pipeline 分布式任务调度器
目标: 基于docker的布式任务调度器, 比quartzs
,xxl-job
更强大的分布式任务调度器。
可以将要执行的任务打包为docker镜像,或者选择已有镜像,自定义脚本程序,通过pipeline框架来实现调度。
开源地址: https://github.com/jadepeng/docker-pipeline
架构
- pipeline master 中心节点,管理和调度任务
- pipeline agent 执行任务的节点,接收到任务后,调用docker执行pipeline任务
功能特性 && TODO List
- [x] 分布式框架,高可用,服务注册与状态维护
- [x] Agent执行任务
- [x] rolling日志接口
- [x] 运行老版本pipeline任务
- [x] 支持定时执行任务(固定周期和cron表达式)
- [ ] 快速创建任务,支持python、node等脚本程序直接执行
- [x] python、java等基础镜像
- [x] 快速docker镜像任务API
- [ ] 快速创建脚本任务
- [ ] 根据资源配额(内存、CPU)调度任务, 运行任务需要指定资源配额
- [ ] agent 增加label标识,调度时可以调度到指定label的agent,比如gpu=true
- [ ] 增加任务管理web, 管理提交任务、查询运行日志等
- [x] 复用腾讯bk-job 网页
- [ ] 修改bk-job前端,适配pipeline
进展
2021.07.31
- 支持定时执行任务(固定周期和cron表达式)
- 增加分布式mongodb锁,多master时,同时只能有一个master schedule任务
2021.07.28
- 新增运行老版本pipeline任务能力
- 增加日志接口
2021.07.27
- 引入bk-job的ui,待修改
2021.07.21
- Master 调用 agent执行任务
- agnet 启动docker执行任务
2021.07.19
- 基于jhipster搭建框架
- 分布式实现
数据结构
一个pipeline 任务:
- 支持多个pipelineTask
- 一个pipelineTask 包含多个Step
@Data
public class Pipeline {
@Id
private String id;
private String name;
@JSONField(name = "pipeline")
private List<PipelineTask> pipelineTasks = new ArrayList<>();
private List<Network> networks = Lists.newArrayList(new Network());
private List<Volume> volumes = Lists.newArrayList(new Volume());
private String startNode;
/**
* 调度类型:
* 1) CRON, 设置cronExpression
* 2) FIX_RATE, 设置fixRateInSeconds
*/
private ScheduleType scheduleType = ScheduleType.NONE;
/**
* CRON表达式,在scheduleType=CRON 时生效
*/
private String cronExpression;
/**
* 固定周期运行,比如每隔多少s,在scheduleType=FIX_RATE 时生效
*/
private int fixRateInSeconds;
/**
* 是否需要调度,为true时,才调度
*/
@Indexed
private boolean enableTrigger;
private long lastTriggerTime;
@Indexed
private long nextTriggerTime;
/**
* 执行超时时间
*/
private int executorTimeout;
/**
* 重试次数
*/
private int executorFailRetryCount;
/**
* 内存限制
*/
private String memory;
/**
* CPU 限制
*/
private String cpu;
@Data
@Builder
public static class PipelineTask {
/**
* 名称
*/
String name;
/**
* 别名
*/
String alias;
/**
* 依赖的pipelines,必须依赖的执行完成才能运行该PipelineTask
*/
List<String> dependencies;
/**
* 任务步骤,顺序执行
*/
List<Step> steps;
}
@Data
public static class Network {
String name = "pipeline_default";
String driver = "bridge";
}
@Data
public static class Volume {
String name = "pipeline_default";
String driver = "local";
}
@Data
public static class StepNetwork {
private String name;
private List<String> aliases = Lists.newArrayList("default");
public StepNetwork(String name) {
this.name = name;
}
}
}
举例:
代码语言:javascript复制{
