From 977beeec5582d5a838ffc0ca6f98c8f2d1ad3414 Mon Sep 17 00:00:00 2001 From: TeslaZhao Date: Tue, 18 May 2021 19:40:48 +0800 Subject: [PATCH] cherry-pick #1238 --- doc/PIPELINE_SERVING.md | 36 +++++++++++++++++ doc/PIPELINE_SERVING_CN.md | 39 +++++++++++++++++-- python/examples/pipeline/bert/README.md | 19 +++++++++ python/examples/pipeline/bert/README_CN.md | 19 +++++++++ python/examples/pipeline/imagenet/README.md | 4 +- .../examples/pipeline/imagenet/README_CN.md | 5 +-- .../pipeline/imdb_model_ensemble/README.md | 8 ++-- .../pipeline/imdb_model_ensemble/README_CN.md | 8 ++-- python/examples/pipeline/ocr/README.md | 31 +++++++++++---- python/examples/pipeline/ocr/README_CN.md | 25 +++++++++--- python/pipeline/logger.py | 12 ++++-- 11 files changed, 173 insertions(+), 33 deletions(-) create mode 100644 python/examples/pipeline/bert/README.md create mode 100644 python/examples/pipeline/bert/README_CN.md diff --git a/doc/PIPELINE_SERVING.md b/doc/PIPELINE_SERVING.md index 7bfd2f51..aeda8d04 100644 --- a/doc/PIPELINE_SERVING.md +++ b/doc/PIPELINE_SERVING.md @@ -726,6 +726,42 @@ There are two kinds of IDs in the pipeline for concatenating requests, `data_id` The log printed by the Pipeline framework will carry both data_id and log_id. After auto-batching is turned on, the first `data_id` in the batch will be used to mark the whole batch, and the framework will print all data_ids in the batch in a log. + +### 5.2 Log Rotating +Log module of Pipeline Serving is defined in file `logger.py`.`logging.handlers.RotatingFileHandler` is used to support the rotation of disk log files. Set `maxBytes` and `backupCount` according to different file levels and daily quality. When the predetermined size is about to be exceeded, the old file will be closed and a new file will be opened for output. + + +```python +"handlers": { + "f_pipeline.log": { + "class": "logging.handlers.RotatingFileHandler", + "level": "INFO", + "formatter": "normal_fmt", + "filename": os.path.join(log_dir, "pipeline.log"), + "maxBytes": 512000000, + "backupCount": 20, + }, + "f_pipeline.log.wf": { + "class": "logging.handlers.RotatingFileHandler", + "level": "WARNING", + "formatter": "normal_fmt", + "filename": os.path.join(log_dir, "pipeline.log.wf"), + "maxBytes": 512000000, + "backupCount": 10, + }, + "f_tracer.log": { + "class": "logging.handlers.RotatingFileHandler", + "level": "INFO", + "formatter": "tracer_fmt", + "filename": os.path.join(log_dir, "pipeline.tracer"), + "maxBytes": 512000000, + "backupCount": 5, + }, +}, +``` + +*** + ## 6.Performance analysis and optimization diff --git a/doc/PIPELINE_SERVING_CN.md b/doc/PIPELINE_SERVING_CN.md index da7d7d2c..340b66a2 100644 --- a/doc/PIPELINE_SERVING_CN.md +++ b/doc/PIPELINE_SERVING_CN.md @@ -705,9 +705,9 @@ Pipeline Serving支持低精度推理，CPU、GPU和TensoRT支持的精度类型 ## 5.日志追踪 Pipeline服务日志在当前目录的PipelineServingLogs目录下，有3种类型日志，分别是pipeline.log日志、pipeline.log.wf日志、pipeline.tracer日志。 -- pipeline.log日志 : 记录 debug & info日志信息 -- pipeline.log.wf日志 : 记录 warning & error日志 -- pipeline.tracer日志 : 统计各个阶段耗时、channel堆积信息 +- `pipeline.log` : 记录 debug & info日志信息 +- `pipeline.log.wf` : 记录 warning & error日志 +- `pipeline.tracer` : 统计各个阶段耗时、channel堆积信息 在服务发生异常时，错误信息会记录在pipeline.log.wf日志中。打印tracer日志要求在config.yml的DAG属性中添加tracer配置。 @@ -718,6 +718,39 @@ Pipeline中有2种id用以串联请求，分别时data_id和log_id，二者区 通常，Pipeline框架打印的日志会同时带上data_id和log_id。开启auto-batching后，会使用批量中的第一个data_id标记batch整体，同时框架会在一条日志中打印批量中所有data_id。 +### 5.2 日志滚动 +Pipeline的日志模块在`logger.py`中定义，使用了`logging.handlers.RotatingFileHandler`支持磁盘日志文件的轮换。根据不同文件级别和日质量分别设置了`maxBytes` 和 `backupCount`，当即将超出预定大小时，将关闭旧文件并打开一个新文件用于输出。 + +```python +"handlers": { + "f_pipeline.log": { + "class": "logging.handlers.RotatingFileHandler", + "level": "INFO", + "formatter": "normal_fmt", + "filename": os.path.join(log_dir, "pipeline.log"), + "maxBytes": 512000000, + "backupCount": 20, + }, + "f_pipeline.log.wf": { + "class": "logging.handlers.RotatingFileHandler", + "level": "WARNING", + "formatter": "normal_fmt", + "filename": os.path.join(log_dir, "pipeline.log.wf"), + "maxBytes": 512000000, + "backupCount": 10, + }, + "f_tracer.log": { + "class": "logging.handlers.RotatingFileHandler", + "level": "INFO", + "formatter": "tracer_fmt", + "filename": os.path.join(log_dir, "pipeline.tracer"), + "maxBytes": 512000000, + "backupCount": 5, + }, +}, +``` + +*** ## 6.性能分析与优化 diff --git a/python/examples/pipeline/bert/README.md b/python/examples/pipeline/bert/README.md new file mode 100644 index 00000000..6074aa2b --- /dev/null +++ b/python/examples/pipeline/bert/README.md @@ -0,0 +1,19 @@ +# Imagenet Pipeline WebService + +This document will takes Imagenet service as an example to introduce how to use Pipeline WebService. + +## Get model +``` +sh get_model.sh +``` + +## Start server + +``` +python3 web_service.py &>log.txt & +``` + +## RPC test +``` +python3 pipeline_rpc_client.py +``` diff --git a/python/examples/pipeline/bert/README_CN.md b/python/examples/pipeline/bert/README_CN.md new file mode 100644 index 00000000..ace7b76f --- /dev/null +++ b/python/examples/pipeline/bert/README_CN.md @@ -0,0 +1,19 @@ +# Imagenet Pipeline WebService + +这里以 Imagenet 服务为例来介绍 Pipeline WebService 的使用。 + +## 获取模型 +``` +sh get_model.sh +``` + +## 启动服务 + +``` +python3 web_service.py &>log.txt & +``` + +## 测试 +``` +python3 pipeline_rpc_client.py +``` diff --git a/python/examples/pipeline/imagenet/README.md b/python/examples/pipeline/imagenet/README.md index d0fa99e6..6fbe0c4c 100644 --- a/python/examples/pipeline/imagenet/README.md +++ b/python/examples/pipeline/imagenet/README.md @@ -10,10 +10,10 @@ sh get_model.sh ## Start server ``` -python resnet50_web_service.py &>log.txt & +python3 resnet50_web_service.py &>log.txt & ``` ## RPC test ``` -python pipeline_rpc_client.py +python3 pipeline_rpc_client.py ``` diff --git a/python/examples/pipeline/imagenet/README_CN.md b/python/examples/pipeline/imagenet/README_CN.md index 335c96b2..c204c3c6 100644 --- a/python/examples/pipeline/imagenet/README_CN.md +++ b/python/examples/pipeline/imagenet/README_CN.md @@ -10,11 +10,10 @@ sh get_model.sh ## 启动服务 ``` -python resnet50_web_service.py &>log.txt & +python3 resnet50_web_service.py &>log.txt & ``` ## 测试 ``` -python pipeline_rpc_client.py +python3 pipeline_rpc_client.py ``` - diff --git a/python/examples/pipeline/imdb_model_ensemble/README.md b/python/examples/pipeline/imdb_model_ensemble/README.md index f90024f6..c72eab66 100644 --- a/python/examples/pipeline/imdb_model_ensemble/README.md +++ b/python/examples/pipeline/imdb_model_ensemble/README.md @@ -8,12 +8,12 @@ sh get_data.sh ## Start servers ``` -python -m paddle_serving_server.serve --model imdb_cnn_model --port 9292 &> cnn.log & -python -m paddle_serving_server.serve --model imdb_bow_model --port 9393 &> bow.log & -python test_pipeline_server.py &>pipeline.log & +python3 -m paddle_serving_server.serve --model imdb_cnn_model --port 9292 &> cnn.log & +python3 -m paddle_serving_server.serve --model imdb_bow_model --port 9393 &> bow.log & +python3 test_pipeline_server.py &>pipeline.log & ``` ## Start clients ``` -python test_pipeline_client.py +python3 test_pipeline_client.py ``` diff --git a/python/examples/pipeline/imdb_model_ensemble/README_CN.md b/python/examples/pipeline/imdb_model_ensemble/README_CN.md index fd478529..79ed5c0c 100644 --- a/python/examples/pipeline/imdb_model_ensemble/README_CN.md +++ b/python/examples/pipeline/imdb_model_ensemble/README_CN.md @@ -8,12 +8,12 @@ sh