This sample explores the observability capabilities of Dapr. Observability includes metric collection, tracing, logging and health checks. In this sample you'll be enabling distributed tracing on an application without changing any application code or creating a dependency on any specific tracing system. Since Dapr uses OpenCensus, a variety of observability tools can be used to view and capture the traces. In this sampple you'll be using Zipkin.
In this sample you will:
- Deploy Zipkin and configure it as a tracing provider for Dapr in Kubernetes.
- Instrument an application for tracing and then deploy it.
- Troubleshoot a performance issue.
This sample builds on the distributed calculator sample and requires Dapr to be installed on a Kubernetes cluster along with state store. It is suggested to go through the distributed calculator sample before this one, which sets you up ready for this sample. If you have not done this then:
-
Clone this repo using
git clone https://github.com/dapr/samples.gitand go to the directory named /8.obervability -
Configure Redis as a state store for Dapr.
Review the Dapr configuration file ./deploy/appconfig.yaml below:
apiVersion: dapr.io/v1alpha1
kind: Configuration
metadata:
name: appconfig
spec:
tracing:
samplingRate: "1"samplingRate is used to enable or disable the tracing. To disable the sampling rate ,
set samplingRate : "0" in the configuration. The valid range of samplingRate is between 0 and 1 inclusive. The sampling rate determines whether a trace span should be sampled or not based on value. samplingRate : "1" will always sample the traces. By default, the sampling rate is 1 in 10,000
This configuration file enables Dapr tracing. Deploy the configuration by running:
kubectl apply -f ./deploy/appconfig.yamlYou can see that now a new Dapr configuration which enables tracing has been added. Run the command:
dapr configurations --kubernetesYou should see output that looks like this:
NAME TRACING-ENABLED MTLS-ENABLED MTLS-WORKLOAD-TTL MTLS-CLOCK-SKEW
appconfig true false
default false true 24h 15m You can see that appconfig has TRACING-ENABLED set to true.
In this sample Zipkin is used for tracing. Examine ./deploy/zipkin.yaml and see how it includes three sections:
- A Deployment for Zipkin using the openzipkin/zipkin docker image.
- A Service which will expose Zipkin internally as a ClusterIP in Kubernetes.
- A Component that defines Zipkin as the tracing provider for Dapr.
Deploy Zipkin to your cluster by running:
kubectl apply -f ./deploy/zipkin.yamlNow that Zipkin is deployed, you can access the Zipkin UI by creating a tunnel to the internal Zipkin service you just created by running:
kubectl port-forward svc/zipkin 9411:9411On your browser go to http://localhost:9411. You should be able to see the Zipkin dashboard.
To instrument a service for tracing with Dapr, no code changes are required, Dapr handles all of the tracing using the Dapr side-car. All that is needed is to add the Dapr annotation for the configuration you deployed earlier (which enables tracing) in the application deployment yaml along with the other Dapr annotations. The configuration annotation looks like this:
...
annotations:
...
dapr.io/config: "appconfig"
...For this sample, a configuration has already been enabled for every service in the distributed calculator app. You can find the annotation in each one of the calculator yaml files. For example review the yaml file for the calculator front end service here.
Note you did not introduce any dependency on Zipkin into the calculator app code or deployment yaml files. The Zipkin Dapr component is configured to read tracing events and write these to an exporter.
Now deploy the distributed calculator application to your cluster following the instructions found in sample 3.distributed-calculator. Then browse to the calculator UI.
Note: If the distributed calculator is already running on your cluster you will need to restart it for the tracing to take effect. You can do so by running:
kubectl rollout restart deployment addapp calculator-front-end divideapp multiplyapp subtractapp
To show how observability can help discover and troubleshoot issues on a distributed application, you'll update one of the services in the calculator app. This updated version simulates a performance degradation in the multiply operation of the calculator that you can then investigate using the traces emitted by the Dapr sidecar. Run the following to apply a new version of the python-multiplier service:
kubectl apply -f ./deploy/python-multiplier.yamlNow go to the calculator UI and perform several calculations. Make sure to use all operands. For example, do the following:
9 + 3 * 2 / 4 - 1 =
Now go to the Zipkin dashboard by running:
kubectl port-forward svc/zipkin 9411:9411And browsing to http://localhost:9411. Click the search button to view tracing coming from the application:
Dapr adds a HTTP/gRPC middleware to the Dapr sidecar. The middleware intercepts all Dapr and application traffic and automatically injects correlation IDs to trace events. You can see a lot of transactions are being captured including the regular health checks done by Kubernetes:
Now look for any performance issues by filtering on any requests that have taken longer than 250 ms using the minDuration criteria:
You can quickly see that the multiply method invocation is unusually slow (takes over 1 second). Since the problem may be either at the calculator-frontend service or the python-multiplier service you can dig further by clicking on the entry:
Now you can see which specific call was delayed via the data field (here it's the 12 * 2 operation) and confirm that it is the multiplier service which you updated that is causing the slowdown (You can find the code for the slow multiplier under the python directory).
- To remove the distributed calculator application from your cluster run:
kubectl delete -f ..\3.distributed-calculator\deploy- To remove the Zipkin installation and tracing configuration run:
kubectl delete -f deploy\appconfig.yaml -f deploy\zipkin.yaml- Learn more about observability.
- Learn more on how Dapr does distributed tracing.
- As an alternative to Zipkin see how to configure tracing with Application Insights.