### Install Go Environment

Source: https://paddlepaddle.github.io/Serving/doc/DEPLOY.html

Commands to download, extract, and configure the Go 1.12 environment.

```bash
wget https://studygolang.com/dl/golang/go1.12.7.linux-amd64.tar.gz --no-check-certificate
```

```bash
tar zxvf go1.12.7.linux-amd64.tar.gz -C /path/to/go
```

```bash
export GOROOT=/path/to/go
export PATH=/path/to/go/bin:$PATH
```

--------------------------------

### Start cube-transfer service

Source: https://paddlepaddle.github.io/Serving/doc/DEPLOY.html

Launch the service using the specified port, log level, and configuration file path.

```bash
./cube-transfer -p 8099 -l 4 --config conf/transfer.conf
```

--------------------------------

### Configure Cube Server

Source: https://paddlepaddle.github.io/Serving/doc/DEPLOY.html

Example configuration for the cube server, including port settings and memory mode.

```text
--port=8000
--dict_split=1
--in_mem=true
```

--------------------------------

### Start Cube Server and Agent

Source: https://paddlepaddle.github.io/Serving/doc/DEPLOY.html

Commands to launch the cube server and the associated cube-agent with a specified port.

```bash
nohup bin/cube &
nohup bin/cube-agent -P 8001 &
```

--------------------------------

### Install Volcano

Source: https://paddlepaddle.github.io/Serving/doc/ELASTIC_CTR.html

Deploy the Volcano batch scheduling system to the Kubernetes cluster.

```bash
$ kubectl apply -f https://raw.githubusercontent.com/volcano-sh/volcano/master/installer/volcano-development.yaml
```

--------------------------------

### Install and Manage Volcano

Source: https://paddlepaddle.github.io/Serving/doc/DEPLOY.html

Commands to install the Volcano framework and handle potential installation failures.

```bash
kubectl apply -f https://raw.githubusercontent.com/volcano-sh/volcano/master/installer/volcano-development.yaml
```

```bash
kubectl delete -f https://raw.githubusercontent.com/volcano-sh/volcano/master/installer/volcano-development.yaml
```

--------------------------------

### GET /instance/status

Source: https://paddlepaddle.github.io/Serving/doc/DEPLOY.html

Retrieves the current status of the cube-transfer instance.

```APIDOC
## GET /instance/status

### Description
Retrieves the current status of the cube-transfer instance.

### Method
GET

### Endpoint
/instance/status
```

--------------------------------

### GET /dict/info

Source: https://paddlepaddle.github.io/Serving/doc/DEPLOY.html

Retrieves the current status of the dictionary.

```APIDOC
## GET /dict/info

### Description
Retrieves the current status of the dictionary managed by the cube-transfer service.

### Method
GET

### Endpoint
/dict/info
```

--------------------------------

### Example Plaintext Input Format

Source: https://paddlepaddle.github.io/Serving/doc/DEPLOY.html

The required format for plaintext input files, where each line contains a key-value pair separated by a colon.

```text
1:1
2:2
10:10 11 12
11:this is eleven
12:value can string
1676869128226002114:48241    37064        91    -539    114    51    -122    269    229    -134    -282
1657749292782759014:167    40        98    27    117    10    -29    15    74    67    -54
```

--------------------------------

### GET /dict/deploy/history

Source: https://paddlepaddle.github.io/Serving/doc/DEPLOY.html

Retrieves the deployment history from the most recent base to the currently deploying delta.

```APIDOC
## GET /dict/deploy/history

### Description
Retrieves the deployment history from the most recent base to the currently deploying delta.

### Method
GET

### Endpoint
/dict/deploy/history
```

--------------------------------

### 更新模型文件

Source: https://paddlepaddle.github.io/Serving/doc/DEPLOY.html

下载并替换Serving目录下的模型文件，随后重启服务。

```bash
$ cd data/model/paddle/fluid
$ wget http://${FILE_SERVER_IP}:${FILE_SERVER_PORT}/data/ctr_model.tar.gz # `FILE_SERVER_IP`与`FILE_SERVER_PORT`请参考1.4节获取。
$ tar zxvf ctr_model.tar.gz # 假设解压出一个inference_only目录
$ rm -rf ctr_prediction/*   # 删除旧的ctr_prediction目录下内容
$ cp inference_only/* ctr_prediction/
$ cd ../../../../           # 切换至serving所在目录
$ ls
bin  conf  data  kvdb  log
$ killall serving           # 杀死旧的serving进程
$ bin/serving &             # 重启serving
```

