adobe / athena Goto Github PK

A Performance and Functional Testing Engine for APIs

• About
  • What can Athena do?
  • How it Works
  • Getting Started
• Developer Guide
  • Coding Standards
• Distributed Load Testing
  • Creating a new Cluster
    • Standalone
    • Docker Compose
    • Accessing Kibana and Elasticsearch
  • Process Management
  • Aggregation and Reporting
    • Kibana Dashboard - Performance Results
  • Optimizing Your System for Performance
• Performance Tests
  • Hooks
    • skip
    • onInit
    • onRequest
    • onResponse
    • onDestroy
  • Mockup Responses
  • Fault Injection
  • Configuration
    • Runs
    • Patterns
    • Tests
• Functional Tests
  • Entities
    • Tests
    • Suites
• Plugins and Fixtures
  • Configuration
    • Fixtures
      • name
      • type
      • config:type
      • config:source
    • Plugins
    • Dependencies
• Roadmap
  • Sidecar for Kubernetes
  • Sidecar for Docker Images
  • Git Hooks
  • RESTful API
  • Management via UI Dashboard
• Troubleshooting
• Frequently Asked Questions
• Contributing
• Licensing

Athena is a performance and functional testing engine that aims to reduce the time and effort required to define and run tests. Its main goal is to act as a unified, but extensible tool for managing and running functional as well as performance test suites.

• Increase confidence in each release by using an integrated testing framework (performance/functional).
• Allow support for defining tests in a modular, but configurable way using YAML files.
• Aggregate test results and provide in-depth reports via Elasticsearch and predefined Kibana dashboards.
• Provide support for tests version management.
• Run tests independent of their location.
• Allow support for defining assertions in a programmatic way (functional).
• Allow support for easily extending the core functionality of the framework through plugins.
• Allow support for defining reusable fixture modules among tests.
• Allow support for creating complex performance mix patterns using functional tests. (on the roadmap)

📝Note: A thorough list of upcoming features is available in the Roadmap.

Behind the scenes, Athena uses Autocannon for performance testing and Chakram for functional API testing, however it is capable of supporting almost any other testing engine via extension.

You can start using Athena right away and run either performance and functional tests using the following command:

node athena.js run -t ./custom_tests_path --[performance|functional]

TBD

Athena uses ESLint for linting and it is following the Google JavaScript Style Guide with a couple of minor adjustments. If you are using IntelliJ as your preferred IDE, make sure to follow this guide in order to learn more about integrating the two for a better development experience.

📝 Note: This feature is currently available only for performance testing. Clustering support for functional testing is on the roadmap.

Athena supports clustering out of the box via multiple deployment strategies. Its clustering model is straightforward, requiring a Manager node and at least one Agent node.

Cluster management is fully integrated, therefore you can use the Athena CLI to create a new cluster, join an existing cluster using a secret access token (generated at initialization) and delegate suites of tests to all available workers without the need of additional management software.

Reporting and aggregation inside the cluster is also provided out of the box. The cluster manager constantly monitors the cluster state, pending and running jobs and aggregates all data inside Elasticsearch. The complete state of the cluster (available agents, previous job reports, etc) can be easily visualised in custom Kibana dashboards.

Management via a UI Dashboard provides an easy to use solution for defining test suites as well as managing previous test runs. (roadmap feature)

There are multiple ways of bootstrapping a new cluster.

Creating a standalone cluster can be achieved using the following command:

node athena.js cluster --init --addr 0.0.0.0

Once the cluster manager is bootstrapped, you will see a standard message and the necessary instructions in order to join the cluster from another node. There is no need to specify as port, as Athena will automatically assign one in the 5000-5100 range.

ℹ️  INFO:  Creating a new Athena cluster on: 0.0.0.0:5000 ...
ℹ️  INFO:  Athena cluster successfully initiated: current node (qa8NK1-QfWNq) is now a manager.
        
        👋 Hello! My name is "tall-coral" and I was assigned to manage this Athena cluster!
        
        To add more workers to this cluster, run the following command:
        node athena.js cluster --join --token DwJHjvTpmE73b-sgfkIcpZCDbiN8MMp6xdZHSb-N01Zp949M-YKcQUOS7w3-fi-u --addr 0.0.0.0:5000

📝 Note: Monitoring and reporting while bootstrapping a new standalone cluster requires extra configuration for Elasticsearch indexing.

The preffered way of bootstrapping a new cluster is via Docker Compose. Using the provided docker-compose.yaml configuration file you can easily kickstart a complete Athena cluster Manager on the fly.

docker-compose up

This will start a new Athena process running in Manager mode, an Elasticsearch cluster, a Filebeat service and Kibana for visualisation.

Once all the services are bootstrapped, you can use the generated access token to join the current cluster from other nodes.

