closed_errror_data = ChannelData(
-                            ecode=ChannelDataEcode.CLOSED_ERROR.value,
+                            error_code=ChannelDataErrcode.CLOSED_ERROR.value,
                             error_info="dag closed.",
                             data_id=data_id)
                         with cv:
                             self._fetch_buffer[data_id] = closed_errror_data
                             cv.notify_all()
                 break
-
             if len(channeldata_dict) != 1:
                 _LOGGER.critical(
                     "[DAG Executor] Failed to fetch result: out_channel "
@@ -172,6 +249,16 @@ class DAGExecutor(object):
                 cond_v.notify_all()
 
     def _get_channeldata_from_fetch_buffer(self, data_id, cond_v):
+        """
+        Getting the channel data from _fetch_buffer.
+
+        Args:
+            data_id: search key
+            cond_v: conditional variable
+
+        Returns:
+            ready_data: one channel data processed
+        """
         ready_data = None
 
         with cond_v:
@@ -188,45 +275,82 @@ class DAGExecutor(object):
                     ready_data = self._fetch_buffer[data_id]
                     self._cv_pool.pop(data_id)
                     self._fetch_buffer.pop(data_id)
-        _LOGGER.debug("(logid={}) [resp thread] Got data".format(data_id))
+        _LOGGER.debug("(data_id={}) [resp thread] Got data".format(data_id))
         return ready_data
 
     def _pack_channeldata(self, rpc_request, data_id):
+        """
+        Unpacking data from RPC request. and creating one channelData.
+
+        Args:
+           rpc_request: one RPC request
+           data_id: data id, unique
+
+        Returns:
+            ChannelData: one channel data to be processed
+        """
         dictdata = None
+        log_id = None
         try:
-            dictdata = self._unpack_rpc_func(rpc_request)
+            dictdata, log_id, prod_errcode, prod_errinfo = self._unpack_rpc_func(
+                rpc_request)
         except Exception as e:
             _LOGGER.error(
                 "(logid={}) Failed to parse RPC request package: {}"
                 .format(data_id, e),
                 exc_info=True)
             return ChannelData(
-                ecode=ChannelDataEcode.RPC_PACKAGE_ERROR.value,
+                error_code=ChannelDataErrcode.RPC_PACKAGE_ERROR.value,
                 error_info="rpc package error: {}".format(e),
-                data_id=data_id)
+                data_id=data_id,
+                log_id=log_id)
         else:
-            # because unpack_rpc_func is rewritten by user, we need
-            # to look for client_profile_key field in rpc_request
+            # because unpack_rpc_func is rewritten by user, we need to look
+            # for product_errcode in returns, and  client_profile_key field
+            # in rpc_request
+            if prod_errcode is not None:
+                # product errors occured
+                _LOGGER.error("unpack_rpc_func prod_errcode:{}".format(
+                    prod_errcode))
+                return ChannelData(
+                    error_code=ChannelDataErrcode.PRODUCT_ERROR.value,
+                    error_info="",
+                    prod_error_code=prod_errcode,
+                    prod_error_info=prod_errinfo,
+                    data_id=data_id,
+                    log_id=log_id)
+
             profile_value = None
-            for idx, key in enumerate(rpc_request.key):
-                if key == self._client_profile_key:
-                    profile_value = rpc_request.value[idx]
-                    break
+            profile_value = dictdata.get(self._client_profile_key)
             client_need_profile = (profile_value == self._client_profile_value)
-            _LOGGER.debug("(logid={}) Need profile in client: {}".format(
-                data_id, client_need_profile))
             return ChannelData(
                 datatype=ChannelDataType.DICT.value,
                 dictdata=dictdata,
                 data_id=data_id,
+                log_id=log_id,
                 client_need_profile=client_need_profile)
 
     def call(self, rpc_request):
+        """
+        DAGExcutor enterance function. There are 5 steps:
+        1._get_next_data_id: Generate an incremental ID
+        2._pack_channeldata: pack the channel data from request.
+        3.retry loop: 
+            a. push channel_data into _in_channel
+            b. get_channeldata_from_fetch_buffer: get results.
+        4._pack_for_rpc_resp: pack RPC responses
+        5.profile: generte profile string and pack into response.
+
+        Args:
+            rpc_request: one RPC request
+   
+        Returns:
+            rpc_resp: one RPC response
+        """
         if self._tracer is not None:
             trace_buffer = self._tracer.data_buffer()
 
         data_id, cond_v = self._get_next_data_id()
-        _LOGGER.info("(logid={}) Succ generate id".format(data_id))
 
         start_call, end_call = None, None
         if not self._is_thread_op:
@@ -235,45 +359,64 @@ class DAGExecutor(object):
         else:
             start_call = self._profiler.record("call_{}#DAG_0".format(data_id))
 
-        _LOGGER.debug("(logid={}) Parsing RPC request package".format(data_id))
         self._profiler.record("prepack_{}#{}_0".format(data_id, self.name))
         req_channeldata = self._pack_channeldata(rpc_request, data_id)
         self._profiler.record("prepack_{}#{}_1".format(data_id, self.name))
 
+        log_id = req_channeldata.log_id
+        _LOGGER.info("(data_id={} log_id={}) Succ Generate ID ".format(data_id,
+                                                                       log_id))
+
         resp_channeldata = None
         for i in range(self._retry):
-            _LOGGER.debug("(logid={}) Pushing data into Graph engine".format(
+            _LOGGER.debug("(data_id={}) Pushing data into Graph engine".format(
                 data_id))
             try:
+                if req_channeldata is None:
+                    _LOGGER.critical(
+                        "(data_id={} log_id={}) req_channeldata is None"
+                        .format(data_id, log_id))
+                if not isinstance(self._in_channel,
+                                  (ThreadChannel, ProcessChannel)):
+                    _LOGGER.critical(
+                        "(data_id={} log_id={})[DAG Executor] Failed to "
+                        "set in_channel: in_channel must be Channel type, but get {}".
+                        format(data_id, log_id, type(self._in_channel)))
                 self._in_channel.push(req_channeldata, self.name)
             except ChannelStopError:
-                _LOGGER.debug("[DAG Executor] Stop")
+                _LOGGER.error("(data_id:{} log_id={})[DAG Executor] Stop".
+                              format(data_id, log_id))
                 with self._cv_for_cv_pool:
                     self._cv_pool.pop(data_id)
                 return self._pack_for_rpc_resp(
                     ChannelData(
-                        ecode=ChannelDataEcode.CLOSED_ERROR.value,
+                        error_code=ChannelDataErrcode.CLOSED_ERROR.value,
                         error_info="dag closed.",
                         data_id=data_id))
 
-            _LOGGER.debug("(logid={}) Wait for Graph engine...".format(data_id))
+            _LOGGER.debug("(data_id={} log_id={}) Wait for Graph engine...".
+                          format(data_id, log_id))
             resp_channeldata = self._get_channeldata_from_fetch_buffer(data_id,
                                                                        cond_v)
 
-            if resp_channeldata.ecode == ChannelDataEcode.OK.value:
-                _LOGGER.info("(logid={}) Succ predict".format(data_id))
+            if resp_channeldata.error_code == ChannelDataErrcode.OK.value:
+                _LOGGER.info("(data_id={} log_id={}) Succ predict".format(
+                    data_id, log_id))
                 break
             else:
-                _LOGGER.error("(logid={}) Failed to predict: {}"
-                              .format(data_id, resp_channeldata.error_info))
-                if resp_channeldata.ecode != ChannelDataEcode.TIMEOUT.value:
+                _LOGGER.error("(data_id={} log_id={}) Failed to predict: {}"
+                              .format(data_id, log_id,
+                                      resp_channeldata.error_info))
+                if resp_channeldata.error_code != ChannelDataErrcode.TIMEOUT.value:
                     break
 
             if i + 1 < self._retry:
-                _LOGGER.warning("(logid={}) DAGExecutor retry({}/{})".format(
-                    data_id, i + 1, self._retry))
+                _LOGGER.warning(
+                    "(data_id={} log_id={}) DAGExecutor retry({}/{})"
+                    .format(data_id, log_id, i + 1, self._retry))
 
-        _LOGGER.debug("(logid={}) Packing RPC response package".format(data_id))
+        _LOGGER.debug("(data_id={} log_id={}) Packing RPC response package"
+                      .format(data_id, log_id))
         self._profiler.record("postpack_{}#{}_0".format(data_id, self.name))
         rpc_resp = self._pack_for_rpc_resp(resp_channeldata)
         self._profiler.record("postpack_{}#{}_1".format(data_id, self.name))
@@ -287,7 +430,8 @@ class DAGExecutor(object):
             trace_buffer.put({
                 "name": "DAG",
                 "id": data_id,
-                "succ": resp_channeldata.ecode == ChannelDataEcode.OK.value,
+                "succ":
+                resp_channeldata.error_code == ChannelDataErrcode.OK.value,
                 "actions": {
                     "call_{}".format(data_id): end_call - start_call,
                 },
@@ -308,6 +452,15 @@ class DAGExecutor(object):
         return rpc_resp
 
     def _pack_for_rpc_resp(self, channeldata):
+        """
+        Packing one RPC response
+
+        Args:
+            channeldata: one channel data to be packed
+
+        Returns:
+            resp: one RPC response
+        """
         try:
             return self._pack_rpc_func(channeldata)
         except Exception as e:
@@ -316,12 +469,16 @@ class DAGExecutor(object):
                 .format(channeldata.id, e),
                 exc_info=True)
             resp = pipeline_service_pb2.Response()
-            resp.ecode = ChannelDataEcode.RPC_PACKAGE_ERROR.value
-            resp.error_info = "rpc package error: {}".format(e)
+            resp.err_no = ChannelDataErrcode.RPC_PACKAGE_ERROR.value
+            resp.err_msg = "rpc package error: {}".format(e)
             return resp
 
 
 class DAG(object):
+    """
+    Directed Acyclic Graph(DAG) engine, builds one DAG topology.
+    """
+
     def __init__(self, request_name, response_op, use_profile, is_thread_op,
                  channel_size, build_dag_each_worker, tracer):
         self._request_name = request_name
@@ -337,6 +494,18 @@ class DAG(object):
 