--------------------------------

### View Cube-Builder Help and Flags

Source: https://paddlepaddle.github.io/Serving/doc/DEPLOY.html

Display the available command-line flags and configuration options for the cube-builder tool.

```bash
cube-builder: Usage : ./cube-builder --help

Flags from /home/work/cube-builder/src/main.cpp:
-cur_version (current version, no need) type: int32 default: 0                //单机builder模式下不需要
-depend_version (depend version, job mode delta need) type: int32 default: 0  //单机builder base模式下不需要，patch模式找到meta_info里的base的key
-dict_name (dict name, no need) type: string default: ""                      //词典名，单机builder模式下不加默认空，用来和版本拼接生成索引文件名
-input_path (source data input path) type: string default: ""               //必须，源数据所在路径，仅支持本地路径
-job_mode (job mode base/delta default:base) type: string default: "base"    //默认base模式，delta模式：-job_mode=delta
-last_version (last version, job mode delta need) type: int32 default: 0     //单机builder base模式下不需要，patch模式找到meta_info里的base的id
-master_address (master address, no need) type: string default: ""           //单机builder模式不需要，会把index meta信息写到本地output/meta_info目录下
-only_build (wheather build need transfer) type: bool default: true         //单机builder模式不需要，代表是不是单机builder，如果false会向master_address发送请求，将index meta信息写到远程
-output_path (source data input path) type: string default: ""       //必须，索引建库数据的输出路径，仅支持本地路径
-shard_num (shard num) type: int32 default: -1             //必须，数据切分的分片数量
```

--------------------------------

### Deploy Components via Script

Source: https://paddlepaddle.github.io/Serving/doc/ELASTIC_CTR.html

Execute the automated deployment script for all components.

```bash
$ bash paddle-suite.sh
```

--------------------------------

### Execute Training Tasks

Source: https://paddlepaddle.github.io/Serving/doc/DEPLOY.html

Commands to set up permissions, label nodes, and launch the CTR training job.

```bash
kubectl create -f defaultserviceaccountclusterrole.yaml
```

```bash
kubectl label node ${NODE NAME} nodeType=model
```

```bash
kubectl apply -f volcano-ctr-demo-baiduyun.yaml
```

--------------------------------

### 配置model_toolkit.prototxt

Source: https://paddlepaddle.github.io/Serving/doc/DEPLOY.html

定义Serving引擎配置，包括模型路径及CTR预估相关的远程参数服务设置。

```protobuf
engines {
name: "image_classification_resnet"
type: "FLUID_CPU_NATIVE_DIR"
reloadable_meta: "./data/model/paddle/fluid_time_file"
reloadable_type: "timestamp_ne"
model_data_path: "./data/model/paddle/fluid/SE_ResNeXt50_32x4d"
runtime_thread_num: 0
batch_infer_size: 0
enable_batch_align: 0
}
engines {
name: "text_classification_bow"
type: "FLUID_CPU_ANALYSIS_DIR"
reloadable_meta: "./data/model/paddle/fluid_time_file"
reloadable_type: "timestamp_ne"
model_data_path: "./data/model/paddle/fluid/text_classification_lstm"
runtime_thread_num: 0
batch_infer_size: 0
enable_batch_align: 0
}

engines {
name: "ctr_prediction"
type: "FLUID_CPU_ANALYSIS_DIR"
reloadable_meta: "./data/model/paddle/fluid_time_file"
reloadable_type: "timestamp_ne"
model_data_path: "./data/model/paddle/fluid/ctr_prediction"
runtime_thread_num: 0
batch_infer_size: 0
enable_batch_align: 0
sparse_param_service_type: REMOTE
sparse_param_service_table_name: "test_dict"
}
```

--------------------------------

### Manage File Server

Source: https://paddlepaddle.github.io/Serving/doc/ELASTIC_CTR.html

Commands to deploy and verify the file server.

```bash
$ kubectl apply -f fileserver.yaml
```

```bash
$ kubectl get pod
```

```bash
$ kubectl get service
```

