Testing guides

Injecting faults with xk6-disruptor

xk6-disruptor examples

Inject HTTP faults into Pod

Open source

Inject HTTP faults into Pod

This example shows how PodDisruptor can be used for testing the effect of faults injected in the HTTP requests served by a pod.

You will find the complete source code at the end of this document. Next sections examine the code in detail.

The example uses httpbin, a simple request/response application that offers endpoints for testing different HTTP requests.

The test requires httpbin to be deployed in a cluster in the namespace httpbin and exposed with an external IP. the IP address is expected in the environment variable SVC_IP.

You will find the Kubernetes manifests and the instructions of how to deploy it to a cluster in the test setup section at the end of this document. You can learn more about how to get the external IP address in the expose your application section.

Initialization

The initialization code imports the external dependencies required by the test. The PodDisruptor class imported from the xk6-disruptor extension provides functions for injecting faults in pods. The k6/http module provides functions for executing HTTP requests.

import { PodDisruptor } from 'k6/x/disruptor';
import http from 'k6/http';

Test Load

The test load is generated by the default function, which executes a request to the httpbin pod using the IP obtained from the environment variable SVC_IP. The test makes requests to the endpoint delay/0.1 which will return after 0.1 seconds (100ms).

export default function (data) {
  http.get(`http://${data.SVC_IP}/delay/0.1`);
}

Note
The test uses the delay endpoint which return after the requested delay. It requests a 0.1s (100ms) delay to ensure the baseline scenario (see scenarios below) has meaningful statistics for the request duration. If we were simply calling a locally deployed http server (for example nginx), the response time would exhibit a large variation between a few microseconds to a few milliseconds. Having 100ms as baseline response time has proved to offer more consistent results.

Fault injection

The disrupt function creates a PodDisruptor using a selector that matches pods in the namespace httpbin with the label app: httpbin.

The http faults are injected by calling the PodDisruptor.injectHTTPFaults method using a fault definition that introduces a delay of 50ms on each request and an error code 500 in 10% of the requests.

export function disrupt(data) {
  if (__ENV.SKIP_FAULTS == '1') {
    return;
  }
  const selector = {
    namespace: namespace,
    select: {
      labels: {
        app: 'httpbin',
      },
    },
  };
  const podDisruptor = new PodDisruptor(selector);
  // delay traffic from one random replica of the deployment
  const fault = {
    averageDelay: '50ms',
    errorCode: 500,
    errorRate: 0.1,
  };
  podDisruptor.injectHTTPFaults(fault, '30s');
}

Notice the following code snippet in the injectFaults function above:

export function disrupt(data) {
  if (__ENV.SKIP_FAULTS == '1') {
    return;
  }
  //...
}

This code makes the function return without injecting faults if the SKIP_FAULTS environment variable is passed to the execution of the test with a value of “1”. We will use this option to obtain a baseline execution without faults.

Scenarios

This test defines two scenarios to be executed. The load scenario applies the test load to the httpbin application for 30s invoking the default function. The disrupt scenario invokes the disrupt function to inject a fault in the HTTP requests of the target application.

export const options = {
  scenarios: {
    load: {
      executor: 'constant-arrival-rate',
      rate: 100,
      preAllocatedVUs: 10,
      maxVUs: 100,
      exec: 'default',
      startTime: '0s',
      duration: '30s',
    },
    disrupt: {
      executor: 'shared-iterations',
      iterations: 1,
      vus: 1,
      exec: 'disrupt',
      startTime: '0s',
    },
  },
};

Note
Notice that the disrupt scenario uses a shared-iterations executor with one iteration and one VU. This setting ensures the disrupt function is executed only once. Executing this function multiples times concurrently may have unpredictable results.

Executions

Note
The commands in this section assume the xk6-disruptor binary is available in your current directory. This location can change depending on the installation process and the platform. Refer to the installation section for details on how to install it in your environment.

Baseline execution

We will first execute the test without introducing faults to have an baseline using the following command:

xk6-disruptor run --env SKIP_FAULTS=1 --env SVC_IP=$SVC_IP --summary-mode=legacy disrupt-pod.js

xk6-disruptor run --env SKIP_FAULTS=1 --env "SVC_IP=$Env:SVC_IP" --summary-mode=legacy disrupt-pod.js

Notice the argument --env SKIP_FAULT=1, which makes the disrupt function return without injecting any fault as explained in the fault injection section. Also notice the --env SVC_IP argument, which passes the external IP used to access the httpbin application.

You should get an output similar to the one shown below (click Expand button to see all output).

