Plugin automatic test framework

The plugin test framework is designed to verify the function and compatibility of plugins. As there are dozens of plugins and hundreds of versions that need to be verified, it is impossible to do it manually. The test framework uses container-based tech stack and requires a set of real services with the agents installed. Then, the test mock OAP backend runs to check the segments data sent from agents.

Every plugin maintained in the main repo requires corresponding test cases as well as matching versions in the supported list doc.

Environment Requirements

1. MacOS/Linux
2. JDK 8+
3. Docker
4. Docker Compose

Case Base Image Introduction

The test framework provides JVM-container and Tomcat-container base images including JDK8 and JDK17. You can choose the best one for your test case. If both are suitable for your case, JVM-container is preferred.

JVM-container Image Introduction

JVM-container uses adoptopenjdk/openjdk8:alpine-jre as the base image. JVM-container supports JDK8 and JDK17 as well in CI, which inherits adoptopenjdk/openjdk8:alpine-jre and eclipse-temurin:17-alpine. It is supported to custom the base Java docker image by specify base_image_java. The test case project must be packaged as project-name.zip, including startup.sh and uber jar, by using mvn clean package.

Take the following test projects as examples:

• sofarpc-scenario is a single project case.
• webflux-scenario is a case including multiple projects.
• jdk17-with-gson-scenario is a single project case with JDK17.

Tomcat-container Image Introduction

Tomcat-container uses tomcat:8.5-jdk8-openjdk, tomcat:8.5-jdk17-openjdk as the base image. It is supported to custom the base Tomcat docker image by specify base_image_tomcat. The test case project must be packaged as project-name.war by using mvn package.

Take the following test project as an example

• spring-4.3.x-scenario

Test project hierarchical structure

The test case is an independent maven project, and it must be packaged as a war tar ball or zip file, depending on the chosen base image. Also, two external accessible endpoints usually two URLs) are required.

All test case codes should be in the org.apache.skywalking.apm.testcase.* package. If there are some codes expected to be instrumented, then the classes could be in the test.org.apache.skywalking.apm.testcase.* package.

JVM-container test project hierarchical structure

[plugin-scenario]
    |- [bin]
        |- startup.sh
    |- [config]
        |- expectedData.yaml
    |- [src]
        |- [main]
            |- ...
        |- [resource]
            |- log4j2.xml
    |- pom.xml
    |- configuration.yml
    |- support-version.list

[] = directory

Tomcat-container test project hierarchical structure

[plugin-scenario]
    |- [config]
        |- expectedData.yaml
    |- [src]
        |- [main]
            |- ...
        |- [resource]
            |- log4j2.xml
        |- [webapp]
            |- [WEB-INF]
                |- web.xml
    |- pom.xml
    |- configuration.yml
    |- support-version.list

[] = directory

Test case configuration files

The following files are required in every test case.

* support-version.list format requires every line for a single version (contains only the last version number of each minor version). You may use # to comment out this version.

configuration.yml

Note:, docker-compose activates only when dependencies is blank.

runningMode option description.

with_optional/with_bootstrap supports multiple selectors, separated by ;.

File Format

type:
entryService:
healthCheck:
startScript:
runningMode:
withPlugins:
environment:
  ...
depends_on:
  ...
dependencies:
  service1:
    image:
    hostname:
    expose:
      ...
    environment:
      ...
    depends_on:
      ...
    links:
      ...
    entrypoint:
      ...
    healthcheck:
      ...

• dependencies support docker compose healthcheck. But the format is a little different. We need to have - as the start of every config item, and describe it as a string line.

For example, in the official document, the health check is:

healthcheck:
  test: ["CMD", "curl", "-f", "http://localhost"]
  interval: 1m30s
  timeout: 10s
  retries: 3
  start_period: 40s

Here you should write:

healthcheck:
  - 'test: ["CMD", "curl", "-f", "http://localhost"]'
  - "interval: 1m30s"
  - "timeout: 10s"
  - "retries: 3"
  - "start_period: 40s"

In some cases, the dependency service (usually a third-party server like the SolrJ server) is required to keep the same version as the client lib version, which is defined as ${test.framework.version} in pom. You may use ${CASE_SERVER_IMAGE_VERSION} as the version number, which will be changed in the test for each version.