     @staticmethod
     def get_use_ops(response_op):
+        """
+        Starting from ResponseOp, recursively traverse the front OPs. Getting
+        all used ops and the post op list of each op (excluding ResponseOp)
+
+        Args:
+            response_op: ResponseOp
+
+        Returns:
+            used_ops: used ops, set
+            succ_ops_of_use_op: op and the next op list, dict.
+            
+        """
         unique_names = set()
         used_ops = set()
         succ_ops_of_use_op = {}  # {op_name: succ_ops}
@@ -362,6 +531,15 @@ class DAG(object):
         return used_ops, succ_ops_of_use_op
 
     def _gen_channel(self, name_gen):
+        """
+        Generate one ThreadChannel or ProcessChannel.
+
+        Args:
+            name_gen: channel name
+
+        Returns:
+            channel: one channel generated
+        """
         channel = None
         if self._is_thread_op:
             channel = ThreadChannel(
@@ -373,11 +551,37 @@ class DAG(object):
         return channel
 
     def _gen_virtual_op(self, name_gen):
+        """
+        Generate one virtual Op
+
+        Args:
+            name_gen: Op name
+
+        Returns:
+            vir_op: one virtual Op object.
+        """
         vir_op = VirtualOp(name=name_gen.next())
         _LOGGER.debug("[DAG] Generate virtual_op: {}".format(vir_op.name))
         return vir_op
 
     def _topo_sort(self, used_ops, response_op, out_degree_ops):
+        """
+        Topological sort of DAG, creates inverted multi-layers views.
+
+        Args:
+            used_ops: op used in DAG
+            response_op: response op
+            out_degree_ops: Next op list for each op, dict. the output of 
+                get_use_ops()
+
+        Returns:
+            dag_views: the inverted hierarchical topology list. examples:
+                DAG :[A -> B -> C -> E]
+                            \-> D /
+                dag_views: [[E], [C, D], [B], [A]]
+                         
+            last_op:the last op front of ResponseOp
+        """
         out_degree_num = {
             name: len(ops)
             for name, ops in out_degree_ops.items()
@@ -421,6 +625,23 @@ class DAG(object):
         return dag_views, last_op
 
     def _build_dag(self, response_op):
+        """
+        Building DAG, the most important function in class DAG. Core steps:
+        1.get_use_ops: Getting used ops, and out degree op list for each op.
+        2._topo_sort: Topological sort creates inverted multi-layers views.
+        3.create channels and virtual ops.
+
+        Args:
+            response_op: ResponseOp
+
+        Returns:
+            actual_ops: all OPs used in DAG, including virtual OPs
+            channels: all channels used in DAG 
+            input_channel: the channel of first OP 
+            output_channel: the channel of last OP
+            pack_func: pack_response_package function of response_op
+            unpack_func: unpack_request_package function of request_op
+        """
         if response_op is None:
             _LOGGER.critical("Failed to build DAG: ResponseOp"
                              " has not been set.")
@@ -546,6 +767,18 @@ class DAG(object):
         return self._channels
 
     def build(self):
+        """
+        Interface for building one DAG outside.
+
+        Args:
+            None
+
+        Returns:
+            _input_channel: the channel of first OP
+            _output_channel:  the channel of last OP
+            _pack_func: pack_response_package function of response_op
+            _unpack_func: unpack_request_package function of request_op
+        """
         (actual_ops, channels, input_channel, output_channel, pack_func,
          unpack_func) = self._build_dag(self._response_op)
         _LOGGER.info("[DAG] Succ build DAG")
@@ -563,6 +796,15 @@ class DAG(object):
         return self._input_channel, self._output_channel, self._pack_func, self._unpack_func
 
     def start(self):
+        """
+        Each OP starts a thread or process by _is_thread_op 
+
+        Args:
+            None
+
+        Returns:
+            _threads_or_proces: threads or process list.
+        """
         self._threads_or_proces = []
         for op in self._actual_ops:
             op.use_profiler(self._use_profile)
@@ -577,11 +819,29 @@ class DAG(object):
         return self._threads_or_proces
 
     def join(self):
+        """
+        All threads or processes join.
+
+        Args:
+            None
+
+        Returns:
+            None
+        """
         for x in self._threads_or_proces:
             if x is not None:
                 x.join()
 
     def stop(self):
+        """
+        Stopping and cleanning all channels.
+
+        Args:
+            None
+
+        Returns:
+            None 
+        """
         for chl in self._channels:
             chl.stop()
         for op in self._actual_ops:

python/pipeline/gateway/proto/gateway.proto

@@ -19,22 +19,25 @@ option go_package = ".;pipeline_serving";
 import "google/api/annotations.proto";
 
 message Response {
-  repeated string key = 1;
-  repeated string value = 2;
-  int32 ecode = 3;
-  string error_info = 4;
+  int32 err_no = 1;
+  string err_msg = 2;
+  repeated string key = 3;
+  repeated string value = 4;
 };
 
 message Request {
   repeated string key = 1;
   repeated string value = 2;
   string name = 3;
-}
+  string method = 4;
+  int64 logid = 5;
+  string clientip = 6;
+};
 
 service PipelineService {
   rpc inference(Request) returns (Response) {
     option (google.api.http) = {
-      post : "/{name=*}/prediction"
+      post : "/{name=*}/{method=*}"
       body : "*"
     };
   }

python/pipeline/gateway/proxy_server.go

@@ -38,7 +38,8 @@ func run_proxy_server(grpc_port int, http_port int) error {
   ctx, cancel := context.WithCancel(ctx)
   defer cancel()
 
-  mux := runtime.NewServeMux()
+  //EmitDefaults=true, does not filter out the default inputs 
+  mux := runtime.NewServeMux(runtime.WithMarshalerOption(runtime.MIMEWildcard, &runtime.JSONPb{OrigName: true, EmitDefaults: true}))
   opts := []grpc.DialOption{grpc.WithInsecure()}
   err := gw.RegisterPipelineServiceHandlerFromEndpoint(ctx, mux, *pipelineEndpoint, opts)
   if err != nil {

python/pipeline/local_service_handler.py

@@ -15,111 +15,203 @@
 import os
 import logging
 import multiprocessing
-try:
-    from paddle_serving_server_gpu import OpMaker, OpSeqMaker, Server
-    PACKAGE_VERSION = "GPU"
-except ImportError:
-    from paddle_serving_server import OpMaker, OpSeqMaker, Server
-    PACKAGE_VERSION = "CPU"
+#from paddle_serving_server_gpu import OpMaker, OpSeqMaker
+#from paddle_serving_server_gpu import Server as GpuServer
+#from paddle_serving_server import Server as CpuServer
 from . import util
-from paddle_serving_app.local_predict import LocalPredictor
+#from paddle_serving_app.local_predict import LocalPredictor
 
 _LOGGER = logging.getLogger(__name__)
 _workdir_name_gen = util.NameGenerator("workdir_")
 
 
 class LocalServiceHandler(object):
+    """
+    LocalServiceHandler is the processor of the local service, contains
+    three client types, brpc, grpc and local_predictor.If you use the 
+    brpc or grpc, serveing startup ability is provided.If you use
+    local_predictor, local predict ability is provided by paddle_serving_app.
+    """
+
     def __init__(self,
                  model_config,
                  client_type='local_predictor',
                  workdir="",
                  thread_num=2,
                  devices="",
+                 fetch_names=None,
                  mem_optim=True,
                  ir_optim=False,
-                 available_port_generator=None):
+                 available_port_generator=None,
+                 use_trt=False,
+                 use_profile=False):
+        """
+        Initialization of localservicehandler
+
+        Args:
+           model_config: model config path
+           client_type: brpc, grpc and local_predictor[default]
+           workdir: work directory
+           thread_num: number of threads, concurrent quantity.
+           devices: gpu id list[gpu], "" default[cpu]
+           fetch_names: get fetch names out of LocalServiceHandler in 
+               local_predictor mode. fetch_names_ is compatible for Client().
+           mem_optim: use memory/graphics memory optimization, True default.
+           ir_optim: use calculation chart optimization, False default.
+           available_port_generator: generate available ports
+           use_trt: use nvidia tensorRt engine, False default.
+           use_profile: use profiling, False default.
+
+        Returns:
+           None
+        """
         if available_port_generator is None:
             available_port_generator = util.GetAvailablePortGenerator()
 