--------------------------------

### Manage Cube Parameter Server

Source: https://paddlepaddle.github.io/Serving/doc/ELASTIC_CTR.html

Commands to deploy and verify the Cube sparse parameter server.

```bash
$ kubectl apply -f cube.yaml
```

```bash
$ kubectl get service
```

--------------------------------

### Execute Client Binary

Source: https://paddlepaddle.github.io/Serving/doc/DEPLOY.html

Run the compiled client binary to initiate the prediction service, with logs generated in the ./log directory.

```bash
./bin/ctr_prediction
```

--------------------------------

### Configure kv_to_seqfile Paths

Source: https://paddlepaddle.github.io/Serving/doc/DEPLOY.html

Set the base donefile and source file paths within the kv_to_seqfile.py script.

```python
BASE_DONEFILE = DATA_PATH + "donefile/base.txt"  #base donefile文件地址
SOURCE_FILE = './source/file.txt' #明文源数据路径
```

--------------------------------

### Compile Cube Components

Source: https://paddlepaddle.github.io/Serving/doc/DEPLOY.html

Standard compilation steps for Paddle Serving including Cube components, requiring a Go compiler.

```bash
$ git clone https://github.com/PaddlePaddle/Serving.git
$ cd Serving
$ mkdir build
$ cd build
$ cmake -DWITH_GPU=OFF .. # 不需要GPU
$ make -jN                # 这里可修改并发编译线程数
$ make install
$ cd output/
$ ls bin
cube  cube-builder  cube-transfer  pdcodegen
$ ls conf
gflags.conf  transfer.conf
```

--------------------------------

### Run Cube-Builder in Base Mode

Source: https://paddlepaddle.github.io/Serving/doc/DEPLOY.html

Execute the builder in base mode using absolute paths for input and output.

```bash
./cube-builder -input_path=${source} -output_path=${output} -shard_num=2 -dict_name=test
```

--------------------------------

### Verify cube-transfer file structure

Source: https://paddlepaddle.github.io/Serving/doc/DEPLOY.html

The directory structure should include the executable, configuration, and necessary working directories.

```bash
$ tree
.
|-- cube-transfer
|-- output
|-- tmp
`-- conf
|-- transfer.conf
```

--------------------------------

### Manage Paddle Serving

Source: https://paddlepaddle.github.io/Serving/doc/ELASTIC_CTR.html

Commands to deploy and verify the Paddle Serving service.

```bash
$ kubectl apply -f paddleserving.yaml
```

```bash
$ kubectl get pod
```

```bash
$ kubectl get service
```

--------------------------------

### 编译Paddle Serving

Source: https://paddlepaddle.github.io/Serving/doc/DEPLOY.html

从源码克隆并编译Paddle Serving，生成可执行文件及配置文件。

```bash
$ git clone https://github.com/PaddlePaddle/Serving.git
$ cd Serving
$ mkdir build
$ cd build
$ cmake -DWITH_GPU=OFF .. # 不需要GPU
$ make -jN                # 这里可修改并发编译线程数
$ make install
$ cd output/demo/serving
$ ls
bin  conf  data  kvdb
```

--------------------------------

### Cube-Builder Directory Structure

Source: https://paddlepaddle.github.io/Serving/doc/DEPLOY.html

The expected directory layout for the cube-builder tool.

```bash
$ tree
`-- cube-builder
|-- source
|   `-- test_seqfile
|-- output
```

--------------------------------

### Manage Cube Transfer Tool

Source: https://paddlepaddle.github.io/Serving/doc/ELASTIC_CTR.html

Commands to deploy, verify, and monitor the Cube transfer tool.

```bash
$ kubectl apply -f transfer.yaml
```

```bash
$ kubectl get pod
```

```bash
$ kubectl logs cube-transfer
```

--------------------------------

### Configure kubectl for Kubernetes

Source: https://paddlepaddle.github.io/Serving/doc/ELASTIC_CTR.html

Move the downloaded cluster configuration file to the default kubectl path.

```bash
$ mv kubectl.conf  ~/.kube/config
```

--------------------------------

### Cube File Structure

Source: https://paddlepaddle.github.io/Serving/doc/DEPLOY.html

Expected file structure after copying binaries and configuration files to physical nodes.

```text
$ tree
.
|-- bin
|   `-- cube
|   `-- cube-agent
|-- conf
|   `-- gflags.conf
```