         /\      Grafana   /‾‾/
    /\  /  \     |\  __   /  /
   /  \/    \    | |/ /  /   ‾‾\
  /          \   |   (  |  (‾)  |
 / __________ \  |_|\_\  \_____/

  execution: local
     script: test.js
     output: -

  scenarios: (100.00%) 2 scenarios, 101 max VUs, 10m30s max duration (incl. graceful stop):
           * disrupt: 1 iterations shared among 1 VUs (maxDuration: 10m0s, exec: disrupt, gracefulStop: 30s)
           * load: 100.00 iterations/s for 30s (maxVUs: 10-100, exec: default, gracefulStop: 30s)


running (00m30.1s), 000/014 VUs, 2998 complete and 0 interrupted iterations
disrupt ✓ [======================================] 1 VUs        00m00.0s/10m0s  1/1 shared iters
load    ✓ [======================================] 000/013 VUs  30s             100.00 iters/s

     data_received..................: 1.4 MB 46 kB/s
     data_sent......................: 267 kB 8.9 kB/s
     dropped_iterations.............: 4      0.132766/s
     http_req_blocked...............: avg=8.08µs   min=2.36µs   med=5.8µs    max=543.79µs p(90)=8.68µs   p(95)=10.5µs
     http_req_connecting............: avg=1.25µs   min=0s       med=0s       max=418.63µs p(90)=0s       p(95)=0s
     http_req_duration..............: avg=103.22ms min=101.65ms med=103.13ms max=121.7ms  p(90)=104.01ms p(95)=104.4ms
       { expected_response:true }...: avg=103.22ms min=101.65ms med=103.13ms max=121.7ms  p(90)=104.01ms p(95)=104.4ms
     http_req_failed................: 0.00%  ✓ 0         ✗ 2997
     http_req_receiving.............: avg=133.5µs  min=45.06µs  med=131.23µs max=879.43µs p(90)=193.48µs p(95)=223.04µs
     http_req_sending...............: avg=31.14µs  min=11.46µs  med=29.37µs  max=171.68µs p(90)=40.48µs  p(95)=47.53µs
     http_req_tls_handshaking.......: avg=0s       min=0s       med=0s       max=0s       p(90)=0s       p(95)=0s
     http_req_waiting...............: avg=103.05ms min=101.54ms med=102.98ms max=121.56ms p(90)=103.82ms p(95)=104.2ms
     http_reqs......................: 2997   99.474844/s
     iteration_duration.............: avg=103.34ms min=109.86µs med=103.28ms max=121.92ms p(90)=104.19ms p(95)=104.63ms
     iterations.....................: 2998   99.508035/s
     vus............................: 13     min=11      max=13
     vus_max........................: 14     min=12      max=14

Fault injection

We repeat the execution injecting the faults. Notice we have removed the --env SKIP_FAULTS=1 argument.

xk6-disruptor run --env SVC_IP=$SVC_IP --summary-mode=legacy disrupt-pod.js

xk6-disruptor run --env "SVC_IP=$Env:SVC_IP" --summary-mode=legacy disrupt-pod.js

         /\      Grafana   /‾‾/
    /\  /  \     |\  __   /  /
   /  \/    \    | |/ /  /   ‾‾\
  /          \   |   (  |  (‾)  |
 / __________ \  |_|\_\  \_____/

  execution: local
     script: disrupt-pod.js
     output: -

  scenarios: (100.00%) 2 scenarios, 101 max VUs, 10m30s max duration (incl. graceful stop):
           * disrupt: 1 iterations shared among 1 VUs (maxDuration: 10m0s, exec: disrupt, gracefulStop: 30s)
           * load: 100.00 iterations/s for 30s (maxVUs: 10-100, exec: default, gracefulStop: 30s)


running (00m31.1s), 000/018 VUs, 2995 complete and 0 interrupted iterations
disrupt ✓ [======================================] 1 VUs        00m31.1s/10m0s  1/1 shared iters
load    ✓ [======================================] 000/017 VUs  30s             100.00 iters/s