         self._model_config = model_config
         self._port_list = []
+        self._device_type = "cpu"
         if devices == "":
             # cpu
             devices = [-1]
+            self._device_type = "cpu"
             self._port_list.append(available_port_generator.next())
             _LOGGER.info("Model({}) will be launch in cpu device. Port({})"
                          .format(model_config, self._port_list))
         else:
             # gpu
-            if PACKAGE_VERSION == "CPU":
-                raise ValueError(
-                    "You are using the CPU version package("
-                    "paddle-serving-server), unable to set devices")
+            self._device_type = "gpu"
             devices = [int(x) for x in devices.split(",")]
             for _ in devices:
                 self._port_list.append(available_port_generator.next())
             _LOGGER.info("Model({}) will be launch in gpu device: {}. Port({})"
                          .format(model_config, devices, self._port_list))
-        self.client_type = client_type
+        self._client_type = client_type
         self._workdir = workdir
         self._devices = devices
         self._thread_num = thread_num
         self._mem_optim = mem_optim
         self._ir_optim = ir_optim
-        self.local_predictor_client = None
+        self._local_predictor_client = None
         self._rpc_service_list = []
         self._server_pros = []
-        self._fetch_vars = None
+        self._use_trt = use_trt
+        self._use_profile = use_profile
+        self.fetch_names_ = fetch_names
 
     def get_fetch_list(self):
-        return self._fetch_vars
+        return self.fetch_names_
 
     def get_port_list(self):
         return self._port_list
 
-    def get_client(self):  # for local_predictor_only
-        if self.local_predictor_client is None:
-            self.local_predictor_client = LocalPredictor()
-            self.local_predictor_client.load_model_config(
-                "{}".format(self._model_config), gpu=False, profile=False)
-        return self.local_predictor_client
+    def get_client(self):
+        """
+        Function get_client is only used for local predictor case, creates one
+        LocalPredictor object, and initializes the paddle predictor by function
+        load_model_config.
+
+        Args:
+            None
+
+        Returns:
+            _local_predictor_client
+        """
+        from paddle_serving_app.local_predict import LocalPredictor
+        if self._local_predictor_client is None:
+            self._local_predictor_client = LocalPredictor()
+            use_gpu = False
+            if self._device_type == "gpu":
+                use_gpu = True
+            self._local_predictor_client.load_model_config(
+                model_path=self._model_config,
+                use_gpu=use_gpu,
+                gpu_id=self._devices[0],
+                use_profile=self._use_profile,
+                thread_num=self._thread_num,
+                mem_optim=self._mem_optim,
+                ir_optim=self._ir_optim,
+                use_trt=self._use_trt)
+        return self._local_predictor_client
 
     def get_client_config(self):
         return os.path.join(self._model_config, "serving_server_conf.prototxt")
 
     def _prepare_one_server(self, workdir, port, gpuid, thread_num, mem_optim,
                             ir_optim):
-        device = "gpu"
-        if gpuid == -1:
-            device = "cpu"
-        op_maker = OpMaker()
-        read_op = op_maker.create('general_reader')
-        general_infer_op = op_maker.create('general_infer')
-        general_response_op = op_maker.create('general_response')
-
-        op_seq_maker = OpSeqMaker()
-        op_seq_maker.add_op(read_op)
-        op_seq_maker.add_op(general_infer_op)
-        op_seq_maker.add_op(general_response_op)
-
-        server = Server()
+        """
+        According to _device_type, generating one CpuServer or GpuServer, and
+        setting the model config amd startup params.
+
+        Args:
+            workdir: work directory
+            port: network port
+            gpuid: gpu id
+            thread_num: thread num
+            mem_optim: use memory/graphics memory optimization
+            ir_optim: use calculation chart optimization
+
+        Returns:
+            server: CpuServer/GpuServer
+        """
+        if self._device_type == "cpu":
+            from paddle_serving_server import OpMaker, OpSeqMaker, Server
+            op_maker = OpMaker()
+            read_op = op_maker.create('general_reader')
+            general_infer_op = op_maker.create('general_infer')
+            general_response_op = op_maker.create('general_response')
+
+            op_seq_maker = OpSeqMaker()
+            op_seq_maker.add_op(read_op)
+            op_seq_maker.add_op(general_infer_op)
+            op_seq_maker.add_op(general_response_op)
+
+            server = Server()
+        else:
+            #gpu
+            from paddle_serving_server_gpu import OpMaker, OpSeqMaker, Server
+            op_maker = OpMaker()
+            read_op = op_maker.create('general_reader')
+            general_infer_op = op_maker.create('general_infer')
+            general_response_op = op_maker.create('general_response')
+
+            op_seq_maker = OpSeqMaker()
+            op_seq_maker.add_op(read_op)
+            op_seq_maker.add_op(general_infer_op)
+            op_seq_maker.add_op(general_response_op)
+
+            server = Server()
+            if gpuid >= 0:
+                server.set_gpuid(gpuid)
+
         server.set_op_sequence(op_seq_maker.get_op_sequence())
         server.set_num_threads(thread_num)
         server.set_memory_optimize(mem_optim)
         server.set_ir_optimize(ir_optim)
 
         server.load_model_config(self._model_config)
-        if gpuid >= 0:
-            server.set_gpuid(gpuid)
-        server.prepare_server(workdir=workdir, port=port, device=device)
-        if self._fetch_vars is None:
-            self._fetch_vars = server.get_fetch_list()
+        server.prepare_server(
+            workdir=workdir, port=port, device=self._device_type)
+        if self.fetch_names_ is None:
+            self.fetch_names_ = server.get_fetch_list()
         return server
 
     def _start_one_server(self, service_idx):
+        """
+        Start one server
+     
+        Args:
+            service_idx: server index
+ 
+        Returns:
+            None
+        """
         self._rpc_service_list[service_idx].run_server()
 
     def prepare_server(self):
+        """
+        Prepare all servers to be started, and append them into list. 
+        """
         for i, device_id in enumerate(self._devices):
             if self._workdir != "":
                 workdir = "{}_{}".format(self._workdir, i)
@@ -135,6 +227,9 @@ class LocalServiceHandler(object):
                     ir_optim=self._ir_optim))
 
     def start_server(self):
+        """
+        Start multiple processes and start one server in each process
+        """
         for i, service in enumerate(self._rpc_service_list):
             p = multiprocessing.Process(
                 target=self._start_one_server, args=(i, ))

python/pipeline/operator.py

@@ -24,6 +24,7 @@ import os
 import sys
 import collections
 import numpy as np
+import json
 from numpy import *
 if sys.version_info.major == 2:
     import Queue
@@ -33,9 +34,9 @@ else:
     raise Exception("Error Python version")
 
 from .proto import pipeline_service_pb2
-from .channel import (ThreadChannel, ProcessChannel, ChannelDataEcode,
+from .channel import (ThreadChannel, ProcessChannel, ChannelDataErrcode,
                       ChannelData, ChannelDataType, ChannelStopError,
-                      ChannelTimeoutError)
+                      ChannelTimeoutError, ProductErrCode)
 from .util import NameGenerator
 from .profiler import UnsafeTimeProfiler as TimeProfiler
 from . import local_service_handler
@@ -88,6 +89,18 @@ class Op(object):
         self._succ_close_op = False
 
     def init_from_dict(self, conf):
+        """
+        Initializing one Op from config.yaml. If server_endpoints exist,
+        which is remote RPC mode, otherwise it is local RPC mode. There
+        are three types of predictios in local RPC mode, brpc, grpc and
+        local_predictor.
+
+        Args:
+            conf: config.yaml
+
+        Returns:
+            None
+        """
         # init op
         if self.concurrency is None:
             self.concurrency = conf["concurrency"]
@@ -118,34 +131,46 @@ class Op(object):
         else:
             self._auto_batching_timeout = self._auto_batching_timeout / 1000.0
 
+        self.model_config = None
+        self.workdir = None
+        self.thread_num = self.concurrency
+        self.devices = ""
+        self.mem_optim = False
+        self.ir_optim = False
         if self._server_endpoints is None:
             server_endpoints = conf.get("server_endpoints", [])
             if len(server_endpoints) != 0:
                 # remote service
                 self.with_serving = True
                 self._server_endpoints = server_endpoints
+                self.client_type = conf["client_type"]
             else:
                 if self._local_service_handler is None:
                     local_service_conf = conf.get("local_service_conf")
                     _LOGGER.info("local_service_conf: {}".format(
                         local_service_conf))
-                    model_config = local_service_conf.get("model_config")
+                    self.model_config = local_service_conf.get("model_config")
                     self.client_type = local_service_conf.get("client_type")
-                    _LOGGER.info("model_config: {}".format(model_config))
-                    if model_config is None:
+                    self.workdir = local_service_conf.get("workdir")
+                    self.thread_num = local_service_conf.get("thread_num")
+                    self.devices = local_service_conf.get("devices")
+                    self.mem_optim = local_service_conf.get("mem_optim")
+                    self.ir_optim = local_service_conf.get("ir_optim")
+                    self._fetch_names = local_service_conf.get("fetch_list")
+                    if self.model_config is None:
                         self.with_serving = False
                     else:
                         # local rpc service
                         self.with_serving = True
                         if self.client_type == "brpc" or self.client_type == "grpc":
                             service_handler = local_service_handler.LocalServiceHandler(
-                                model_config=model_config,
+                                model_config=self.model_config,
                                 client_type=self.client_type,
-                                workdir=local_service_conf["workdir"],
-                                thread_num=local_service_conf["thread_num"],
-                                devices=local_service_conf["devices"],
-                                mem_optim=local_service_conf["mem_optim"],
-                                ir_optim=local_service_conf["ir_optim"])
+                                workdir=self.workdir,
+                                thread_num=self.thread_num,
+                                devices=self.devices,
+                                mem_optim=self.mem_optim,
+                                ir_optim=self.ir_optim)
                             service_handler.prepare_server()  # get fetch_list
                             serivce_ports = service_handler.get_port_list()
                             self._server_endpoints = [
@@ -159,19 +184,15 @@ class Op(object):
                                 )
                         elif self.client_type == "local_predictor":
                             service_handler = local_service_handler.LocalServiceHandler(
-                                model_config=model_config,
+                                model_config=self.model_config,
                                 client_type=self.client_type,
-                                workdir=local_service_conf["workdir"],
-                                thread_num=local_service_conf["thread_num"],
-                                devices=local_service_conf["devices"])
-                            #service_handler.prepare_server()  # get fetch_list
-                            self.local_predictor = service_handler.get_client()
+                                workdir=self.workdir,
+                                thread_num=self.thread_num,
+                                devices=self.devices,
+                                fetch_names=self._fetch_names)
                             if self._client_config is None:
                                 self._client_config = service_handler.get_client_config(
                                 )
-                            if self._fetch_names is None:
-                                self._fetch_names = service_handler.get_fetch_list(
-                                )
                         self._local_service_handler = service_handler
                 else:
                     self.with_serving = True
@@ -208,6 +229,15 @@ class Op(object):
                               self._batch_size, self._auto_batching_timeout)))
 
