Simulate HTTP faults

This document describes how to simulate HTTP faults by creating HTTPChaos experiments in Chaos Mesh.

HTTPChaos introduction

HTTPChaos is a fault type provided by Chaos Mesh. By creating HTTPChaos experiments, you can simulate the fault scenarios during the HTTP request and response processing. Currently, HTTPChaos supports simulating the following fault types:

  • abort: interrupts the connection
  • delay: injects latency into the request or response
  • replace: replaces part of content in HTTP request or response messages
  • patch: adds additional content to HTTP request or response messages

HTTPChaos supports combinations of different fault types. If you have configured multiple HTTP fault types at the same time when creating HTTPChaos experiments, the order set to inject the faults when the experiments start running is abort -> delay -> replace -> patch. When the abort fault cause short circuits, the connection will be directly interrupted.

For the detailed description of HTTPChaos configuration, see Field description below.

Notes

Before injecting the faults supported by HTTPChaos, note the followings:

  • There is no control manager of Chaos Mesh running on the target Pod.
  • The rules will affect both of clients and servers in the Pod, if you want to affect only one side, please refer to the specify side section.
  • HTTPS accesses should be disabled, because injecting HTTPS connections is not supported currently.
  • For HTTPChaos injection to take effect, the client should avoid reusing TCP socket. This is because HTTPChaos does not affect the HTTP requests that are sent via TCP socket before the fault injection.
  • Use non-idempotent requests (such as most of the POST requests) with caution in production environments. If such requests are used, the target service may not return to normal status by repeating requests after the fault injection.

Create experiments using Chaos Dashboard

  1. Open Chaos Dashboard, and click NEW EXPERIMENT on the page to create a new experiment:

    create an experiment

  2. In the Choose a Target area, choose HTTP FAULT and select a specific behavior, such as RESPONSE ABORT. Then fill out specific configurations.

    create HTTP fault

  3. Submit the experiment.

    In the example above, you have configured injecting the “Response abort” fault into all requests of Port 80.

Create experiments using YAML files

Chaos Mesh also supports using YAML configuration files to create HTTPChaos experiments. In a YAML file, you can simulate either one HTTP fault type or a combination of different HTTP fault types.

Example of abort

  1. Write the experimental configuration to the http-abort-failure.yaml file as the example below:

    1. apiVersion: chaos-mesh.org/v1alpha1
    2. kind: HTTPChaos
    3. metadata:
    4. name: test-http-chaos
    5. spec:
    6. mode: all
    7. selector:
    8. labelSelectors:
    9. app: nginx
    10. target: Request
    11. port: 80
    12. method: GET
    13. path: /api
    14. abort: true
    15. duration: 5m

    Based on this configuration example, Chaos Mesh will inject the abort fault into the specified pod for 5 minutes. During the fault injection, the GET requests sent through port 80 in the /api path of the target Pod will be interrupted.

  2. After the configuration file is prepared, use kubectl to create the experiment:

    1. kubectl apply -f ./http-abort-failure.yaml

Example of fault combinations

  1. Write the experimental configuration to http-failure.yaml file as the example below:

    1. apiVersion: chaos-mesh.org/v1alpha1
    2. kind: HTTPChaos
    3. metadata:
    4. name: test-http-chaos
    5. spec:
    6. mode: all
    7. selector:
    8. labelSelectors:
    9. app: nginx
    10. target: Request
    11. port: 80
    12. method: GET
    13. path: /api/*
    14. delay: 10s
    15. replace:
    16. path: /api/v2/
    17. method: DELETE
    18. patch:
    19. headers:
    20. - ['Token', '<one token>']
    21. - ['Token', '<another token>']
    22. body:
    23. type: JSON
    24. value: '{"foo": "bar"}'
    25. duration: 5m

    Based on this configuration example, Chaos Mesh will inject the delay fault, replace fault, and patch fault consecutively.

  2. After the configuration file is prepared, use kubectl to create the experiment:

    1. kubectl apply -f ./http-failure.yaml

Field description

Description for common fields

Common fields are meaningful when the target of fault injection is Request or Response.