get_data.sh ## 启动服务 ``` -python -m paddle_serving_server.serve --model imdb_cnn_model --port 9292 &> cnn.log & -python -m paddle_serving_server.serve --model imdb_bow_model --port 9393 &> bow.log & -python test_pipeline_server.py &>pipeline.log & +python3 -m paddle_serving_server.serve --model imdb_cnn_model --port 9292 &> cnn.log & +python3 -m paddle_serving_server.serve --model imdb_bow_model --port 9393 &> bow.log & +python3 test_pipeline_server.py &>pipeline.log & ``` ## 启动客户端 ``` -python test_pipeline_client.py +python3 test_pipeline_client.py ``` diff --git a/python/examples/pipeline/ocr/README.md b/python/examples/pipeline/ocr/README.md index de7bcaa2..4c669b3c 100644 --- a/python/examples/pipeline/ocr/README.md +++ b/python/examples/pipeline/ocr/README.md @@ -4,11 +4,13 @@ This document will take OCR as an example to show how to use Pipeline WebService to start multi-model tandem services. +This OCR example only supports Process OP. + ## Get Model ``` -python -m paddle_serving_app.package --get_model ocr_rec +python3 -m paddle_serving_app.package --get_model ocr_rec tar -xzvf ocr_rec.tar.gz -python -m paddle_serving_app.package --get_model ocr_det +python3 -m paddle_serving_app.package --get_model ocr_det tar -xzvf ocr_det.tar.gz ``` @@ -18,14 +20,16 @@ wget --no-check-certificate https://paddle-serving.bj.bcebos.com/ocr/test_imgs.t tar xf test_imgs.tar ``` -## Start Service +## Run services + +### 1.Start a single server and client. ``` -python web_service.py &>log.txt & +python3 web_service.py &>log.txt & ``` -## Test +Test ``` -python pipeline_http_client.py +python3 pipeline_http_client.py ``` + + +### 2.Run benchmark +``` +python3 web_service.py &>log.txt & +``` + +Test +``` +sh benchmark.sh +``` diff --git a/python/examples/pipeline/ocr/README_CN.md b/python/examples/pipeline/ocr/README_CN.md index c7058e02..c6c2064a 100644 --- a/python/examples/pipeline/ocr/README_CN.md +++ b/python/examples/pipeline/ocr/README_CN.md @@ -3,12 +3,13 @@ ([English](./README.md)|简体中文) 本文档将以 OCR 为例，介绍如何使用 Pipeline WebService 启动多模型串联的服务。 +本示例仅支持进程OP模式。 ## 获取模型 ``` -python -m paddle_serving_app.package --get_model ocr_rec +python3 -m paddle_serving_app.package --get_model ocr_rec tar -xzvf ocr_rec.tar.gz -python -m paddle_serving_app.package --get_model ocr_det +python3 -m paddle_serving_app.package --get_model ocr_det tar -xzvf ocr_det.tar.gz ``` @@ -19,13 +20,15 @@ tar xf test_imgs.tar ``` ## 启动 WebService + +### 1.启动单server、单client ``` -python web_service.py &>log.txt & +python3 web_service.py &>log.txt & ``` ## 测试 ``` -python pipeline_http_client.py +python3 pipeline_http_client.py ``` + +### 2.启动 benchmark +``` +python3 web_service.py &>log.txt & +``` + +Test +``` +sh benchmark.sh +``` diff --git a/python/pipeline/logger.py b/python/pipeline/logger.py index b566c012..ebf31564 100644 --- a/python/pipeline/logger.py +++ b/python/pipeline/logger.py @@ -42,22 +42,28 @@ logger_config = { }, "handlers": { "f_pipeline.log": { - "class": "logging.FileHandler", + "class": "logging.handlers.RotatingFileHandler", "level": "INFO", "formatter": "normal_fmt", "filename": os.path.join(log_dir, "pipeline.log"), + "maxBytes": 512000000, + "backupCount": 20, }, "f_pipeline.log.wf": { - "class": "logging.FileHandler", + "class": "logging.handlers.RotatingFileHandler", "level": "WARNING", "formatter": "normal_fmt", "filename": os.path.join(log_dir, "pipeline.log.wf"), + "maxBytes": 512000000, + "backupCount": 10, }, "f_tracer.log": { - "class": "logging.FileHandler", + "class": "logging.handlers.RotatingFileHandler", "level": "INFO", "formatter": "tracer_fmt", "filename": os.path.join(log_dir, "pipeline.tracer"), + "maxBytes": 512000000, + "backupCount": 5, }, }, "loggers": { -- GitLab