     def launch_local_rpc_service(self):
+        """
+        Launching multiple local rpc servers.
+
+        Args:
+            None
+
+        Returns:
+            None
+        """
         if self._local_service_handler is None:
             _LOGGER.warning(
                 self._log("Failed to launch local rpc"
@@ -222,6 +252,15 @@ class Op(object):
                      .format(self.name, port))
 
     def use_default_auto_batching_config(self):
+        """
+        Set the auto batching config default.
+
+        Args:
+            None
+
+        Returns:
+            None
+        """
         if self._batch_size != 1:
             _LOGGER.warning("Op({}) reset batch_size=1 (original: {})"
                             .format(self.name, self._batch_size))
@@ -239,6 +278,18 @@ class Op(object):
         self._tracer = tracer
 
     def init_client(self, client_config, server_endpoints):
+        """
+        Initialize the client object. There are three types of clients, brpc,
+        grpc and local_predictor. In grpc or brpc mode, the client connects 
+        endpoints.
+
+        Args:
+            client_config: client config info
+            server_endpoints: server IP/Port list.
+
+        Returns:
+            client: client object.
+        """
         if self.with_serving == False:
             _LOGGER.info("Op({}) has no client (and it also do not "
                          "run the process function)".format(self.name))
@@ -266,6 +317,16 @@ class Op(object):
         return self._input_ops
 
     def set_input_ops(self, ops):
+        """
+        Set input ops.Each op have many input ops, but only one input
+        channel.
+
+        Args:
+            ops: op list
+
+        Returns:
+            None.
+        """
         if not isinstance(ops, list):
             ops = [] if ops is None else [ops]
         self._input_ops = []
@@ -278,6 +339,10 @@ class Op(object):
             self._input_ops.append(op)
 
     def add_input_channel(self, channel):
+        """
+        Adding one input channel to the Op. Each op have many front op,
+        but, only one input channel.
+        """
         if not isinstance(channel, (ThreadChannel, ProcessChannel)):
             _LOGGER.critical(
                 self._log("Failed to set input_channel: input "
@@ -294,6 +359,16 @@ class Op(object):
         return self._input
 
     def add_output_channel(self, channel):
+        """
+        Adding one output channel to the Op. Each op have many output channels,
+        But only one front channel.
+
+        Args:
+            channel: an output channel object.
+
+        Returns:
+            None
+        """
         if not isinstance(channel, (ThreadChannel, ProcessChannel)):
             _LOGGER.critical(
                 self._log("Failed to add output_channel: output channel "
@@ -308,7 +383,23 @@ class Op(object):
     def _get_output_channels(self):
         return self._outputs
 
-    def preprocess(self, input_dicts):
+    def preprocess(self, input_dicts, data_id=0, log_id=0):
+        """
+        In preprocess stage, assembling data for process stage. users can 
+        override this function for model feed features.
+
+        Args:
+            input_dicts: input data to be preprocessed
+            data_id: inner unique id, 0 default
+            log_id: global unique id for RTT, 0 default
+
+        Return:
+            input_dict: data for process stage
+            is_skip_process: skip process stage or not, False default
+            prod_errcode: None default, otherwise, product errores occured.
+                          It is handled in the same way as exception. 
+            prod_errinfo: "" default
+        """
         # multiple previous Op
         if len(input_dicts) != 1:
             _LOGGER.critical(
@@ -318,9 +409,20 @@ class Op(object):
             os._exit(-1)
 
         (_, input_dict), = input_dicts.items()
-        return input_dict
-
-    def process(self, feed_batch, typical_logid):
+        return input_dict, False, None, ""
+
+    def process(self, feed_batch, typical_logid=0):
+        """
+        In process stage, send requests to the inference server or predict locally.
+        users do not need to inherit this function
+        Args:
+            feed_batch: data to be fed to inference server
+            typical_logid: mark batch predicts, usually the first logid in batch,
+                0 default.
+
+        Returns:
+            call_result: predict result
+        """
         err, err_info = ChannelData.check_batch_npdata(feed_batch)
         if err != 0:
             _LOGGER.critical(
@@ -345,27 +447,54 @@ class Op(object):
             call_result.pop("serving_status_code")
         return call_result
 
-    def postprocess(self, input_dict, fetch_dict):
-        return fetch_dict
+    def postprocess(self, input_dict, fetch_dict, log_id=0):
+        """
+        In postprocess stage, assemble data for next op or output.
+        Args:
+            input_dict: data returned in preprocess stage.
+            fetch_dict: data returned in process stage.
+            log_id: logid, 0 default
+
+        Returns: 
+            fetch_dict: return fetch_dict default
+            prod_errcode: None default, otherwise, product errores occured.
+                          It is handled in the same way as exception.
+            prod_errinfo: "" default
+        """
+        return fetch_dict, None, ""
 
     def _parse_channeldata(self, channeldata_dict):
+        """
+        Parse one channeldata 
+        Args:
+            channeldata_dict : channel data to be parsed, dict type
+        
+        Return:
+            data_id: created by dag._id_generator, unique
+            error_channeldata: error channeldata
+            parsed_data: get np/dict data from channeldata
+            client_need_profile: need profile info
+            profile_set: profile info
+            log_id: logid for tracing a request 
+        """
         data_id, error_channeldata = None, None
         client_need_profile, profile_set = False, set()
         parsed_data = {}
 
         key = list(channeldata_dict.keys())[0]
         data_id = channeldata_dict[key].id
+        log_id = channeldata_dict[key].log_id
         client_need_profile = channeldata_dict[key].client_need_profile
 
         for name, data in channeldata_dict.items():
-            if data.ecode != ChannelDataEcode.OK.value:
+            if data.error_code != ChannelDataErrcode.OK.value:
                 error_channeldata = data
                 break
             parsed_data[name] = data.parse()
             if client_need_profile:
                 profile_set |= data.profile_data_set
         return (data_id, error_channeldata, parsed_data, client_need_profile,
-                profile_set)
+                profile_set, log_id)
 
     def _push_to_output_channels(self,
                                  data,
@@ -374,6 +503,20 @@ class Op(object):
                                  profile_str=None,
                                  client_need_profile=False,
                                  profile_set=None):
+        """
+        Push data to output channels, Do not run the later stage(preprocess,
+        process, postprocess)
+        Args:
+            data: channeldata, to be pushed
+            channels: output channels
+            name: op name  
+            profile_str: one profile message
+            client_need_profile: False default
+            profile_set: profile message collections
+
+        Returns:
+            None
+        """
         if name is None:
             name = self.name
 
@@ -387,6 +530,16 @@ class Op(object):
             channel.push(data, name)
 
     def start_with_process(self):
+        """
+        Each OP creates a process to run the main loop, initializes the CUDA
+        environment in each individual process.
+
+        Args:
+            None
+
+        Returns:
+            process array
+        """
         trace_buffer = None
         if self._tracer is not None:
             trace_buffer = self._tracer.data_buffer()
@@ -395,22 +548,42 @@ class Op(object):
             p = multiprocessing.Process(
                 target=self._run,
                 args=(concurrency_idx, self._get_input_channel(),
-                      self._get_output_channels(), False, trace_buffer))
+                      self._get_output_channels(), False, trace_buffer,
+                      self.model_config, self.workdir, self.thread_num,
+                      self.devices, self.mem_optim, self.ir_optim))
             p.daemon = True
             p.start()
             process.append(p)
         return process
 
     def start_with_thread(self):
+        """
+        Each OP creates a thread to run the main loop, initializes the CUDA 
+        environment in the main thread.
+
+        Args:
+            None
+ 
+        Returns:
+            thread array
+        """
         trace_buffer = None
         if self._tracer is not None:
             trace_buffer = self._tracer.data_buffer()
+
+        #Init cuda env in main thread
+        if self.client_type == "local_predictor":
+            _LOGGER.info("Init cuda env in main thread")
+            self.local_predictor = self._local_service_handler.get_client()
+
         threads = []
         for concurrency_idx in range(self.concurrency):
             t = threading.Thread(
                 target=self._run,
                 args=(concurrency_idx, self._get_input_channel(),
-                      self._get_output_channels(), True, trace_buffer))
+                      self._get_output_channels(), True, trace_buffer,
+                      self.model_config, self.workdir, self.thread_num,
+                      self.devices, self.mem_optim, self.ir_optim))
             # When a process exits, it attempts to terminate
             # all of its daemonic child processes.
             t.daemon = True
@@ -421,52 +594,109 @@ class Op(object):
     def init_op(self):
         pass
 