It does not support resource related configurations, such as volumes, ports, and ulimits. The reason for this is that in test scenarios, no mapping is required for any port to the host VM, or to mount any folder.

Take the following test cases as examples:

• dubbo-2.7.x with JVM-container
• jetty with JVM-container
• gateway with runningMode
• canal with docker-compose

expectedData.yaml

Operator for number

Operator for String

Expected Data Format Of The Segment

segmentItems:
-
  serviceName: SERVICE_NAME(string)
  segmentSize: SEGMENT_SIZE(int)
  segments:
  - segmentId: SEGMENT_ID(string)
    spans:
        ...

Expected Data Format Of The Span

Note: The order of span list should follow the order of the span finish time.

    operationName: OPERATION_NAME(string)
    parentSpanId: PARENT_SPAN_ID(int)
    spanId: SPAN_ID(int)
    startTime: START_TIME(int)
    endTime: END_TIME(int)
    isError: IS_ERROR(string: true, false)
    spanLayer: SPAN_LAYER(string: DB, RPC_FRAMEWORK, HTTP, MQ, CACHE)
    spanType: SPAN_TYPE(string: Exit, Entry, Local)
    componentId: COMPONENT_ID(int)
    tags:
    - {key: TAG_KEY(string), value: TAG_VALUE(string)}
    ...
    logs:
    - {key: LOG_KEY(string), value: LOG_VALUE(string)}
    ...
    peer: PEER(string)
    refs:
    - {
       traceId: TRACE_ID(string),
       parentTraceSegmentId: PARENT_TRACE_SEGMENT_ID(string),
       parentSpanId: PARENT_SPAN_ID(int),
       parentService: PARENT_SERVICE(string),
       parentServiceInstance: PARENT_SERVICE_INSTANCE(string),
       parentEndpoint: PARENT_ENDPOINT_NAME(string),
       networkAddress: NETWORK_ADDRESS(string),
       refType:  REF_TYPE(string: CrossProcess, CrossThread)
     }
   ...

The verify description for SegmentRef

Expected Data Format Of The Meter Items

meterItems:
-
  serviceName: SERVICE_NAME(string)
  meterSize: METER_SIZE(int)
  meters:
  - ...

Expected Data Format Of The Meter

    meterId:
        name: NAME(string)
        tags:
        - {name: TAG_NAME(string), value: TAG_VALUE(string)}
    singleValue: SINGLE_VALUE(double)
    histogramBuckets:
    - HISTOGRAM_BUCKET(double)
    ...

The verify description for MeterId

Expected Data Format Of The Log Items

logItems:
-
  serviceName: SERVICE_NAME(string)
  logSize: LOG_SIZE(int)
  logs:
  - ...

Expected Data Format Of The Log

    timestamp: TIMESTAMP_VALUE(int)
    endpoint: ENDPOINT_VALUE(int)
    traceContext:
      traceId: TRACE_ID_VALUE(string)
      traceSegmentId: TRACE_SEGMENT_ID_VALUE(string)
      spanId: SPAN_ID_VALUE(int)
    body:
      type: TYPE_VALUE(string)
      content: # Choose one of three (text, json or yaml)
        text: TEXT_VALUE(string)
        # json: JSON_VALUE(string)
        # yaml: YAML_VALUE(string)
    tags:
      data:
        - key: TAG_KEY(string)
          value: TAG_VALUE(string)
        ...
    layer: LAYER_VALUE(string)
    ...

The verify description for Log

startup.sh

This script provide a start point to JVM based service, most of them starts by a java -jar, with some variables. The following system environment variables are available in the shell.

${agent_opts} is required to add into your java -jar command, which including the parameter injected by test framework, and make agent installed. All other parameters should be added after ${agent_opts}.

The test framework will set the service name as the test case folder name by default, but in some cases, there are more than one test projects are required to run in different service codes, could set it explicitly like the following example.

Example

home="$(cd "$(dirname $0)"; pwd)"

java -jar ${agent_opts} "-Dskywalking.agent.service_name=jettyserver-scenario" ${home}/../libs/jettyserver-scenario.jar &
sleep 1

java -jar ${agent_opts} "-Dskywalking.agent.service_name=jettyclient-scenario"  ${home}/../libs/jettyclient-scenario.jar &

Only set this or use other skywalking options when it is really necessary.