--------------------------------

### Execute Redis Performance Test

Source: https://paddlepaddle.github.io/Serving/doc/ELASTIC_CTR.html

Run the Redis client benchmark tool with specified host, thread count, request count, and batch size.

```bash
$ ./get_values -h 192.168.1.1 -t 3 -r 10000 -b 1000
```

--------------------------------

### Base Mode Output Directory Structure

Source: https://paddlepaddle.github.io/Serving/doc/DEPLOY.html

The resulting directory structure after running a base mode build.

```bash
$ tree
`-- cube-builder
|-- source
|   `-- test_seqfile
`-- output
|-- 1565323045_1565323045
|   |-- test_part0
|   |   |-- data.0
|   |   |-- data.n
|   |   |-- index.0
|   |   `-- index.n
|   |-- test_part0.tar
|   |-- test_part0.tar.md5
|   |-- test_part1
|   |   |-- data.0
|   |   |-- data.n
|   |   |-- index.0
|   |   `-- index.n
|   |-- test_part1.tar
|   `-- test_part1.tar.md5
`-- meta_info
|-- 1565323045_1565323045_0_0.json
`-- 1565323045_1565323045_1_0.json
```

--------------------------------

### 修改gflags配置

Source: https://paddlepaddle.github.io/Serving/doc/DEPLOY.html

启用Cube服务支持。

```bash
--enable_cube=true

```

--------------------------------

### Run Cube-Builder in Delta Mode

Source: https://paddlepaddle.github.io/Serving/doc/DEPLOY.html

Execute the builder in delta mode, requiring dependency on previous version IDs.

```bash
./cube-builder -input_path=${source} -output_path=${output} -shard_num=2 -depend_version=1565323045 -last_version=1565323045 -job_mode=delta -dict_name=test
```

--------------------------------

### HTTP File Server Configuration

Source: https://paddlepaddle.github.io/Serving/doc/DEPLOY.html

YAML configuration for deploying a static file server and a load balancer service to expose model files.

```yaml
apiVersion: v1
kind: Pod
metadata:
  name: file-server
  labels:
    app: file-server
spec:
  nodeSelector:
    nodeType: model
  volumes:
  - hostPath:
      path: /home/work
      type: ""
    name: file-home
  containers:
  - name: file-server
    image: halverneus/static-file-server
    ports:
    - containerPort: 8080
    volumeMounts:
    - mountPath: /web
      name: file-home
---
kind: Service
apiVersion: v1
metadata:
  name: loadbalancer
spec:
  type: LoadBalancer
  ports:
    - name: file-server
      port: 8080
      targetPort: 8080
  selector:
    app: file-server
```

--------------------------------

### 下载模型参数

Source: https://paddlepaddle.github.io/Serving/doc/DEPLOY.html

通过HTTP服务下载CTR预估模型压缩包。

```bash
wget http://${FILE_SERVER_IP}:${FILE_SERVER_PORT}/data/ctr_model.tar.gz

```

--------------------------------

### Configure RBAC for Training

Source: https://paddlepaddle.github.io/Serving/doc/DEPLOY.html

YAML configuration to create a ClusterRole and ClusterRoleBinding for the default service account.

```yaml
kind: ClusterRole
apiVersion: rbac.authorization.k8s.io/v1
metadata:
name: default
namespace: default
rules:
- apiGroups: [""]
resources: ["pods"]
verbs: ["get", "list", "watch"]

---
kind: ClusterRoleBinding
apiVersion: rbac.authorization.k8s.io/v1
metadata:
name: default
namespace: default
subjects:
- kind: ServiceAccount
name: default
namespace: default
roleRef:
kind: ClusterRole
name: default
apiGroup: rbac.authorization.k8s.io
```

--------------------------------

### Execute Distributed Training

Source: https://paddlepaddle.github.io/Serving/doc/ELASTIC_CTR.html

Commands to deploy and monitor the CTR distributed training job.

```bash
$ kubectl apply -f ctr.yaml
```

```bash
$ kubectl get pod
```

```bash
$ kubectl logs edl-demo-trainer-0
```

--------------------------------

### Read Seqfile with kvtool

Source: https://paddlepaddle.github.io/Serving/doc/DEPLOY.html

Use the kvtool.py script to read and output the contents of a seqfile.

```python
python kvtool.py /home/work/test/seqfile
```