-    def _run_preprocess(self, parsed_data_dict, op_info_prefix):
+    def _run_preprocess(self, parsed_data_dict, op_info_prefix, logid_dict):
+        """
+        Run preprocess stage
+        Args:
+            parsed_data_dict: data to be pre-processed
+            op_info_prefix: input op info
+            logid_dict: logid dict
+
+        Returns:
+            preped_data_dict: data preprocessed, to be processed 
+            err_channeldata_dict: when exceptions occurred, putting errors in it.
+            skip_process_dict: skip process stage or not
+
+        """
         _LOGGER.debug("{} Running preprocess".format(op_info_prefix))
         preped_data_dict = collections.OrderedDict()
         err_channeldata_dict = collections.OrderedDict()
+        skip_process_dict = {}
         for data_id, parsed_data in parsed_data_dict.items():
             preped_data, error_channeldata = None, None
+            is_skip_process = False
+            prod_errcode, prod_errinfo = None, None
+            log_id = logid_dict.get(data_id)
             try:
-                preped_data = self.preprocess(parsed_data)
+                preped_data, is_skip_process, prod_errcode, prod_errinfo = self.preprocess(
+                    parsed_data, data_id, logid_dict.get(data_id))
+                # Set skip_process_dict
+                if is_skip_process is True:
+                    skip_process_dict[data_id] = True
             except TypeError as e:
                 # Error type in channeldata.datatype
-                error_info = "(logid={}) {} Failed to preprocess: {}".format(
-                    data_id, op_info_prefix, e)
+                error_info = "(data_id={} log_id={}) {} Failed to preprocess: {}".format(
+                    data_id, log_id, op_info_prefix, e)
                 _LOGGER.error(error_info, exc_info=True)
                 error_channeldata = ChannelData(
-                    ecode=ChannelDataEcode.TYPE_ERROR.value,
+                    error_code=ChannelDataErrcode.TYPE_ERROR.value,
                     error_info=error_info,
-                    data_id=data_id)
+                    data_id=data_id,
+                    log_id=log_id)
             except Exception as e:
-                error_info = "(logid={}) {} Failed to preprocess: {}".format(
-                    data_id, op_info_prefix, e)
+                error_info = "(data_id={} log_id={}) {} Failed to preprocess: {}".format(
+                    data_id, log_id, op_info_prefix, e)
                 _LOGGER.error(error_info, exc_info=True)
                 error_channeldata = ChannelData(
-                    ecode=ChannelDataEcode.UNKNOW.value,
+                    error_code=ChannelDataErrcode.UNKNOW.value,
                     error_info=error_info,
-                    data_id=data_id)
+                    data_id=data_id,
+                    log_id=log_id)
+
+            if prod_errcode is not None:
+                # product errors occured
+                error_channeldata = ChannelData(
+                    error_code=ChannelDataErrcode.PRODUCT_ERROR.value,
+                    error_info="",
+                    prod_error_code=prod_errcode,
+                    prod_error_info=prod_errinfo,
+                    data_id=data_id,
+                    log_id=log_id)
+
             if error_channeldata is not None:
                 err_channeldata_dict[data_id] = error_channeldata
             else:
                 preped_data_dict[data_id] = preped_data
         _LOGGER.debug("{} Succ preprocess".format(op_info_prefix))
-        return preped_data_dict, err_channeldata_dict
-
-    def _run_process(self, preped_data_dict, op_info_prefix):
+        return preped_data_dict, err_channeldata_dict, skip_process_dict
+
+    def _run_process(self, preped_data_dict, op_info_prefix, skip_process_dict,
+                     logid_dict):
+        """
+        Run process stage
+        Args:
+            preped_data_dict: feed the data to be predicted by the model.  
+            op_info_prefix: prefix op info
+            skip_process_dict: skip process stage or not
+            logid_dict: logid dict
+
+        Returns:
+            midped_data_dict: data midprocessed, to be post-processed 
+            err_channeldata_dict: when exceptions occurred, putting errors in it 
+        """
         _LOGGER.debug("{} Running process".format(op_info_prefix))
         midped_data_dict = collections.OrderedDict()
         err_channeldata_dict = collections.OrderedDict()
-        if self.with_serving:
-            data_ids = preped_data_dict.keys()
+        ### if (batch_num == 1 && skip == True) ,then skip the process stage.
+        is_skip_process = False
+        data_ids = preped_data_dict.keys()
+        if len(data_ids) == 1 and skip_process_dict.get(data_ids[0]) == True:
+            is_skip_process = True
+            _LOGGER.info("(data_id={} log_id={}) skip process stage".format(
+                data_ids[0], logid_dict.get(data_ids[0])))
+
+        if self.with_serving is True and is_skip_process is False:
+            # use typical_logid to mark batch data
             typical_logid = data_ids[0]
             if len(data_ids) != 1:
                 for data_id in data_ids:
                     _LOGGER.info(
-                        "(logid={}) {} During access to PaddleServingService,"
+                        "(data_id={} logid={}) {} During access to PaddleServingService,"
                         " we selected logid={} (from batch: {}) as a "
                         "representative for logging.".format(
-                            data_id, op_info_prefix, typical_logid, data_ids))
+                            data_id,
+                            logid_dict.get(data_id), op_info_prefix,
+                            typical_logid, data_ids))
 
             # combine samples to batch
             one_input = preped_data_dict[data_ids[0]]
@@ -486,64 +716,70 @@ class Op(object):
                     input_offset.append(offset)
             else:
                 _LOGGER.critical(
-                    "{} Failed to process: expect input type is dict(sample"
-                    " input) or list(batch input), but get {}".format(
-                        op_info_prefix, type(one_input)))
+                    "(data_id={} log_id={}){} Failed to process: expect input type is dict(sample"
+                    " input) or list(batch input), but get {}".format(data_ids[
+                        0], typical_logid, op_info_prefix, type(one_input)))
                 os._exit(-1)
 
             midped_batch = None
-            ecode = ChannelDataEcode.OK.value
+            error_code = ChannelDataErrcode.OK.value
             if self._timeout <= 0:
                 try:
                     midped_batch = self.process(feed_batch, typical_logid)
                 except Exception as e:
-                    ecode = ChannelDataEcode.UNKNOW.value
-                    error_info = "(logid={}) {} Failed to process(batch: {}): {}".format(
-                        typical_logid, op_info_prefix, data_ids, e)
+                    error_code = ChannelDataErrcode.UNKNOW.value
+                    error_info = "(data_id={} log_id={}) {} Failed to process(batch: {}): {}".format(
+                        data_ids[0], typical_logid, op_info_prefix, data_ids, e)
                     _LOGGER.error(error_info, exc_info=True)
             else:
+                # retry N times configed in yaml files.
                 for i in range(self._retry):
                     try:
+                        # time out for each process
                         midped_batch = func_timeout.func_timeout(
                             self._timeout,
                             self.process,
                             args=(feed_batch, typical_logid))
                     except func_timeout.FunctionTimedOut as e:
                         if i + 1 >= self._retry:
-                            ecode = ChannelDataEcode.TIMEOUT.value
-                            error_info = "(logid={}) {} Failed to process(batch: {}): " \
+                            error_code = ChannelDataErrcode.TIMEOUT.value
+                            error_info = "(log_id={}) {} Failed to process(batch: {}): " \
                                     "exceeded retry count.".format(
                                             typical_logid, op_info_prefix, data_ids)
                             _LOGGER.error(error_info)
                         else:
                             _LOGGER.warning(
-                                "(logid={}) {} Failed to process(batch: {}): timeout,"
+                                "(log_id={}) {} Failed to process(batch: {}): timeout,"
                                 " and retrying({}/{})...".format(
                                     typical_logid, op_info_prefix, data_ids, i +
                                     1, self._retry))
                     except Exception as e:
-                        ecode = ChannelDataEcode.UNKNOW.value
-                        error_info = "(logid={}) {} Failed to process(batch: {}): {}".format(
+                        error_code = ChannelDataErrcode.UNKNOW.value
+                        error_info = "(log_id={}) {} Failed to process(batch: {}): {}".format(
                             typical_logid, op_info_prefix, data_ids, e)
                         _LOGGER.error(error_info, exc_info=True)
                         break
                     else:
                         break
-            if ecode != ChannelDataEcode.OK.value:
+            if error_code != ChannelDataErrcode.OK.value:
                 for data_id in data_ids:
                     err_channeldata_dict[data_id] = ChannelData(
-                        ecode=ecode, error_info=error_info, data_id=data_id)
+                        error_code=error_code,
+                        error_info=error_info,
+                        data_id=data_id,
+                        log_id=logid_dict.get(data_id))
             elif midped_batch is None:
                 # op client return None
-                error_info = "(logid={}) {} Failed to predict, please check if " \
+                error_info = "(log_id={}) {} Failed to predict, please check if " \
                         "PaddleServingService is working properly.".format(
                                 typical_logid, op_info_prefix)
                 _LOGGER.error(error_info)
                 for data_id in data_ids:
                     err_channeldata_dict[data_id] = ChannelData(
-                        ecode=ChannelDataEcode.CLIENT_ERROR.value,
+                        error_code=ChannelDataErrcode.CLIENT_ERROR.value,
                         error_info=error_info,
-                        data_id=data_id)
+                        data_id=data_id,
+                        log_id=logid_dict.get(data_id))
             else:
                 # transform np format to dict format
                 var_names = midped_batch.keys()
@@ -552,7 +788,7 @@ class Op(object):
                 for name in var_names:
                     lod_offset_name = "{}.lod".format(name)
                     if lod_offset_name in var_names:
-                        _LOGGER.debug("(logid={}) {} {} is LodTensor".format(
+                        _LOGGER.debug("(log_id={}) {} {} is LodTensor".format(
                             typical_logid, op_info_prefix, name))
                         lod_var_names.add(name)
                         lod_offset_names.add(lod_offset_name)
@@ -588,38 +824,67 @@ class Op(object):
         return midped_data_dict, err_channeldata_dict
 