     data_received..................: 1.1 MB 34 kB/s
     data_sent......................: 267 kB 8.6 kB/s
     dropped_iterations.............: 7      0.224798/s
     http_req_blocked...............: avg=9.81µs   min=2.59µs   med=5.93µs   max=489.67µs p(90)=7.88µs   p(95)=9.5µs
     http_req_connecting............: avg=2.48µs   min=0s       med=0s       max=367.63µs p(90)=0s       p(95)=0s
     http_req_duration..............: avg=142.15ms min=50.33ms  med=153.79ms max=165.85ms p(90)=154.8ms  p(95)=155.12ms
       { expected_response:true }...: avg=151.9ms  min=101.81ms med=153.9ms  max=165.85ms p(90)=154.86ms p(95)=155.17ms
     http_req_failed................: 9.65%  ✓ 289       ✗ 2705
     http_req_receiving.............: avg=80.92µs  min=28.32µs  med=77.33µs  max=352.19µs p(90)=105.09µs p(95)=123.68µs
     http_req_sending...............: avg=30.43µs  min=11.27µs  med=29.37µs  max=287.71µs p(90)=37.42µs  p(95)=41.84µs
     http_req_tls_handshaking.......: avg=0s       min=0s       med=0s       max=0s       p(90)=0s       p(95)=0s
     http_req_waiting...............: avg=142.04ms min=50.25ms  med=153.68ms max=165.76ms p(90)=154.69ms p(95)=155.01ms
     http_reqs......................: 2994   96.149356/s
     iteration_duration.............: avg=152.64ms min=50.43ms  med=153.93ms max=31.12s   p(90)=154.97ms p(95)=155.29ms
     iterations.....................: 2995   96.18147/s
     vus............................: 1      min=1       max=18
     vus_max........................: 18     min=12      max=18

Comparison

Let’s take a closer look at the results for the requests on each scenario. We can observe that he base scenario has an average of 103ms and an error rate of 0% while the faults scenario has a median around 151.9ms and an error rate of nearly 10%, matching the definition of the faults defined in the disruptor.

Execution	Avg. Response	Failed requests
Baseline	103.22ms	0.00%
Fault injection	151.9 ms	9.65%

Note
Notice we have used the average response time reported as expected_response:true because this metric only consider successful requests while http_req_duration considers all requests, including those returning a fault.

Source Code

import { PodDisruptor } from 'k6/x/disruptor';
import http from 'k6/http';

export default function (data) {
  http.get(`http://${__ENV.SVC_IP}/delay/0.1`);
}

export function disrupt(data) {
  if (__ENV.SKIP_FAULTS == '1') {
    return;
  }

  const selector = {
    namespace: 'httpbin',
    select: {
      labels: {
        app: 'httpbin',
      },
    },
  };
  const podDisruptor = new PodDisruptor(selector);

  // delay traffic from one random replica of the deployment
  const fault = {
    averageDelay: '50ms',
    errorCode: 500,
    errorRate: 0.1,
  };
  podDisruptor.injectHTTPFaults(fault, '30s');
}

export const options = {
  scenarios: {
    load: {
      executor: 'constant-arrival-rate',
      rate: 100,
      preAllocatedVUs: 10,
      maxVUs: 100,
      exec: 'default',
      startTime: '0s',
      duration: '30s',
    },
    disrupt: {
      executor: 'shared-iterations',
      iterations: 1,
      vus: 1,
      exec: 'disrupt',
      startTime: '0s',
    },
  },
};

Test setup

The tests requires the deployment of the httpbin application. The application must also be accessible using an external IP available in the SVC_IP environment variable.

The manifests below define the resources required for deploying the application and exposing it as a LoadBalancer service.

You can deploy the application using the following commands:

# Create Namespace
kubectl apply -f namespace.yaml
namespace/httpbin created

# Deploy Pod
kubectl apply -f pod.yaml
pod/httpbin created

# Expose Pod as a Service
kubectl apply -f service.yaml
service/httpbin created

You can retrieve the resources using the following command:

kubectl -n httpbin get all
NAME          READY   STATUS    RESTARTS   AGE
pod/httpbin   1/1     Running   0          1m

NAME              TYPE           CLUSTER-IP     EXTERNAL-IP      PORT(S)        AGE
service/httpbin   LoadBalancer   10.96.169.78   172.18.255.200   80:31224/TCP   1m

You must set the environment variable SVC_IP with the external IP address and port used to access the httpbin service from the test script.

You can learn more about how to get the external IP address in the expose your application section.

Manifests

kind: 'Pod'
apiVersion: 'v1'
metadata:
  name: httpbin
  namespace: httpbin
  labels:
    app: httpbin
spec:
  containers:
    - name: httpbin
      image: kennethreitz/httpbin
      ports:
        - name: http
          containerPort: 80

Was this page helpful?

Suggest an edit in GitHub

Create a GitHub issue

Email docs@grafana.com

Help and support

Community

Inject HTTP faults into Pod

Initialization

Test Load

Fault injection

Scenarios

Executions

Baseline execution

Fault injection

Comparison

Source Code

Test setup

Manifests

Was this page helpful?

Still have questions?

Get every update

Inject HTTP faults into Pod

Initialization

Test Load

Fault injection

Scenarios

Executions

Baseline execution

Fault injection

Comparison

Source Code

Test setup

Manifests

Was this page helpful?

Related resources from Grafana Labs