"_id" : "29103d5e4a77409b9f6050eea8110bb3",
"name" : "docker image pipeline",
"pipelineTasks" : [
{
"name" : "docker image pipeline",
"steps" : [
{
"name" : "defaultJob",
"image" : "java-pipeline:1.0.1",
"workingDir" : "/workspace",
"environment" : {},
"networks" : [
{
"name" : "pipeline_default",
"aliases" : [
"default"
]
}
],
"onSuccess" : false,
"authConfig" : {}
}
]
}
],
"networks" : [
{
"name" : "pipeline_default",
"driver" : "bridge"
}
],
"volumes" : [
{
"name" : "pipeline_default",
"driver" : "local"
}
],
"cronExpression" : "0 0 * * * ?",
"fixRateInSeconds" : 0,
"scheduleType" : "CRON",
"enableTrigger" : true,
"lastTriggerTime" : 1627744509047,
"nextTriggerTime" : 1627747200000,
"executorTimeout" : 0,
"executorFailRetryCount" : 0,
"isAvailable" : 1,
"runningPipelines" : [],
"finishedPipeliens" : [],
"created_by" : "admin",
"created_date" : "2021-07-20T04:33:16.477Z",
"last_modified_by" : "system",
"last_modified_date" : "2021-07-31T15:15:09.048Z"
}
使用说明
安装部署
编译
使用mvn编译
代码语言:javascript复制mvn package -DskipTests
部署master
根据需要,修改master的prod配置文件application-prod.yml
。
包含kafka
配置,server
端口,mongodb
地址,jwt secret
配置。
mongodb 会自动新建collection和初始化数据,无需手动导入数据。
代码语言:javascript复制kafka:
producer:
bootstrap-servers: 127.0.0.1:9092
retries: 3
batch-size: 2000
buffer-memory: 33554432
consumer:
group-id: consumer-pipeline
auto-offset-reset: earliest
enable-auto-commit: true
bootstrap-servers: 172.31.161.38:9092
server:
port: 8080
spring:
data:
mongodb:
uri: mongodb://127.0.0.1:28017
database: pipeline
jhipster:
security:
authentication:
jwt:
base64-secret:
注意master的jwt secret需要和agent的保持一致。
配置好后,启动:
代码语言:javascript复制nohup java -jar pipeline-master-$version.jar --spring.profiles.active=prod &
可以将application-prod.yml
放在和jar包同一目录。
部署agent
根据需要,修改master的prod配置文件application-prod.yml
。
包含:
- eureka的defaultZone,配置master的地址
- 端口
- docker地址
- docker-tls-verify: 是否启动tls验证
- docker-cert-path:启动tls验证的ca证书
- pipeline-log-path: 运行日志存储路径
eureka:
instance:
prefer-ip-address: true
client:
service-url:
defaultZone: http://admin:${jhipster.registry.password}@127.0.0.1:8080/eureka/
server:
port: 8081
application:
docker-server:
docker-tls-verify: true
docker-cert-path: /mnt/parastor/pipeline/ca/
pipeline-log-path: /mnt/parastor/pipeline/logs/
jhipster:
security:
authentication:
jwt:
base64-secret:
执行老版本任务
代码语言:javascript复制POST /api/pipelines/exec-old
Body:
代码语言:javascript复制{
"networks":[
{
"driver":"bridge",
"name":"pipeline_network_3eac4b36209a41e58a5f22dd403fee50"
}
],
"pipeline":[