     def _run_postprocess(self, parsed_data_dict, midped_data_dict,
-                         op_info_prefix):
+                         op_info_prefix, logid_dict):
+        """
+        Run postprocess stage.
+        Args:
+            parsed_data_dict: data returned in preprocess stage 
+            midped_data_dict: data returned in process stage
+            op_info_prefix: prefix op info
+            logid_dict: logid dict
+
+        Returns:
+            postped_data_dict: data postprocessed 
+            err_channeldata_dict: when exceptions occurred, putting errors in it
+ 
+        """
         _LOGGER.debug("{} Running postprocess".format(op_info_prefix))
         postped_data_dict = collections.OrderedDict()
         err_channeldata_dict = collections.OrderedDict()
         for data_id, midped_data in midped_data_dict.items():
+            log_id = logid_dict.get(data_id)
             postped_data, err_channeldata = None, None
+            prod_errcode, prod_errinfo = None, None
             try:
-                postped_data = self.postprocess(parsed_data_dict[data_id],
-                                                midped_data)
+                postped_data, prod_errcode, prod_errinfo = self.postprocess(
+                    parsed_data_dict[data_id], midped_data,
+                    logid_dict.get(data_id))
             except Exception as e:
-                error_info = "(logid={}) {} Failed to postprocess: {}".format(
-                    data_id, op_info_prefix, e)
+                error_info = "(data_id={} log_id={}) {} Failed to postprocess: {}".format(
+                    data_id, log_id, op_info_prefix, e)
                 _LOGGER.error(error_info, exc_info=True)
                 err_channeldata = ChannelData(
-                    ecode=ChannelDataEcode.UNKNOW.value,
+                    error_code=ChannelDataErrcode.UNKNOW.value,
                     error_info=error_info,
-                    data_id=data_id)
+                    data_id=data_id,
+                    log_id=log_id)
+
+            if prod_errcode is not None:
+                # product errors occured
+                err_channeldata = ChannelData(
+                    error_code=ChannelDataErrcode.PRODUCT_ERROR.value,
+                    error_info="",
+                    prod_error_code=prod_errcode,
+                    prod_error_info=prod_errinfo,
+                    data_id=data_id,
+                    log_id=log_id)
+
             if err_channeldata is not None:
                 err_channeldata_dict[data_id] = err_channeldata
                 continue
             else:
                 if not isinstance(postped_data, dict):
-                    error_info = "(logid={}) {} Failed to postprocess: " \
+                    error_info = "(log_id={} log_id={}) {} Failed to postprocess: " \
                             "output of postprocess funticon must be " \
                             "dict type, but get {}".format(
-                                data_id, op_info_prefix,
+                                data_id, log_id, op_info_prefix,
                                 type(postped_data))
                     _LOGGER.error(error_info)
                     err_channeldata = ChannelData(
-                        ecode=ChannelDataEcode.UNKNOW.value,
+                        error_code=ChannelDataErrcode.UNKNOW.value,
                         error_info=error_info,
-                        data_id=data_id)
+                        data_id=data_id,
+                        log_id=log_id)
                     err_channeldata_dict[data_id] = err_channeldata
                     continue
 
@@ -629,18 +894,36 @@ class Op(object):
                     output_data = ChannelData(
                         ChannelDataType.CHANNEL_NPDATA.value,
                         npdata=postped_data,
-                        data_id=data_id)
+                        data_id=data_id,
+                        log_id=log_id)
                 else:
                     output_data = ChannelData(
                         ChannelDataType.DICT.value,
                         dictdata=postped_data,
-                        data_id=data_id)
+                        data_id=data_id,
+                        log_id=log_id)
                 postped_data_dict[data_id] = output_data
         _LOGGER.debug("{} Succ postprocess".format(op_info_prefix))
         return postped_data_dict, err_channeldata_dict
 
     def _auto_batching_generator(self, input_channel, op_name, batch_size,
                                  timeout, op_info_prefix):
+        """
+        Merge batch_size requests for one prediction.Taking one piece of data 
+        from the input channel each time until equals batch_size, or the waiting 
+        time exceeds auto_batching_timeout.
+
+        Args:
+            input_channel: the input channel of Op
+            op_name: op name
+            batch_size: batch size, Less than worker_num
+            timeout: batch timeout, seconds, If timeout is None, and the quantity 
+                taken from the front is less than batch_size, blocking occured.
+            op_info_prefix: op link info.
+
+        Returns:
+            None
+        """
         while True:
             batch = []
             while len(batch) == 0:
@@ -661,6 +944,9 @@ class Op(object):
                         else:
                             channeldata_dict = input_channel.front(op_name)
                         batch.append(channeldata_dict)
+                        _LOGGER.debug(
+                            "_auto_batching_generator get {} channeldata from op:{} into batch, batch_size:{}".
+                            format(idx, op_name, batch_size))
                     except ChannelTimeoutError:
                         _LOGGER.debug("{} Failed to generate batch: "
                                       "timeout".format(op_info_prefix))
@@ -670,37 +956,91 @@ class Op(object):
             yield batch
 
     def _parse_channeldata_batch(self, batch, output_channels):
+        """
+        Parse channeldatas batch
+        Args:
+            batch: auto-batching batch datas
+            output_channels: output channels 
+
+        Returns:
+            parsed_data_dict: parsed from channeldata in batch
+            need_profile_dict: need profile dict in batch 
+            profile_dict: profile info dict in batch
+            logid_dict: trace each request in batch
+        """
         parsed_data_dict = collections.OrderedDict()
         need_profile_dict = {}
         profile_dict = {}
+        logid_dict = {}
         for channeldata_dict in batch:
             (data_id, error_channeldata, parsed_data,
-                    client_need_profile, profile_set) = \
+                    client_need_profile, profile_set, log_id) = \
                             self._parse_channeldata(channeldata_dict)
             if error_channeldata is None:
                 parsed_data_dict[data_id] = parsed_data
                 need_profile_dict[data_id] = client_need_profile
                 profile_dict[data_id] = profile_set
+                logid_dict[data_id] = log_id
             else:
                 # error data in predecessor Op
                 # (error_channeldata with profile info)
                 self._push_to_output_channels(error_channeldata,
                                               output_channels)
 
-        return parsed_data_dict, need_profile_dict, profile_dict
+        return parsed_data_dict, need_profile_dict, profile_dict, logid_dict
 
     def _run(self, concurrency_idx, input_channel, output_channels,
-             is_thread_op, trace_buffer):
+             is_thread_op, trace_buffer, model_config, workdir, thread_num,
+             devices, mem_optim, ir_optim):
+        """
+        _run() is the entry function of OP process / thread model.When client 
+        type is local_predictor in process mode, the CUDA environment needs to 
+        be initialized by LocalServiceHandler[child process], otherwise, Cuda
+        error(3), initialization error is occured. Preprocess, process and 
+        postprocess are executed in the main loop. The preprocess and postprocess
+        function is usually rewrited by users. Trace data is recorded by trace_que.
+
+        Args:
+            concurrency_idx: thread/process index
+            input_channel: input channel, take the data to be processed
+            output_channels: output channel, store processed data
+            is_thread_op: False, It's process op; True, It's thread op
+            trace_buffer: store trace infomations
+            model_config: model config path
+            workdir: work directory
+            thread_num: number of threads, concurrent quantity
+            devices: gpu id list[gpu], "" default[cpu]
+            mem_optim: use memory/graphics memory optimization, True default.
+            ir_optim: use calculation chart optimization, False default. 
+
+        Returns:
+            None
+        """
         op_info_prefix = "[{}|{}]".format(self.name, concurrency_idx)
         tid = threading.current_thread().ident
 
-        # init op
+        # init ops
         profiler = None
         try:
+            if is_thread_op == False and self.client_type == "local_predictor":
+                self.service_handler = local_service_handler.LocalServiceHandler(
+                    model_config=model_config,
+                    client_type="local_predictor",
+                    workdir=workdir,
+                    thread_num=thread_num,
+                    devices=devices,
+                    mem_optim=mem_optim,
+                    ir_optim=ir_optim)
+
+                _LOGGER.info("Init cuda env in process {}".format(
+                    concurrency_idx))
+                self.local_predictor = self.service_handler.get_client()
+            # check all ops initialized successfully.
             profiler = self._initialize(is_thread_op, concurrency_idx)
+
         except Exception as e:
             _LOGGER.critical(
-                "{} Failed to init op: {}".format(op_info_prefix, e),
+                "{} failed to init op: {}".format(op_info_prefix, e),
                 exc_info=True)
             os._exit(-1)
         _LOGGER.info("{} Succ init".format(op_info_prefix))
@@ -727,7 +1067,7 @@ class Op(object):
 
             # parse channeldata batch
             try:
-                parsed_data_dict, need_profile_dict, profile_dict \
+                parsed_data_dict, need_profile_dict, profile_dict, logid_dict\
                         = self._parse_channeldata_batch(
                                 channeldata_dict_batch, output_channels)
             except ChannelStopError:
@@ -740,11 +1080,12 @@ class Op(object):
 