Kibana

Once the Compose stack is up and running, you can access Kibana at:

http://localhost:5601

Elasticsearch

Also, Elasticsearch can be accessed at:

http://localhost:9200

Athena uses the PM2 process manager behind the scenes for managing the cluster Manager and Agent processes. Therefore, you can get useful information about the running processes and manage them easily via the CLI.

Each report is indexed in ElasticSearch and can be aggregated and previewed using Kibana.

Athena provides a custom Kibana Dashboard that aggregates performance job results. The aggregated results can provide insights for a single performance job executed by a single Agent inside the cluster or for the entire cluster results.

You can access the Performance Reports dashboard inside Kibana > Dashboard > Performance Reports. The following visualizations are included inside the Performance Report dashboard:

• Connections Goal
• Average RPS
• RPS in the 99th Percentile
• 2xx Responses
• non-2xx Responses
• Duration (seconds)
• Total Requests
• RPS Over Time (area chart)
• (RIOT) Requests Increase Over Time (area chart)
• RPS Percentiles (bar chart)

Isolating Performance Reports

• Use job_id : "<JOB_ID>" to aggregate the results to a specific job (provides results for the entire cluster if the job ran that way).
• Use agent_name: "<AGENT_NAME>" to aggregate the results to a specific agent.

In order to get the best performance from your nodes while running Athena, make sure to fine tune your open-file limits, network as well as your kernel settings.

Depending on your operating system, you can change your open-file limits using the following command:

ulimit -n 65536 200000

Furthermore, the following network and kernel settings are recommended inside sysctl.conf:

net.ipv4.tcp_max_syn_backlog = 40000
net.core.somaxconn = 40000
net.core.wmem_default = 8388608
net.core.rmem_default = 8388608
net.ipv4.tcp_sack = 1
net.ipv4.tcp_window_scaling = 1
net.ipv4.tcp_fin_timeout = 15
net.ipv4.tcp_keepalive_intvl = 30
net.ipv4.tcp_tw_reuse = 1
net.ipv4.tcp_moderate_rcvbuf = 1
net.core.rmem_max = 134217728
net.core.wmem_max = 134217728
net.ipv4.tcp_mem  = 134217728 134217728 134217728
net.ipv4.tcp_rmem = 4096 277750 134217728
net.ipv4.tcp_wmem = 4096 277750 134217728
net.core.netdev_max_backlog = 300000

Athena allows you to define and specify various flexible performance testing scenarios for traffic-shaping, namely:

• Stress Testing - Constant traffic at specified parameters.
• Load Testing - Steady increase of traffic until a threshold is reached.
• Spike Testing - Short bursts of high traffic.
• Soak Testing - Reliable long-duration tests.

Performance tests within Athena are composed from 3 types of modules that can be defined in individual yaml configuration files and can be used to compose complex performace tests.

1. Performance Runs - The most granular unit of work in performance tests. They allow you to define a standard or linear performance test.
2. Performance Patterns - Are composed from multiple performance runs and allow you to define complex weighted pattern mixes.
3. Performance Tests - Are composed from performance patterns and provide support for defining complex scenarios while controlling the rampUp, coolDown and spike behavior.

All performance test definitions support multiple hooks that can be used to dynamically manipulate a test's behavior. When a hook is used, the assigned function will receive the test's current context, which can be used for further decisions.

Whether to skip the current test or not. The value can be provided dynamically via a fixture function.

Runs when the test is first initialised.

Runs before the HTTP request.

Runs when the HTTP response is received.

Runs when the test case has finished.

TBD

TBD

The following section describes the configuration model defined for Athena's performance test types.

📝Notes: The config and hooks objects will cascade between performance test types and the provided values will be overriden depending on the specificity level.

The following config properties are available for performance runs.

name: string
version: string
description: string
engine: string
type: perfRun # Required perf run identifier.
config:
    url: string
    socketPath: string
    connections: number
    duration: number # in seconds
    amount: number # overrides duration
    timeout: number
    pipelining: object
    bailout: 
    method:
    title:
    body:
    headers: "object"
    maxConnectionRequests: number
    connectionRate: number
    overallRate: number
    reconnectRate: number
    requests: "[object]"
    idReplacement: string
    forever: boolean
    servername: string
    excludeErrorStats: boolean
hooks:
  onInit:
  onDestroy:
  onRequest:
  onResponse:

The following config properties are available for performance patterns.

name: string
version: string
description: string
engine: autocannon      # required
type: perfPattern       # required
pattern:
  - ref: string         # the performance run reference   
    version: string
    weight: string      # percentage (eg. 20%)
    config:             # object
                        # see perf. run config for example.
    hooks:              # object
                        # see perf. run hooks for example.