Take the following test cases as examples

• undertow
• webflux

Best Practices

How To Use The Archetype To Create A Test Case Project

We provided archetypes and a script to make creating a project easier. It creates a completed project of a test case. So that we only need to focus on cases. First, we can use followed command to get usage about the script.

bash ${SKYWALKING_HOME}/test/plugin/generator.sh

Then, runs and generates a project, named by scenario_name, in ./scenarios.

Recommendations for pom

    <properties>
        <!-- Provide and use this property in the pom. -->
        <!-- This version should match the library version, -->
        <!-- in this case, http components lib version 4.3. -->
        <test.framework.version>4.3</test.framework.version>
    </properties>
    <dependencies>
        <dependency>
            <groupId>org.apache.httpcomponents</groupId>
            <artifactId>httpclient</artifactId>
            <version>${test.framework.version}</version>
        </dependency>
        ...
    </dependencies>

    <build>
        <!-- Set the package final name as same as the test case folder case. -->
        <finalName>httpclient-4.3.x-scenario</finalName>
        ....
    </build>

How To Implement Heartbeat Service

Heartbeat service is designed for checking the service available status. This service is a simple HTTP service, returning 200 means the target service is ready. Then the traffic generator will access the entry service and verify the expected data. User should consider to use this service to detect such as whether the dependent services are ready, especially when dependent services are database or cluster.

Notice, because heartbeat service could be traced fully or partially, so, segmentSize in expectedData.yaml should use ge as the operator, and don’t include the segments of heartbeat service in the expected segment data.

The example Process of Writing Tracing Expected Data

Expected data file, expectedData.yaml, include SegmentItems part.

We are using the HttpClient plugin to show how to write the expected data.

There are two key points of testing

1. Whether is HttpClient span created.
2. Whether the ContextCarrier created correctly, and propagates across processes.

+-------------+         +------------------+            +-------------------------+
|   Browser   |         |  Case Servlet    |            | ContextPropagateServlet |
|             |         |                  |            |                         |
+-----|-------+         +---------|--------+            +------------|------------+
      |                           |                                  |
      |                           |                                  |
      |       WebHttp            +-+                                 |
      +------------------------> |-|         HttpClient             +-+
      |                          |--------------------------------> |-|
      |                          |-|                                |-|
      |                          |-|                                |-|
      |                          |-| <--------------------------------|
      |                          |-|                                +-+
      | <--------------------------|                                 |
      |                          +-+                                 |
      |                           |                                  |
      |                           |                                  |
      |                           |                                  |
      |                           |                                  |
      +                           +                                  +

segmentItems

By following the flow of HttpClient case, there should be two segments created.

1. Segment represents the CaseServlet access. Let’s name it as SegmentA.
2. Segment represents the ContextPropagateServlet access. Let’s name it as SegmentB.

segmentItems:
  - serviceName: httpclient-case
    segmentSize: ge 2 # Could have more than one health check segments, because, the dependency is not standby.

Because Tomcat plugin is a default plugin of SkyWalking, so, in SegmentA, there are two spans

1. Tomcat entry span
2. HttpClient exit span

SegmentA span list should like following

    - segmentId: not null
      spans:
        - operationName: /httpclient-case/case/context-propagate
          parentSpanId: 0
          spanId: 1
          startTime: nq 0
          endTime: nq 0
          isError: false
          spanLayer: Http
          spanType: Exit
          componentId: eq 2
          tags:
            - {key: url, value: 'http://127.0.0.1:8080/httpclient-case/case/context-propagate'}
            - {key: http.method, value: GET}
            - {key: http.status_code, value: '200'}
          logs: []
          peer: 127.0.0.1:8080
        - operationName: /httpclient-case/case/httpclient
          parentSpanId: -1
          spanId: 0
          startTime: nq 0
          endTime: nq 0
          spanLayer: Http
          isError: false
          spanType: Entry
          componentId: 1
          tags:
            - {key: url, value: 'http://localhost:{SERVER_OUTPUT_PORT}/httpclient-case/case/httpclient'}
            - {key: http.method, value: GET}
            - {key: http.status_code, value: '200'}
          logs: []
          peer: null

SegmentB should only have one Tomcat entry span, but includes the Ref pointing to SegmentA.