{
"alias":"Word",
"dependencies":[],
"name":"pipeline_task_3eac4b36209a41e58a5f22dd403fee50_1",
"nextPipelines":[],
"steps":[
{
"alias":"Word",
"auth_config":{},
"command":[
"echo $CI_SCRIPT | base64 -d | /bin/bash -e"
],
"entrypoint":[
"/bin/bash",
"-c"
],
"environment":{
"CI_SCRIPT":"CmlmIFsgLW4gIiRDSV9ORVRSQ19NQUNISU5FIiBdOyB0aGVuCmNhdCA8PEVPRiA ICRIT01FLy5uZXRyYwptYWNoaW5lICRDSV9ORVRSQ19NQUNISU5FCmxvZ2luICRDSV9ORVRSQ19VU0VSTkFNRQpwYXNzd29yZCAkQ0lfTkVUUkNfUEFTU1dPUkQKRU9GCmNobW9kIDA2MDAgJEhPTUUvLm5ldHJjCmZpCnVuc2V0IENJX05FVFJDX1VTRVJOQU1FCnVuc2V0IENJX05FVFJDX1BBU1NXT1JECnVuc2V0IENJX1NDUklQVAplY2hvICsgamF2YSAtY3AgL2RhdGF2b2x1bWUvcGRmX3RvX3dvcmQvcGRmYm94X3V0aWwtMS4wLVNOQVBTSE9ULmphciBjb20uaWZseXRlay5pbmRleGVyLlJ1bm5lciAtLWlucHV0UERGIC9kYXRhdm9sdW1lL2V4dHJhY3QvZjkyYzJhNzViYWU4NGJiMDg4MzIwNWRiM2YyZGFlNzkvcGRmL2VjNWMwYjk0M2QwYjRmNDI5MzcyMmE1ZGRjNjFlNjZkL0hTNy5wZGYgLS1vdXRwdXRXb3JkIC9kYXRhdm9sdW1lL2V4dHJhY3QvZjkyYzJhNzViYWU4NGJiMDg4MzIwNWRiM2YyZGFlNzkvcGRmVG9Xb3JkL2VjNWMwYjk0M2QwYjRmNDI5MzcyMmE1ZGRjNjFlNjZkLyAtLXNjaGVtYUlucHV0UGF0aCAvZGF0YXZvbHVtZS9leHRyYWN0L2ticWEvZjkyYzJhNzViYWU4NGJiMDg4MzIwNWRiM2YyZGFlNzkgLS1lbnRpdHlJbmRleFBhdGggL2RhdGF2b2x1bWUvZXh0cmFjdC9mOTJjMmE3NWJhZTg0YmIwODgzMjA1ZGIzZjJkYWU3OS9wZGZUb1dvcmQvZW50aXR5IC0tZmllbGRJbmRleFBhdGggL2RhdGF2b2x1bWUvZXh0cmFjdC9mOTJjMmE3NWJhZTg0YmIwODgzMjA1ZGIzZjJkYWU3OS9wZGZUb1dvcmQvZmllbGQgLS10eXBlIGx1Y2VuZSAtLW91dHB1dCAvZGF0YXZvbHVtZS9leHRyYWN0L2Y5MmMyYTc1YmFlODRiYjA4ODMyMDVkYjNmMmRhZTc5L3BkZlRvV29yZC9lYzVjMGI5NDNkMGI0ZjQyOTM3MjJhNWRkYzYxZTY2ZC9lbnRpdHlJbmZvLnR4dApqYXZhIC1jcCAvZGF0YXZvbHVtZS9wZGZfdG9fd29yZC9wZGZib3hfdXRpbC0xLjAtU05BUFNIT1QuamFyIGNvbS5pZmx5dGVrLmluZGV4ZXIuUnVubmVyIC0taW5wdXRQREYgL2RhdGF2b2x1bWUvZXh0cmFjdC9mOTJjMmE3NWJhZTg0YmIwODgzMjA1ZGIzZjJkYWU3OS9wZGYvZWM1YzBiOTQzZDBiNGY0MjkzNzIyYTVkZGM2MWU2NmQvSFM3LnBkZiAtLW91dHB1dFdvcmQgL2RhdGF2b2x1bWUvZXh0cmFjdC9mOTJjMmE3NWJhZTg0YmIwODgzMjA1ZGIzZjJkYWU3OS9wZGZUb1dvcmQvZWM1YzBiOTQzZDBiNGY0MjkzNzIyYTVkZGM2MWU2NmQvIC0tc2NoZW1hSW5wdXRQYXRoIC9kYXRhdm9sdW1lL2V4dHJhY3Qva2JxYS9mOTJjMmE3NWJhZTg0YmIwODgzMjA1ZGIzZjJkYWU3OSAtLWVudGl0eUluZGV4UGF0aCAvZGF0YXZvbHVtZS9leHRyYWN0L2Y5MmMyYTc1YmFlODRiYjA4ODMyMDVkYjNmMmRhZTc5L3BkZlRvV29yZC9lbnRpdHkgLS1maWVsZEluZGV4UGF0aCAvZGF0YXZvbHVtZS9leHRyYWN0L2Y5MmMyYTc1YmFlODRiYjA4ODMyMDVkYjNmMmRhZTc5L3BkZlRvV29yZC9maWVsZCAtLXR5cGUgbHVjZW5lIC0tb3V0cHV0IC9kYXRhdm9sdW1lL2V4dHJhY3QvZjkyYzJhNzViYWU4NGJiMDg4MzIwNWRiM2YyZGFlNzkvcGRmVG9Xb3JkL2VjNWMwYjk0M2QwYjRmNDI5MzcyMmE1ZGRjNjFlNjZkL2VudGl0eUluZm8udHh0Cg=="
},
"image":"registry.iflyresearch.com/aimind/java:v1.0.0",