--------------------------------

### Configure cube-transfer

Source: https://paddlepaddle.github.io/Serving/doc/DEPLOY.html

The configuration file conf/transfer.conf requires absolute paths and specific network settings for the transfer service and cube agents.

```ini
[default]
dict_name: test_dict                                # 词典名
mode: base_delta                                    # 配送模式base_only/base_delta
download_mode: http                                 # 配送方式，可以选择http或ftp
wget_port: 80                                       # http服务的端口
buildtool_local: /path/to/cube-builder              # builder工具位置，必须在本地，绝对路径
donefile_address: http://${FILE_SERVER_IP}:${FILE_SERVER_PORT}/data/ctr_cube/donefile/ # donefile路径，${FILE_SERVER_IP}:${FILE_SERVER_PORT}为1.4节搭建的file server地址。文件夹内包含base.txt, patch.txt和一批Hadoop SequenceFile文件
output_address: /some/path/to/output      # builder产出的数据索引输出位置
tmp_address: /some/path/to/tmp            # transfer工具运行中临时文件存放位置
shard_num: 2                                        # 分片数
copy_num: 1                                         # 每片副本数
deploy_path: /home/work/test_dict                   # 不用修改                          
transfer_address: 10.10.10.5                        # cube-transfer本机的ip

[cube_agent]
agent0_0: 192.168.1.1:8001                       # 0号分片0号副本的agent ip:port
cube0_0: 192.168.1.1:8000:/path/to/cube          # 0号分片0号副本的cube，该路径下会存放配送的数据，需要与启动cube可执行文件时所在的路径一致 格式：ip:port:deploy_path
agent1_0: 192.168.1.2:8001                       # 1号分片0号副本的agent ip:port
cube1_0: 192.168.1.2:8000:/path/to/cube          # 1号分片0号副本的cube，该路径下会存放配送的数据，需要与启动cube可执行文件时所在的路径一致 格式：ip:port:deploy_path
```

--------------------------------

### CTR预估模型配置项

Source: https://paddlepaddle.github.io/Serving/doc/DEPLOY.html

CTR预估模型中关于远程参数服务的关键配置。

```protobuf
sparse_param_service_type: REMOTE
sparse_param_service_table_name: "test_dict"

```

--------------------------------

### Deploy and Execute CTR Prediction Client

Source: https://paddlepaddle.github.io/Serving/doc/ELASTIC_CTR.html

Commands to apply the client configuration and execute the prediction task within the container.

```bash
$ kubectl apply -f paddleclient.yaml
```

```bash
# 进入容器
$ kubectl exec -ti pdservingclient /bin/bash

# 此命令在容器内执行
$ bin/ctr_prediction
```

--------------------------------

### Update and Push Docker Images

Source: https://paddlepaddle.github.io/Serving/doc/ELASTIC_CTR.html

Methods for updating the environment by committing changes to a container or rebuilding from a Dockerfile.

```bash
$ docker commit ${DOCKER_CONTAINER_NAME} ${DOCKER_IMAGE_NAME}
$ docker push  ${DOCKER_IMAGE_NAME}
```

```bash
$ docker build -t ${DOCKER_IMAGE_NAME} .
$ docker push  ${DOCKER_IMAGE_NAME}
```

--------------------------------

### Delta Mode Output Directory Structure

Source: https://paddlepaddle.github.io/Serving/doc/DEPLOY.html

The resulting directory structure after running a delta mode build.

```bash
$ tree
`-- cube-builder
|-- source
|   `-- test_seqfile
`-- output
|-- 1565323045_1565323045
|   |-- test_part0
|   |   |-- data.0
|   |   |-- data.n
|   |   |-- index.0
|   |   `-- index.n
|   |-- test_part0.tar
|   |-- test_part0.tar.md5
|   |-- test_part1
|   |   |-- data.0
|   |   |-- data.n
|   |   |-- index.0
|   |   `-- index.n
|   |-- test_part1.tar
|   `-- test_part1.tar.md5
|-- 1565323045_1565326078
|   |-- test_part0
|   |-- data.0
|   |   |-- data.n
|   |   |-- index.0
|   |   `-- index.n
|   |-- test_part0.tar
|   |-- test_part0.tar.md5
|   |-- test_part1
|   |   |-- data.0
|   |   |-- data.n
|   |   |-- index.0
|   |   `-- index.n
|   |-- test_part1.tar
|   `-- test_part1.tar.md5
`-- meta_info
|-- 1565323045_1565323045_0_0.json
|-- 1565323045_1565323045_0_0.json
|-- 1565326078_1565323045_0_0.json
`-- 1565326078_1565323045_1_0.json
```