SegmentB span list should like following

- segmentId: not null
  spans:
  -
   operationName: /httpclient-case/case/context-propagate
   parentSpanId: -1
   spanId: 0
   tags:
   - {key: url, value: 'http://127.0.0.1:8080/httpclient-case/case/context-propagate'}
   - {key: http.method, value: GET}
   - {key: http.status_code, value: '200'}
   logs: []
   startTime: nq 0
   endTime: nq 0
   spanLayer: Http
   isError: false
   spanType: Entry
   componentId: 1
   peer: null
   refs:
    - {parentEndpoint: /httpclient-case/case/httpclient, networkAddress: 'localhost:8080', refType: CrossProcess, parentSpanId: 1, parentTraceSegmentId: not null, parentServiceInstance: not null, parentService: not null, traceId: not null}

The example Process of Writing Meter Expected Data

Expected data file, expectedData.yaml, include MeterItems part.

We are using the toolkit plugin to demonstrate how to write the expected data. When write the meter plugin, the expected data file keeps the same.

There is one key point of testing

1. Build a meter and operate it.

Such as Counter:

MeterFactory.counter("test_counter").tag("ck1", "cv1").build().increment(1d);
MeterFactory.histogram("test_histogram").tag("hk1", "hv1").steps(1d, 5d, 10d).build().addValue(2d);

+-------------+         +------------------+
|   Plugin    |         |    Agent core    |
|             |         |                  |
+-----|-------+         +---------|--------+
      |                           |
      |                           |
      |    Build or operate      +-+
      +------------------------> |-|
      |                          |-]
      |                          |-|
      |                          |-|
      |                          |-|
      |                          |-|
      | <--------------------------|
      |                          +-+
      |                           |
      |                           |
      |                           |
      |                           |
      +                           +

meterItems

By following the flow of the toolkit case, there should be two meters created.

1. Meter test_counter created from MeterFactory#counter. Let’s name it as MeterA.
2. Meter test_histogram created from MeterFactory#histogram. Let’s name it as MeterB.

meterItems:
  - serviceName: toolkit-case
    meterSize: 2

They’re showing two kinds of meter, MeterA has a single value, MeterB has a histogram value.

MeterA should like following, counter and gauge use the same data format.

- meterId:
    name: test_counter
    tags:
      - {name: ck1, value: cv1}
  singleValue: gt 0

MeterB should like following.

- meterId:
    name: test_histogram
    tags:
      - {name: hk1, value: hv1}
  histogramBuckets:
    - 0.0
    - 1.0
    - 5.0
    - 10.0

Local Test and Pull Request To The Upstream

First of all, the test case project could be compiled successfully, with right project structure and be able to deploy. The developer should test the start script could run in Linux/MacOS, and entryService/health services are able to provide the response.

You could run test by using following commands

cd ${SKYWALKING_HOME}
bash ./test/plugin/run.sh -f ${scenario_name}

Notice，if codes in ./apm-sniffer have been changed, no matter because your change or git update， please recompile the skywalking-agent. Because the test framework will use the existing skywalking-agent folder, rather than recompiling it every time.

Use ${SKYWALKING_HOME}/test/plugin/run.sh -h to know more command options.

If the local test passed, then you could add it to .github/workflows/plugins-test.<n>.yaml file, which will drive the tests running on the GitHub Actions of official SkyWalking repository. Based on your plugin’s name, please add the test case into file .github/workflows/plugins-test.<n>.yaml, by alphabetical orders.

Every test case is a GitHub Actions Job. Please use the scenario directory name as the case name, mostly you’ll just need to decide which file (plugins-test.<n>.yaml) to add your test case, and simply put one line (as follows) in it, take the existed cases as examples. You can run python3 tools/select-group.py to see which file contains the least cases and add your cases into it, in order to balance the running time of each group.

If a test case required to run in JDK 17 environment, please add you test case into file plugins-jdk17-test.<n>.yaml.

jobs:
  PluginsTest:
    name: Plugin
    runs-on: ubuntu-latest
    timeout-minutes: 90
    strategy:
      fail-fast: true
      matrix:
        case:
          # ...
          - <your scenario test directory name>
          # ...