ParameterTypeDescriptionDefault valueRequiredExample
modestringSpecifies the mode of the experiment. The mode options include one (selecting a random pod), all (selecting all eligible pods), fixed (selecting a specified number of eligible pods), fixed-percent (selecting a specified percentage of Pods from the eligible pods), and random-max-percent (selecting the maximum percentage of Pods from the eligible pods).yesone
valuestringProvides parameters for the mode configuration depending on the value of mode.no1
targetstringSpecifies whether the target of fault injuection is Request or Response. The target-related fields should be configured at the same time.yesRequest
portint32The TCP port that the target service listens on.yes80
pathstringThe URI path of the target request. Supports Matching wildcards.Takes effect on all paths by default.no/api/
methodstringThe HTTP method of the target request method.Takes effect for all methods by default.noGET
request_headersmap[string]stringMatches request headers to the target service.Takes effect for all requests by default.noContent-Type: application/json
abortboolIndicates whether to inject the fault that interrupts the connection.falsenotrue
delaystringSpecifies the time for a latency fault.0no10s
replace.headersmap[string]stringSpecifies the key pair used to replace the request headers or response headers.noContent-Type: application/xml
replace.body[]byteSpecifies request body or response body to replace the fault (Base64 encoded).noeyJmb28iOiAiYmFyIn0K
patch.headers[][]stringSpecifies the attached key pair of the request headers or response headers with patch faults.no- [Set-Cookie, one cookie]
patch.body.typestringSpecifies the type of patch faults of the request body or response body. Currently, it only supports JSON.noJSON
patch.body.valuestringSpecifies the fault of the request body or response body with patch faults.no“{“foo”: “bar”}”
durationstringSpecifies the duration of a specific experiment.yes30s
schedulerstringSpecifies the scheduling rules for the time of a specific experiment.no5 *

The Request field is a meaningful when the target set to Request during the fault injection.

ParameterTypeDescriptionDefault valueRequiredExample
replace.pathstringSpecifies the URI path used to replace content.no/api/v2/
replace.methodstringSpecifies the replaced content of the HTTP request method.noDELETE
replace.queriesmap[string]stringSpecifies the replaced key pair of the URI query.nofoo: bar
patch.queries[][]stringSpecifies the attached key pair of the URI query with patch faults.no- [foo, bar]

The Response is a meaningful when the target set to Response during the fault injection.

ParameterTypeDescriptionDefault valueRequiredExample
codeint32Specifies the status code responded by target.Takes effect for all status codes by default.no200
response_headersmap[string]stringMatches request headers to target.Takes effect for all responses by default.noContent-Type: application/json
replace.codeint32Specifies the replaced content of the response status code.no404

Specify side

The rules will affect both of clients and servers in the Pod by default, but you can affect only one side by selecting the request headers.

This section provides some examples to specify the affected side, you can adjust the header selector in rules depend on your particular cases.

Client side

To inject faults into clients in the Pod without affecting servers, you can select the request/response by the Host header in the request.

For example, if you want to interrupt all requests to http://example.com/, you can apply the following YAML config:

  1. apiVersion: chaos-mesh.org/v1alpha1
  2. kind: HTTPChaos
  3. metadata:
  4. name: test-http-client
  5. spec:
  6. mode: all
  7. selector:
  8. labelSelectors:
  9. app: some-http-client
  10. target: Request
  11. port: 80
  12. path: '*'
  13. request_headers:
  14. Host: 'example.com'
  15. abort: true

Server side

To inject faults into servers in the Pod without affecting clients, you can also select the request/response by the Host header in the request.

For example, if you want to interrupt all requests to your server behind service nginx.nginx.svc, you can apply the following YAML config:

  1. apiVersion: chaos-mesh.org/v1alpha1
  2. kind: HTTPChaos
  3. metadata:
  4. name: test-http-server
  5. spec:
  6. mode: all
  7. selector:
  8. labelSelectors:
  9. app: nginx
  10. target: Request
  11. port: 80
  12. path: '*'
  13. request_headers:
  14. Host: 'nginx.nginx.svc'
  15. abort: true

In other cases, especially when injecting the inbound request from outside, you may select the request/response by the X-Forwarded-Host header in the request.

For example, if you want to interrupt all requests to your server behind a public gateway nginx.host.org, you can apply the following YAML config:

  1. apiVersion: chaos-mesh.org/v1alpha1
  2. kind: HTTPChaos
  3. metadata:
  4. name: test-http-server
  5. spec:
  6. mode: all
  7. selector:
  8. labelSelectors:
  9. app: nginx
  10. target: Request
  11. port: 80
  12. path: '*'
  13. request_headers:
  14. X-Forwarded-Host: 'nginx.host.org'
  15. abort: true

Local debugging

If you are not sure of the effects of certain fault injections, you can also test the corresponding features locally using rs-tproxy. Chaos Mesh also provides HTTPChaos by using rs-tproxy.