             # preprecess
             start = profiler.record("prep#{}_0".format(op_info_prefix))
-            preped_data_dict, err_channeldata_dict \
-                    = self._run_preprocess(parsed_data_dict, op_info_prefix)
+            preped_data_dict, err_channeldata_dict, skip_process_dict \
+                    = self._run_preprocess(parsed_data_dict, op_info_prefix, logid_dict)
             end = profiler.record("prep#{}_1".format(op_info_prefix))
             prep_time = end - start
             try:
+                # put error requests into output channel, skip process and postprocess stage
                 for data_id, err_channeldata in err_channeldata_dict.items():
                     self._push_to_output_channels(
                         data=err_channeldata,
@@ -761,7 +1102,7 @@ class Op(object):
             # process
             start = profiler.record("midp#{}_0".format(op_info_prefix))
             midped_data_dict, err_channeldata_dict \
-                    = self._run_process(preped_data_dict, op_info_prefix)
+                    = self._run_process(preped_data_dict, op_info_prefix, skip_process_dict, logid_dict)
             end = profiler.record("midp#{}_1".format(op_info_prefix))
             midp_time = end - start
             try:
@@ -781,8 +1122,7 @@ class Op(object):
             # postprocess
             start = profiler.record("postp#{}_0".format(op_info_prefix))
             postped_data_dict, err_channeldata_dict \
-                    = self._run_postprocess(
-                            parsed_data_dict, midped_data_dict, op_info_prefix)
+                    = self._run_postprocess(parsed_data_dict, midped_data_dict, op_info_prefix, logid_dict)
             end = profiler.record("postp#{}_1".format(op_info_prefix))
             postp_time = end - start
             try:
@@ -838,6 +1178,19 @@ class Op(object):
                         break
 
     def _initialize(self, is_thread_op, concurrency_idx):
+        """
+        Initialize one OP object in the target function of a thread or porcess.
+        Initialize the client object with _client_config and _server_endpoints.
+        Create a TimeProfiler per thread or process for recording profiler info.
+
+        Args:
+            is_thread_op: True, one op runs in one thread; False, one op runs
+                in one process.
+            concurrency_idx: process id, Thread mode does not use this param.
+
+        Returns:
+            TimeProfiler
+        """
         if is_thread_op:
             with self._for_init_op_lock:
                 if not self._succ_init_op:
@@ -877,9 +1230,17 @@ class Op(object):
 
 
 class RequestOp(Op):
-    """ RequestOp do not run preprocess, process, postprocess. """
+    """
+    RequestOp is a special Op, for unpacking one request package. If the
+    request needs one special unpackaging method, you need to inherit class
+    RequestOp and rewrite function unpack_request_package.Notice!!! Class
+    RequestOp does not run preprocess, process, postprocess.
+    """
 
     def __init__(self):
+        """
+        Initialize the RequestOp
+        """
         # PipelineService.name = "@DAGExecutor"
         super(RequestOp, self).__init__(name="@DAGExecutor", input_ops=[])
         # init op
@@ -890,7 +1251,25 @@ class RequestOp(Op):
             os._exit(-1)
 
     def unpack_request_package(self, request):
-        dictdata = {}
+        """
+        Unpack request package by gateway.proto
+        Args:
+            request: HTTP body, JSON format
+
+        Returns:
+            dict_data: json fields in HTTP body
+            log_id: log_id
+            prod_errcode: None or ProductErrCode.SUCC.value default, otherwise,
+                          product errores occured.It is handled in the same way
+                          as exception.
+            prod_errinfo: "" default 
+        """
+        dict_data = {}
+        log_id = None
+        if request is None:
+            _LOGGER.critical("request is None")
+            raise ValueError("request is None")
+
         for idx, key in enumerate(request.key):
             data = request.value[idx]
             try:
@@ -899,14 +1278,27 @@ class RequestOp(Op):
                     data = evaled_data
             except Exception as e:
                 pass
-            dictdata[key] = data
-        return dictdata
+            dict_data[key] = data
+        log_id = request.logid
+        _LOGGER.info("RequestOp unpack one request. log_id:{}, clientip:{} \
+            name:{}, method:{}".format(log_id, request.clientip, request.name,
+                                       request.method))
+
+        return dict_data, log_id, None, ""
 
 
 class ResponseOp(Op):
-    """ ResponseOp do not run preprocess, process, postprocess. """
+    """ 
+    ResponseOp is a special Op, for packing one response package. If the channeldata 
+    needs a special packaging method, you need to inherit class ReponseOp and rewrite
+    pack_response_package function. Notice!!! Class ResponseOp does not run preprocess,
+    process, postprocess.
+    """
 
     def __init__(self, input_ops):
+        """
+        Initialize the ResponseOp
+        """
         super(ResponseOp, self).__init__(
             name="@DAGExecutor", input_ops=input_ops)
         # init op
@@ -918,9 +1310,21 @@ class ResponseOp(Op):
             os._exit(-1)
 
     def pack_response_package(self, channeldata):
+        """
+        Getting channeldata from the last channel, packting the response 
+        package serialized by protobuf.  
+
+        Args:
+            channeldata: Type ChannelData
+
+        Returns:
+            resp: pipeline_service_pb2.Response()
+        """
         resp = pipeline_service_pb2.Response()
-        resp.ecode = channeldata.ecode
-        if resp.ecode == ChannelDataEcode.OK.value:
+        error_code = channeldata.error_code
+        error_info = ""
+        if error_code == ChannelDataErrcode.OK.value:
+            # Framework level errors
             if channeldata.datatype == ChannelDataType.CHANNEL_NPDATA.value:
                 feed = channeldata.parse()
                 # ndarray to string:
@@ -933,8 +1337,8 @@ class ResponseOp(Op):
                 feed = channeldata.parse()
                 for name, var in feed.items():
                     if not isinstance(var, str):
-                        resp.ecode = ChannelDataEcode.TYPE_ERROR.value
-                        resp.error_info = self._log(
+                        error_code = ChannelDataErrcode.TYPE_ERROR.value
+                        error_info = self._log(
                             "fetch var type must be str({}).".format(
                                 type(var)))
                         _LOGGER.error("(logid={}) Failed to pack RPC "
@@ -944,19 +1348,46 @@ class ResponseOp(Op):
                     resp.value.append(var)
                     resp.key.append(name)
             else:
-                resp.ecode = ChannelDataEcode.TYPE_ERROR.value
-                resp.error_info = self._log(
-                    "error type({}) in datatype.".format(channeldata.datatype))
+                error_code = ChannelDataErrcode.TYPE_ERROR.value
+                error_info = self._log("error type({}) in datatype.".format(
+                    channeldata.datatype))
                 _LOGGER.error("(logid={}) Failed to pack RPC response"
-                              " package: {}".format(channeldata.id,
-                                                    resp.error_info))
+                              " package: {}".format(channeldata.id, error_info))
         else:
-            resp.error_info = channeldata.error_info
+            # Product level errors
+            error_info = channeldata.error_info
+            if error_code == ChannelDataErrcode.PRODUCT_ERROR.value:
+                #rewrite error_code when product errors occured
+                error_code = channeldata.prod_error_code
+                error_info = channeldata.prod_error_info
+
+        # pack results
+        if error_code is None:
+            error_code = 0
+        resp.err_no = error_code
+        resp.err_msg = error_info
+
         return resp
 
 
 class VirtualOp(Op):
-    ''' For connecting two channels. '''
+    """ 
+    To connect 2 ops across levels in dag view, we create virtual ops
+    between non-virtual ops, and transfer data only. For examples, 
+    the pred ops of F are D & E.In the process of building DAG, we will
+    create channels layer by layer according to dag views.Op F is not 
+    in the next layer view of [B, E], so we will create a virtual OP 
+    'V1' whose pred OP is E. And so on, we create two virtual op 'V2'
+    and 'V3', Finally, we find the non-virtual op F. we create 4 channels
+    among E, V1, V2, V3 and F, the producer of V1, V2, V3 and F is E.
+    
+        DAG: [A -> B -> C -> D -> F]
+               \-> E ----------/
+
+        DAG view: [[A], [B, E], [C], [D], [F]]
+        BUILD DAG: [A -> B -> C -> D -> E -> F]
+                     \-> E -> V1-> V2-> V3/
+    """
 
     def __init__(self, name, concurrency=1):
         super(VirtualOp, self).__init__(
@@ -964,9 +1395,27 @@ class VirtualOp(Op):
         self._virtual_pred_ops = []
 
     def add_virtual_pred_op(self, op):
+        """
+        Add the front op of current vritual op.
+        
+        Args:
+            op: one op object, may be a virtual op or not.
+
+        Returns:
+            None
+        """
         self._virtual_pred_ops.append(op)
 
     def _actual_pred_op_names(self, op):
+        """
+        Recursively find the front op which is a non-virtual op.
+   
+        Args:
+            op: one op object
+            
+        Returns:
+            names: the name of non-virtual pred ops.
+        """
         # can use disjoint-set, but it's not necessary
         if not isinstance(op, VirtualOp):
             return [op.name]
@@ -976,6 +1425,15 @@ class VirtualOp(Op):
         return names
 
     def add_output_channel(self, channel):
+        """
+        Adding the output channel of non-virtual pred ops.
+
+        Args:
+            channel: one channel.
+          
+        Returns:
+            None.
+        """
         if not isinstance(channel, (ThreadChannel, ProcessChannel)):
             _LOGGER.critical(
                 self._log("Failed to add output_channel: output_channel"
@@ -989,6 +1447,20 @@ class VirtualOp(Op):
 
     def _run(self, concurrency_idx, input_channel, output_channels, client_type,
              is_thread_op):
+        """
+        The target function _run() only transfers data between OPs in one thread
+        or process.
+
+        Args:
+            concurrency_idx: process id, not avaliable in thread mode.
+            input_channel: input channel
+            output_channels: output channels
+            client_type: no use
+            is_thread_op: True, thread mode; False, process mode
+
+        Returns:
+            None
+        """
         op_info_prefix = "[{}|{}]".format(self.name, concurrency_idx)
         log = get_log_func(op_info_prefix)
         tid = threading.current_thread().ident

python/pipeline/pipeline_client.py

@@ -18,7 +18,9 @@ import numpy as np
 from numpy import *
 import logging
 import functools
-from .channel import ChannelDataEcode
+import json
+import socket
+from .channel import ChannelDataErrcode
 from .proto import pipeline_service_pb2
 from .proto import pipeline_service_pb2_grpc
 