--------------------------------

### Configure Helm TLS Environment Variables

Source: https://paddlepaddle.github.io/Serving/doc/DEPLOY.html

Set these environment variables on the development machine to enable secure communication with the Tiller service.

```bash
export HELM_TLS_ENABLE=true

export HELM_TLS_CA_CERT=/etc/kubernetes/pki/ca.pem

export HELM_TLS_CERT=/etc/kubernetes/pki/admin.pem

export HELM_TLS_KEY=/etc/kubernetes/pki/admin-key.pem
```

--------------------------------

### Configure CTR Prediction Service

Source: https://paddlepaddle.github.io/Serving/doc/DEPLOY.html

Modify the predictors.prototxt file to define the service name, endpoint router, and cluster naming configuration.

```protobuf
predictors {
name: "ctr_prediction_service"
service_name: "baidu.paddle_serving.predictor.ctr_prediction.CTRPredictionService"
endpoint_router: "WeightedRandomRender"
weighted_random_render_conf {
variant_weight_list: "50"
}
variants {
tag: "var1"
naming_conf {
cluster: "list://127.0.0.1:8010"
}
}
}
```

--------------------------------

### 配置Cube客户端

Source: https://paddlepaddle.github.io/Serving/doc/DEPLOY.html

定义用于Cube客户端API的配置文件，包含分片、超时及节点列表信息。

```json
[{
"dict_name": "test_dict",
"shard": 2,
"dup": 1,
"timeout": 200,
"retry": 3,
"backup_request": 100,
"type": "ipport_list",
"load_balancer": "rr",
"nodes": [{
"ipport_list": "list://192.168.1.1:8000"
},{
"ipport_list": "list://192.168.1.2:8000"
}]
}]
```

--------------------------------

### Set Kubernetes Role Binding

Source: https://paddlepaddle.github.io/Serving/doc/ELASTIC_CTR.html

Grant necessary permissions for pod-to-pod communication.

```bash
$ kubectl create rolebinding default-view --clusterrole=view --serviceaccount=default:default --namespace=default
```

--------------------------------

### Access Sparse Parameter Donefile

Source: https://paddlepaddle.github.io/Serving/doc/DEPLOY.html

The URL path for the donefile used to describe the sparse parameter directory structure.

```text
http://${FILE_SERVER_IP}:${FILE_SERVER_PORT}/data/ctr_cube/donefile/base.txt
```

--------------------------------

### Define donefile format

Source: https://paddlepaddle.github.io/Serving/doc/DEPLOY.html

The donefile uses JSON format to track base and patch data updates for the cube-transfer service.

```json
{"id":"1562000400","key":"1562000400","input":"/home/work/test_data/input/seqfile"}
```

```json
{"id":"1562000401","key":"1562000400","input":"/home/work/test_data/input/seqfile"}
{"id":"1562000402","key":"1562000400","input":"/home/work/test_data/input/seqfile"}
```

--------------------------------

### Access Dense Model Archive

Source: https://paddlepaddle.github.io/Serving/doc/DEPLOY.html

The URL path for the dense model tarball hosted on the HTTP file server.

```text
http://${FILE_SERVER_IP}:${FILE_SERVER_PORT}/data/ctr_model.tar.gz
```

--------------------------------

### Label Kubernetes Node

Source: https://paddlepaddle.github.io/Serving/doc/ELASTIC_CTR.html

Assign a specific label to a node for model output and file services.

```bash
$ kubectl label nodes $NODE_NAME nodeType=model
```

--------------------------------

### Verify Kubernetes Node Status

Source: https://paddlepaddle.github.io/Serving/doc/ELASTIC_CTR.html

Check the status of nodes in the cluster.

```bash
$ kubectl get node
```

=== COMPLETE CONTENT === This response contains all available snippets from this library. No additional content exists. Do not make further requests.