The following config properties are available for complete performance tests definitions:

name: string
description: string
engine: autocannon      # required
type: perfTest          # required
hooks:                  # object
    # ...
config:                 # object
    # ...
scenario:
  pattern:
    - ref: "prod"
      version: "1.0"
      config:
        # granular config control 
      rampUp:
        every: 10s      # or fixed
        rps: 10
        connections: 10
        threads:
        fixed: 30s      # or every
      coolDown:
        every: 10s      # or fixed
        rps: 10
        threads:
        connections: 10
        fixed: 30s      # or every
      spike:
        every: 10s      # or fixed
        rps: 10
        threads:
        connections: 10
        fixed: 30s      # or "every". if "fixed", you need to specify "after"
        after: 30s

At a granular level, functional tests must also be defined via yaml files that follow a specification, a Gherkin-like schema:

• given - Variables and prerequisites for the API call.
• when - The API call itself.
• then - The validation of expected outcomes.

Functional tests also support hooks, which are simply sections where additional preparations or cleanup can be done (e.g. obtaining an authentication token before using it within a request or setting and resetting resource states).

The order in which these sections will be executed is:

• setup
• given
• beforeWhen
• when
• beforeThen
• then
• teardown

Test Example:

type: test
name: Sample test
engine: chakram
scenario:
  given: >
    host = "https://httpbin.org/get"
    params = {
      headers: {
        "accept: application/json"
      }
    };
  when: >
    response = chakram.get(host, params)
  then: >
    expect(response).to.have.status(200)

You can organize tests together and employ hierarchical configurations via suites. These can flexibly override any or all the hook/scenario items for the tests grouped under them, as in the following example:

type: suite
name: sampleSuite
engine: chakram
hooks:  # affects hooks for all suite tests (with lower precedence)
  setup: console.log("override setup")
  beforeWhen: console.log("override beforeWhen")
  beforeThen: console.log("override beforeThen")
  teardown: console.log("override teardown")
  tests:  # affects hooks only for simpleTest (with higher precedence)
    - ref: simpleTest
      setup: console.log("override setup")
      beforeWhen: console.log("override beforeWhen")
      beforeThen: console.log("override beforeThen")
      teardown: console.log("override teardown")
scenario:  # affects scenario for all suite tests (with lower precedence)
  given: console.log("override given")
  when: console.log("override when")
  then: console.log("override then")
  tests:  # affects scenario only for simpleTest (with higher precedence)
  - ref: simpleTest
    given: console.log("override given")
    when: console.log("override when")
    then: console.log("override then")

Fixtures are helper functions that can be injected in various contexts. Plugins allow you to extend Athena's functionality. A limited set of out-of-the-box plugins will be provided by the framework, but users can define pretty much any functionality over what is already offered (for now, there is a cryptographic utility, but more are in the works).

Fixtures need to be defined and configured first via yaml files in order to be used inside tests. The following configuration options are available while defining fixtures:

(Required) The fixture name in camelCase, which will also act as the provisioned function name.

(Required) The module type (fixture).

(Required) The fixture type. Allowed values: lib, inline.

(Required) The fixture source path if config:type is set to lib. The fixture implementation if the config:type is set to inline.

The following is an example of a valid fixture definition:

name: getUUID
type: fixture
config:
  type: lib
  source: fixtures/getUUIDFixture.js

Plugins allow you to extend Athena's core functionality via setup actions and filters. Using plugins, you can intercept Athena's behaviour at specific times or even override it completely.

📝Note: A thorough list of available filters and actions is in the works.

All plugins and fixture dependencies are automatically detected and installed during runtime. The following example represents a valid fixture without the need for you to manage the package.json file.

const uuid = require("uuid/v1");

function uuidFixture() {
    return uuid();
}

module.exports = uuidFixture;

💡 Note: If a test is marked as unstable during the pre-flight check, its dependencies will not be installed.

• RESTful API.
• Web-based dashboard UI for managing suites, tests and the cluster.
• Ability to run functional tests as complex performance mix patterns.
• Support for Git hooks.
• Extended storage support for multiple adapters.
• Sidecar for Kubernetes.

Injected as a separate pod inside a node via Kubernetes hooks and Kubernetes controller, modifies iptables so all inbound and outbound traffic goes through the Atheena sidecar, for checks and traffic proxying athena uses an envoy proxy that it configures for outbound traffic proxying.

Via a Docker Compose configuration and bash scripting, Athena acts as a sidecar for individual Docker images. The approach is the same for a Kubernetes cluster.

Athena can be configured to listen to Git hooks and run tests in any directory that contains a file called .perf.athena.

Athena has a simple control plane and a powerful RESTful web server that can be used to manage suites of tests. Using the API, you can start/stop different tests as well as manage the collected metrics about a specific test or suite run.

Configuration management should be handled via a web-based custom dashboard UI that takes advantage of the exposed RESTful API.

TBD

• 🤔Question: In terms of performance, how does Athena compare with any other load testing tool?
  • 💬 Answer: Behind the scenes, Athena uses the Autocannon load testing engine which is able to deliver more load than wrk and wrk2. We've benchmarked three load testing tools (Autocannon, WRK2 and Gatling) and published our results in this short article on Medium.

Contributions are welcomed! Read the Contributing Guide for more information.

This project is licensed under the Apache V2 License. See LICENSE for more information.