@@ -26,6 +28,10 @@ _LOGGER = logging.getLogger(__name__)
 
 
 class PipelineClient(object):
+    """
+    PipelineClient provides the basic capabilities of the pipeline SDK
+    """
+
     def __init__(self):
         self._channel = None
         self._profile_key = "pipeline.profile"
@@ -42,6 +48,23 @@ class PipelineClient(object):
 
     def _pack_request_package(self, feed_dict, profile):
         req = pipeline_service_pb2.Request()
+
+        logid = feed_dict.get("logid")
+        if logid is None:
+            req.logid = 0
+        else:
+            req.logid = long(logid)
+            feed_dict.pop("logid")
+
+        clientip = feed_dict.get("clientip")
+        if clientip is None:
+            hostname = socket.gethostname()
+            ip = socket.gethostbyname(hostname)
+            req.clientip = ip
+        else:
+            req.clientip = clientip
+            feed_dict.pop("clientip")
+
         np.set_printoptions(threshold=sys.maxsize)
         for key, value in feed_dict.items():
             req.key.append(key)
@@ -60,29 +83,7 @@ class PipelineClient(object):
         return req
 
     def _unpack_response_package(self, resp, fetch):
-        if resp.ecode != 0:
-            return {
-                "ecode": resp.ecode,
-                "ecode_desc": ChannelDataEcode(resp.ecode),
-                "error_info": resp.error_info,
-            }
-        fetch_map = {"ecode": resp.ecode}
-        for idx, key in enumerate(resp.key):
-            if key == self._profile_key:
-                if resp.value[idx] != "":
-                    sys.stderr.write(resp.value[idx])
-                continue
-            if fetch is not None and key not in fetch:
-                continue
-            data = resp.value[idx]
-            try:
-                evaled_data = eval(data)
-                if isinstance(evaled_data, np.ndarray):
-                    data = evaled_data
-            except Exception as e:
-                pass
-            fetch_map[key] = data
-        return fetch_map
+        return resp
 
     def predict(self, feed_dict, fetch=None, asyn=False, profile=False):
         if not isinstance(feed_dict, dict):

python/pipeline/pipeline_server.py

@@ -32,6 +32,10 @@ _LOGGER = logging.getLogger(__name__)
 
 
 class PipelineServicer(pipeline_service_pb2_grpc.PipelineServiceServicer):
+    """
+    Pipeline Servicer entrance.
+    """
+
     def __init__(self, name, response_op, dag_conf, worker_idx=-1):
         super(PipelineServicer, self).__init__()
         self._name = name
@@ -42,10 +46,16 @@ class PipelineServicer(pipeline_service_pb2_grpc.PipelineServiceServicer):
         _LOGGER.info("[PipelineServicer] succ init")
 
     def inference(self, request, context):
+        _LOGGER.info("(log_id={}) inference request name:{} self.name:{}".
+                     format(request.logid, request.name, self._name))
         if request.name != "" and request.name != self._name:
+            _LOGGER.error("(log_id={}) name dismatch error. request.name:{},"
+                          "server.name={}".format(request.logid, request.name,
+                                                  self._name))
             resp = pipeline_service_pb2.Response()
-            resp.ecode = channel.ChannelDataEcode.NO_SERVICE.value
-            resp.error_info = "Failed to inference: Service name error."
+            resp.err_no = channel.ChannelDataErrcode.NO_SERVICE.value
+            resp.err_msg = "Failed to inference: Service name error."
+            resp.result = ""
             return resp
         resp = self._dag_executor.call(request)
         return resp
@@ -53,7 +63,9 @@ class PipelineServicer(pipeline_service_pb2_grpc.PipelineServiceServicer):
 
 @contextlib.contextmanager
 def _reserve_port(port):
-    """Find and reserve a port for all subprocesses to use."""
+    """
+    Find and reserve a port for all subprocesses to use.
+    """
     sock = socket.socket(socket.AF_INET6, socket.SOCK_STREAM)
     sock.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEPORT, 1)
     if sock.getsockopt(socket.SOL_SOCKET, socket.SO_REUSEPORT) == 0:
@@ -66,6 +78,10 @@ def _reserve_port(port):
 
 
 class PipelineServer(object):
+    """
+    Pipeline Server : grpc gateway + grpc server.
+    """
+
     def __init__(self, name=None):
         self._name = name  # for grpc-gateway path
         self._rpc_port = None
@@ -74,6 +90,16 @@ class PipelineServer(object):
         self._proxy_server = None
 
     def _grpc_gateway(self, grpc_port, http_port):
+        """
+        Running a gateway server, linking libproxy_server.so
+
+        Args:
+            grpc_port: GRPC port
+            http_port: HTTP port
+
+        Returns:
+            None
+        """
         import os
         from ctypes import cdll
         from . import gateway
@@ -83,6 +109,17 @@ class PipelineServer(object):
         proxy_server.run_proxy_server(grpc_port, http_port)
 
     def _run_grpc_gateway(self, grpc_port, http_port):
+        """
+        Starting the GRPC gateway in a new process. Exposing one 
+        available HTTP port outside, and reflecting the data to RPC port.
+
+        Args:
+            grpc_port: GRPC port
+            http_port: HTTP port
+
+        Returns:
+            None
+        """
         if http_port <= 0:
             _LOGGER.info("Ignore grpc_gateway configuration.")
             return
@@ -99,6 +136,15 @@ class PipelineServer(object):
         self._proxy_server.start()
 
     def set_response_op(self, response_op):
+        """
+        Set the response OP.
+
+        Args:
+            response_op: ResponseOp or its subclass object
+
+        Returns:
+            None
+        """
         if not isinstance(response_op, operator.ResponseOp):
             raise Exception("Failed to set response_op: response_op "
                             "must be ResponseOp type.")
@@ -109,6 +155,17 @@ class PipelineServer(object):
         self._used_op, _ = dag.DAG.get_use_ops(self._response_op)
 
     def prepare_server(self, yml_file=None, yml_dict=None):
+        """
+        Reading configures from the yml file(config.yaml), and launching
+        local services.
+
+        Args:
+            yml_file: Reading configures from yaml files
+            yml_dict: Reading configures from yaml dict.
+   
+        Returns:
+            None 
+        """
         conf = ServerYamlConfChecker.load_server_yaml_conf(
             yml_file=yml_file, yml_dict=yml_dict)
 
@@ -158,6 +215,15 @@ class PipelineServer(object):
         self._start_local_rpc_service()
 
     def _init_ops(self, op_conf):
+        """
+        Initializing all OPs from dicetory.
+
+        Args:
+            op_conf: the op configures in yaml dict.
+
+        Returns:
+            None.
+        """
         default_conf = {
             "concurrency": 1,
             "timeout": -1,
@@ -187,12 +253,22 @@ class PipelineServer(object):
                 op.launch_local_rpc_service()
 
     def run_server(self):
+        """
+        If _build_dag_each_worker is True, Starting _worker_num processes and 
+        running one GRPC server in each process. Otherwise, Staring one GRPC
+        server.
+
+        Args:
+            None
+
+        Returns:
+            None
+        """
         if self._build_dag_each_worker:
             with _reserve_port(self._rpc_port) as port:
                 bind_address = 'localhost:{}'.format(port)
                 workers = []
                 for i in range(self._worker_num):
-                    show_info = (i == 0)
                     worker = multiprocessing.Process(
                         target=self._run_server_func,
                         args=(bind_address, self._response_op, self._conf, i))
@@ -220,6 +296,15 @@ class PipelineServer(object):
             server.wait_for_termination()
 
     def _run_server_func(self, bind_address, response_op, dag_conf, worker_idx):
+        """
+        Running one GRPC server with PipelineServicer.
+
+        Args:
+            bind_address: binding IP/Port
+            response_op: ResponseOp or its subclass object
+            dag_conf: DAG config
+            worker_idx: Process index.
+        """
         options = [('grpc.so_reuseport', 1),
                    ('grpc.max_send_message_length', 256 * 1024 * 1024),
                    ('grpc.max_send_message_length', 256 * 1024 * 1024)]
@@ -235,6 +320,10 @@ class PipelineServer(object):
 
 
 class ServerYamlConfChecker(object):
+    """
+    Checking validities of server yaml files.
+    """
+
     def __init__(self):
         pass
 

python/pipeline/proto/pipeline_service.proto

@@ -19,13 +19,16 @@ message Request {
   repeated string key = 1;
   repeated string value = 2;
   optional string name = 3;
+  optional string method = 4;
+  optional int64 logid = 5;
+  optional string clientip = 6;
 };
 
 message Response {
-  repeated string key = 1;
-  repeated string value = 2;
-  required int32 ecode = 3;
-  optional string error_info = 4;
+  optional int32 err_no = 1;
+  optional string err_msg = 2;
+  repeated string key = 3;
+  repeated string value = 4;
 };
 
 service PipelineService {