Support for deploying Jaeger into Kubernetes
Switch branches/tags
Nothing to show
Clone or download
objectiser Update to Jaeger version 1.7 (#105)
Signed-off-by: Gary Brown <gary@brownuk.com>
Latest commit 6a96024 Sep 24, 2018

README.md

Build Status

Jaeger Kubernetes Templates

How to contribute

Please see CONTRIBUTING.md

Development setup

This template uses an in-memory storage with a limited functionality for local testing and development. The image used defaults to the latest version (released)[https://github.com/jaegertracing/jaeger/releases]. Do not use this template in production environments. Note that functionality may differ from the pinned docker versions for production.

Install everything in the current namespace:

kubectl create -f https://raw.githubusercontent.com/jaegertracing/jaeger-kubernetes/master/all-in-one/jaeger-all-in-one-template.yml

Once everything is ready, kubectl get service jaeger-query tells you where to find Jaeger URL. If you are using minikube to setup your Kubernetes cluster, the command minikube service jaeger-query --url can be used instead.

Production setup

Pinned Production Version

The docker image tags are manually pinned and manually updated. You should use the current pinned version for production.

Backing storage

The Jaeger Collector and Query require a backing storage to exist before being started up. As a starting point for your own templates, we provide basic templates deploying Cassandra and Elasticsearch. None of them are ready for production and should be adapted before any real usage.

To use our Cassandra template:

kubectl create -f https://raw.githubusercontent.com/jaegertracing/jaeger-kubernetes/master/production/configmap.yml
kubectl create -f https://raw.githubusercontent.com/jaegertracing/jaeger-kubernetes/master/production/cassandra.yml

For Elasticsearch, use:

kubectl create -f https://raw.githubusercontent.com/jaegertracing/jaeger-kubernetes/master/production-elasticsearch/configmap.yml
kubectl create -f https://raw.githubusercontent.com/jaegertracing/jaeger-kubernetes/master/production-elasticsearch/elasticsearch.yml

The Cassandra template includes also a Kubernetes Job that creates the schema required by the Jaeger components. It's advisable to wait for this job to finish before deploying the Jaeger components. To check the status of the job, run:

kubectl get job jaeger-cassandra-schema-job

The job should have 1 in the SUCCESSFUL column.

Jaeger configuration

The Jaeger Collector, Query and Agent require a ConfigMap to exist on the same namespace, named jaeger-configuration. This ConfigMap is included in the storage templates, as each backing storage have their own specific configuration entries, but in your environment, you'll probably manage it differently.

If changes are required for the configuration, the edit command can be used:

kubectl edit configmap jaeger-configuration

Jaeger components

The main production template deploys the Collector and the Query Service (with UI) as separate individually scalable services, as well as the Agent as DaemonSet.

kubectl create -f https://raw.githubusercontent.com/jaegertracing/jaeger-kubernetes/master/jaeger-production-template.yml

If the backing storage is not ready by the time the Collector/Agent start, they will fail and Kubernetes will reschedule the pod. It's advisable to either wait for the backing storage to stabilize, or to ignore such failures for the first few minutes.

Once everything is ready, kubectl get service jaeger-query tells you where to find Jaeger URL, or minikube service jaeger-query --url when using minikube.

As the agent is deployed as a DaemonSet, the node's IP address can be stored as an environment variable and passed down to the application as:

env:
- name: JAEGER_AGENT_HOST
  valueFrom:
    fieldRef:
      fieldPath: status.hostIP

Deploying the agent as sidecar

The Jaeger Agent is designed to be deployed local to your service, so that it can receive traces via UDP keeping your application's load minimal. By default, the template above installs the agent as a DaemonSet, but this means that all pods running on a given node will send data to the same agent. If that's not suitable for your workload, an alternative is to deploy the agent as a sidecar. To accomplish that, just add it as a container within any struct that supports spec.containers, like a Pod, Deployment and so on. More about this be found on the blog post Deployment strategies for the Jaeger Agent.

Assuming that your application is named myapp and the image is for it is mynamespace/hello-myimage, your Deployment descriptor would be something like:

- apiVersion: apps/v1
  kind: Deployment
  metadata:
    name: myapp
  spec:
    selector:
      matchLabels:
        app: myapp
    template:
      metadata:
        labels:
          app: myapp
      spec:
        containers:
        - image: mynamespace/hello-myimage
          name: myapp
          ports:
          - containerPort: 8080
        - image: jaegertracing/jaeger-agent
          name: jaeger-agent
          ports:
          - containerPort: 5775
            protocol: UDP
          - containerPort: 6831
            protocol: UDP
          - containerPort: 6832
            protocol: UDP
          - containerPort: 5778
            protocol: TCP
          args: ["--collector.host-port=jaeger-collector.jaeger-infra.svc:14267"]

The Jaeger Agent will then be available to your application at localhost:5775/localhost:6831/localhost:6832/localhost:5778. In most cases, you don't need to specify a hostname or port to your Jaeger Tracer, as it will default to the right values already.

Configure UDP/HTTP Senders

As the Jaeger Agent is deployed with the other components, your application needs to tell the Jaeger Client where to find the agent. Refer to your client's documentation for the appropriate mechanism, but most clients allow this to be set via the environment variable JAEGER_AGENT_HOST in environment variable like so:

env:
    - name: JAEGER_SERVICE_NAME
    value: <YOUR SERVICE NAME>
    - name: JAEGER_AGENT_HOST
    value: jaeger-all-in-one-agent
    - name: JAEGER_SAMPLER_TYPE
    value: const
    - name: JAEGER_SAMPLER_PARAM
    value: "1"

The following service names are supported by HTTP sender:

Service Name Port
jaeger-collector 14268
zipkin 9411

The following service names are supported by UDP sender:

  • jaeger-agent

Persistent storage

Even though this template uses a stateful Cassandra, backing storage is set to emptyDir. It's more appropriate to create a PersistentVolumeClaim/PersistentVolume and use it instead. Note that this Cassandra deployment does not support deleting pods or scaling down, as this might require administrative tasks that are dependent on the final deployment architecture.

Service Dependencies

Jaeger production deployment needs an external process to derive dependency links between services. Project spark-dependencies provides this functionality.

This job should be periodically run before end of a day. The following command creates CronJob scheduled 5 minutes before the midnight.

kubectl run jaeger-spark-dependencies --schedule="55 23 * * *" --env="STORAGE=cassandra" --env="CASSANDRA_CONTACT_POINTS=cassandra:9042"  --restart=Never --image=jaegertracing/spark-dependencies

If you want to run the job only once and immediately then remove scheduled flag.

Deploying Docker Tags

The Jaeger project automatically creates new Docker images with tags that mirror the release number. The production manifests uses pinned versions as to not accidentally break people on new releases.

A general tip for deploying docker images (i.e. on kubernetes): it's recommended that you do not use the tag :latest in production but rather pin the latest version. See the kubernetes best practices for more details.

Helm support

A curated Chart for Kubernetes Helm that adds all components required to run Jaeger.

Uninstalling

If you need to remove the Jaeger components created by this template, run:

kubectl delete all,daemonset,configmap -l jaeger-infra

Testing

Tests are based on Arquillian Cube which require an active connection to kubernetes cluster (via kubectl). When executing tests from IDE make sure that template is copied to target/test-classes.

minikube start
./mvnw clean verify -Pcassandra,elasticsearch,all-in-one

License

Apache 2.0 License.