Jaeger依赖分析

一、依赖任务

1.1 k8s运行任务

# https://kubernetes.io/docs/tasks/job/automated-tasks-with-cron-jobs/
# https://github.com/jaegertracing/jaeger-kubernetes
apiVersion: batch/v1beta1
kind: CronJob
metadata:
  name: jaeger-spark-dependencies
spec:
  schedule: "*/1 * * * *"
  jobTemplate:
    spec:
      template:
        spec:
          containers:
          - name: hello
            image: busybox
            args:
            - /bin/sh
            - -c
            - date; echo Hello from the Kubernetes cluster
          - name: dependencies
            image: jaegertracing/spark-dependencies
            env:
            - name: STORAGE
              value: "elasticsearch"
            - name: ES_NODES
              value: "jaeger-elasticsearch:9200"
            - name: ES_USERNAME
              value: "elastic"
            - name: ES_PASSWORD
              value: "changeme"                           
          restartPolicy: OnFailure

1.2 查看任务状态

kubectl get cj
kubectl describe cj jaeger-spark-dependencies 

1.3 依赖关系结果

[
  {
    "parent": "frontend",
    "callCount": 92,
    "source": "jaeger",
    "child": "frontend"
  },
  {
    "parent": "frontend",
    "callCount": 4,
    "source": "jaeger",
    "child": "driver"
  },
  {
    "parent": "frontend",
    "callCount": 40,
    "source": "jaeger",
    "child": "route"
  },
  {
    "parent": "frontend",
    "callCount": 4,
    "source": "jaeger",
    "child": "customer"
  },
  {
    "parent": "mysql",
    "callCount": 4,
    "source": "jaeger",
    "child": "mysql"
  },
  {
    "parent": "boot-service",
    "callCount": 1030,
    "source": "jaeger",
    "child": "boot-service"
  },
  {
    "parent": "customer",
    "callCount": 4,
    "source": "jaeger",
    "child": "mysql"
  },
  {
    "parent": "driver",
    "callCount": 54,
    "source": "jaeger",
    "child": "redis"
  }
]

二、依赖任务实现

io.jaegertracing.spark.dependencies.elastic.ElasticsearchDependenciesJob
整体思路是：
1）从ES集群查询数据，获取span列表
2）根据traceId将Span进行分组
3）计算每个分组中所有span的依赖关系

image.png

4）聚合依赖关系，并写入ES

三、依赖Http接口实现

cmd/query/app/http_handler.go:dependencies
用户需要传入时间范围，然后获取数据。

// 请求示例URL　http://localhost:16686/api/dependencies?endTs=1574408925476&lookback=604800000
func (aH *APIHandler) dependencies(w http.ResponseWriter, r *http.Request) {
    endTsMillis, err := strconv.ParseInt(r.FormValue(endTsParam), 10, 64)
    if aH.handleError(w, errors.Wrapf(err, "unable to parse %s", endTimeParam), http.StatusBadRequest) {
        return
    }
    var lookback time.Duration
    if formValue := r.FormValue(lookbackParam); len(formValue) > 0 {
        lookback, err = time.ParseDuration(formValue + "ms")
        if aH.handleError(w, errors.Wrapf(err, "unable to parse %s", lookbackParam), http.StatusBadRequest) {
            return
        }
    }
    service := r.FormValue(serviceParam)

    if lookback == 0 {
        lookback = defaultDependencyLookbackDuration
    }
    endTs := time.Unix(0, 0).Add(time.Duration(endTsMillis) * time.Millisecond)
    // 查询指定范围内的依赖关系
    dependencies, err := aH.queryService.GetDependencies(endTs, lookback)
    if aH.handleError(w, err, http.StatusInternalServerError) {
        return
    }
    // 如果用户传入服务名，根据服务名过滤依赖
    filteredDependencies := aH.filterDependenciesByService(dependencies, service)
    // 以parent和child为key，聚合数据
    structuredRes := structuredResponse{
        Data: aH.deduplicateDependencies(filteredDependencies),
    }
    aH.writeJSON(w, r, &structuredRes)
}

四、参考文档

https://github.com/openzipkin/zipkin-dependencies
https://github.com/jaegertracing/spark-dependencies

https://segmentfault.com/a/1190000021063578
https://www.jianshu.com/p/e26c5a9b5261