天天觀焦點：【云原生 ? Prometheus】Prometheus 注冊中心Eureka服務發(fā)現(xiàn)原理

2023-04-21 16:23:46 來源：騰訊云

Prometheus 注冊中心Eureka服務發(fā)現(xiàn)原理

概述

Eureka服務發(fā)現(xiàn)協(xié)議允許使用Eureka Rest API檢索出Prometheus需要監(jiān)控的targets，Prometheus會定時周期性的從Eureka調用Eureka Rest API，并將每個應用實例創(chuàng)建出一個target。

(相關資料圖)

Eureka服務發(fā)現(xiàn)協(xié)議支持對如下元標簽進行relabeling：

__meta_eureka_app_name: the name of the app__meta_eureka_app_instance_id: the ID of the app instance__meta_eureka_app_instance_hostname: the hostname of the instance__meta_eureka_app_instance_homepage_url: the homepage url of the app instance__meta_eureka_app_instance_statuspage_url: the status page url of the app instance__meta_eureka_app_instance_healthcheck_url: the health check url of the app instance__meta_eureka_app_instance_ip_addr: the IP address of the app instance__meta_eureka_app_instance_vip_address: the VIP address of the app instance__meta_eureka_app_instance_secure_vip_address: the secure VIP address of the app instance__meta_eureka_app_instance_status: the status of the app instance__meta_eureka_app_instance_port: the port of the app instance__meta_eureka_app_instance_port_enabled: the port enabled of the app instance__meta_eureka_app_instance_secure_port: the secure port address of the app instance__meta_eureka_app_instance_secure_port_enabled: the secure port of the app instance__meta_eureka_app_instance_country_id: the country ID of the app instance__meta_eureka_app_instance_metadata_: app instance metadata__meta_eureka_app_instance_datacenterinfo_name: the datacenter name of the app instance__meta_eureka_app_instance_datacenterinfo_: the datacenter metadata

eureka_sd_configs常見配置如下：

- job_name: "eureka"  eureka_sd_configs:    - server: http://localhost:8761/eureka #eureka server地址      refresh_interval: 1m #刷新間隔，默認30s

eureka_sd_configs官網(wǎng)支持主要配置如下：

server: basic_auth:  [ username:  ]  [ password:  ]  [ password_file:  ]# Configures the scrape request"s TLS settings.tls_config:  [  ]# Optional proxy URL.[ proxy_url:  ]# Configure whether HTTP requests follow HTTP 3xx redirects.[ follow_redirects:  | default = true ]# Refresh interval to re-read the app instance list.[ refresh_interval:  | default = 30s ]

Eureka協(xié)議實現(xiàn)

基于Eureka服務發(fā)現(xiàn)協(xié)議核心邏輯都封裝在discovery/eureka.go的func (d *Discovery) refresh(ctx context.Context) ([]*targetgroup.Group, error)方法中：

func (d *Discovery) refresh(ctx context.Context) ([]*targetgroup.Group, error) { // 通過Eureka REST API接口從eureka拉取元數(shù)據(jù)：http://ip:port/eureka/apps apps, err := fetchApps(ctx, d.server, d.client) if err != nil {  return nil, err } tg := &targetgroup.Group{  Source: "eureka", } for _, app := range apps.Applications {//遍歷app        // targetsForApp()方法將app下每個instance部分轉成target  targets := targetsForApp(&app)        //解析的采集點合入一起  tg.Targets = append(tg.Targets, targets...) } return []*targetgroup.Group{tg}, nil}

refresh方法主要有兩個流程：

1、fetchApps()：從eureka-server的/eureka/apps接口拉取注冊服務信息；

2、targetsForApp()：遍歷app下instance，將每個instance解析出一個target，并添加一堆元標簽數(shù)據(jù)。

如下示例從eureka-server的/eureka/apps接口拉取的注冊服務信息：

    1    UP_1_            SERVICE-PROVIDER-01                    localhost:service-provider-01:8001            192.168.3.121            SERVICE-PROVIDER-01            192.168.3.121            UP            UNKNOWN            8001            443            1                            MyOwn                                        30                90                1629385562130                1629385682050                0                1629385562132                                        8001                true                8080                        http://192.168.3.121:8001/            http://192.168.3.121:8001/actuator/info            http://192.168.3.121:8001/actuator/health            service-provider-01            service-provider-01            false            1629385562132            1629385562039            ADDED