"name":"pipeline_task_3eac4b36209a41e58a5f22dd403fee50_1",
"networks":[
{
"aliases":[
"default"
],
"name":"pipeline_network_3eac4b36209a41e58a5f22dd403fee50"
}
],
"on_success":true,
"volumes":[
"pipeline_default:/aimind",
"/mnt/parastor/aimind/shared/:/share",
"/mnt/parastor/aimind/pipeline-jobs/2021/07/26/3eac4b36209a41e58a5f22dd403fee50:/workspace",
"/mnt/parastor/aimind/datavolumes/carmaster:/datavolume"
],
"working_dir":"/workspace"
}
]
}
],
"volumes":[
{
"driver":"local",
"name":"pipeline_default"
}
]
}
成功返回:
代码语言:javascript复制{
"retcode": "000000",
"desc": "成功",
"data": {
"id": "8137f344-f52d-4595-bdbb-425363847b61",
}
}
可根据id获取日志。
获取job执行日志
代码语言:javascript复制GET /api/pipelines/jobLog/{jobid}/
结果:
代码语言:javascript复制{
"retcode": "000000",
"desc": "成功",
"data": {
"currentTask": null,
"logs": [
{
"id": "e76a686f68b64c0783b7721b058be137",
"jobId": "8137f344-f52d-4595-bdbb-425363847b61",
"status": "FINISHED",
"taskName": "pipeline_task_3eac4b36209a41e58a5f22dd403fee50_1",
"exitedValue": 0,
"logs": [
"proc "pipeline_task_3eac4b36209a41e58a5f22dd403fee50_1" started",
"pipeline_task_3eac4b36209a41e58a5f22dd403fee50_1: java -cp /datavolume/pdf_to_word/pdfbox_util-1.0-SNAPSHOT.jar com.iflytek.indexer.Runner --inputPDF /datavolume/extract/f92c2a75bae84bb0883205db3f2dae79/pdf/ec5c0b943d0b4f4293722a5ddc61e66d/HS7.pdf --outputWord /datavolume/extract/f92c2a75bae84bb0883205db3f2dae79/pdfToWord/ec5c0b943d0b4f4293722a5ddc61e66d/ --schemaInputPath /datavolume/extract/kbqa/f92c2a75bae84bb0883205db3f2dae79 --entityIndexPath /datavolume/extract/f92c2a75bae84bb0883205db3f2dae79/pdfToWord/entity --fieldIndexPath /datavolume/extract/f92c2a75bae84bb0883205db3f2dae79/pdfToWord/field --type lucene --output /datavolume/extract/f92c2a75bae84bb0883205db3f2dae79/pdfToWord/ec5c0b943d0b4f4293722a5ddc61e66d/entityInfo.txt",
"proc "pipeline_task_3eac4b36209a41e58a5f22dd403fee50_1" exited with status 0"
]
}
],
"exitedValue": 0,
"status": "FINISHED",
"pipelineJobSt": 1627477250599,
"pipelineJobFt": 1627477274299
}
}
周期任务
如果pipelien需要周期执行,需要配置enableTrigger
为true,同时设置按照CRON
或者’
FIX_RATE` 运行:
- FIX_RATE: 固定周期,通过
fixRateInSeconds
配置周期运行时间
示例:每360
秒运行一次:
{
// pipeline ...
"pipelineTasks" : [ ],
"fixRateInSeconds" : 360,
"scheduleType" : "FIX_RATE",
"enableTrigger" : true
}
- CRON: 按照CRON表达式周期执行,通过
cronExpression
配置.
示例:每小时开始的时候运行一次:
代码语言:javascript复制{
// pipeline ...
"pipelineTasks" : [ ],
"cronExpression" : "0 0 * * * ?",
"scheduleType" : "CRON",
"enableTrigger" : true
}