instance信息會被解析成采集點target：

func targetsForApp(app *Application) []model.LabelSet { targets := make([]model.LabelSet, 0, len(app.Instances)) // Gather info about the app"s "instances". Each instance is considered a task. for _, t := range app.Instances {  var targetAddress string        // __address__取值方式：instance.hostname和port，沒有port則默認port=80  if t.Port != nil {   targetAddress = net.JoinHostPort(t.HostName, strconv.Itoa(t.Port.Port))  } else {   targetAddress = net.JoinHostPort(t.HostName, "80")  }  target := model.LabelSet{   model.AddressLabel:  lv(targetAddress),   model.InstanceLabel: lv(t.InstanceID),   appNameLabel:                     lv(app.Name),   appInstanceHostNameLabel:         lv(t.HostName),   appInstanceHomePageURLLabel:      lv(t.HomePageURL),   appInstanceStatusPageURLLabel:    lv(t.StatusPageURL),   appInstanceHealthCheckURLLabel:   lv(t.HealthCheckURL),   appInstanceIPAddrLabel:           lv(t.IPAddr),   appInstanceVipAddressLabel:       lv(t.VipAddress),   appInstanceSecureVipAddressLabel: lv(t.SecureVipAddress),   appInstanceStatusLabel:           lv(t.Status),   appInstanceCountryIDLabel:        lv(strconv.Itoa(t.CountryID)),   appInstanceIDLabel:               lv(t.InstanceID),  }  if t.Port != nil {   target[appInstancePortLabel] = lv(strconv.Itoa(t.Port.Port))   target[appInstancePortEnabledLabel] = lv(strconv.FormatBool(t.Port.Enabled))  }  if t.SecurePort != nil {   target[appInstanceSecurePortLabel] = lv(strconv.Itoa(t.SecurePort.Port))   target[appInstanceSecurePortEnabledLabel] = lv(strconv.FormatBool(t.SecurePort.Enabled))  }  if t.DataCenterInfo != nil {   target[appInstanceDataCenterInfoNameLabel] = lv(t.DataCenterInfo.Name)   if t.DataCenterInfo.Metadata != nil {    for _, m := range t.DataCenterInfo.Metadata.Items {     ln := strutil.SanitizeLabelName(m.XMLName.Local)     target[model.LabelName(appInstanceDataCenterInfoMetadataPrefix+ln)] = lv(m.Content)    }   }  }  if t.Metadata != nil {   for _, m := range t.Metadata.Items {                // prometheus label只支持[^a-zA-Z0-9_]字符，其它非法字符都會被替換成下劃線_    ln := strutil.SanitizeLabelName(m.XMLName.Local)    target[model.LabelName(appInstanceMetadataPrefix+ln)] = lv(m.Content)   }  }  targets = append(targets, target) } return targets}

解析比較簡單，就不再分析，解析后的標簽數(shù)據(jù)如下圖：

標簽中有兩個特別說明下：

1、__address__：這個取值instance.hostname和port(默認80)，所以要注意注冊到eureka上的hostname準確性，不然可能無法抓取；

2、metadata-map數(shù)據(jù)會被轉成__meta_eureka_app_instance_metadata_格式標簽，prometheus進行relabeling一般操作metadata-map，可以自定義metric_path、抓取端口等；

3、prometheus的label只支持[a-zA-Z0-9_]，其它非法字符都會被轉換成下劃線，具體參加：strutil.SanitizeLabelName(m.XMLName.Local)；但是eureka的metadata-map標簽含有下劃線時，注冊到eureka-server上變成雙下劃線，如下配置:

eureka:  instance:    metadata-map:      scrape_enable: true      scrape.port: 8080

通過/eureka/apps獲取如下：

總結

基于Eureka服務發(fā)現(xiàn)原理如下圖：

基于eureka_sd_configs服務發(fā)現(xiàn)協(xié)議配置創(chuàng)建Discoverer，并通過協(xié)程運行Discoverer.Run方法，Eureka服務發(fā)現(xiàn)核心邏輯封裝discovery/eureka.go的func (d *Discovery) refresh(ctx context.Context) ([]*targetgroup.Group, error)方法中。

refresh方法中主要調用兩個方法：

1、fetchApps：定時周期從Eureka Server的/eureka/apps接口拉取注冊上來的服務元數(shù)據(jù)信息；

2、targetsForApp：解析上步驟拉取的元數(shù)據(jù)信息，遍歷app下的instance，將每個instance解析成target，并將其它元數(shù)據(jù)信息轉換成target元標簽可以用于relabel_configs操作

標簽：