使用Selenium 4 Docker Grid镜像并行运行Gauge时随机触发TimeoutException问题求助

Hey there, let’s work through this random flakiness you’re hitting when running Gauge tests in parallel with Selenium 4 Docker Grid (versions selenium/node-chrome:4.3.0-20220706 and selenium/hub:4.3.0-20220706). The stack trace points to Netty HTTP handler failures, which often tie to resource constraints, timeouts, or version-specific bugs. Here are actionable fixes to try:

• Fix Docker Node Resource Constraints
  Chrome is resource-heavy, and the default Selenium Docker nodes don’t allocate enough shared memory or CPU/RAM for parallel runs. This causes random crashes or timeouts.

  • Add --shm-size=2g to your node run command (Chrome relies on shared memory to avoid crashes):

    docker run -d --shm-size=2g --link selenium-hub:hub selenium/node-chrome:4.3.0-20220706
    

  • Limit CPU/RAM per node to prevent resource starvation, e.g., --cpus=2 --memory=4g
  • Check the Grid UI (http://<hub-ip>:4444/ui) for node health, and inspect node logs for Chrome crash messages (docker logs <node-container-id>)

• Tune Selenium Client Timeouts
  The Netty errors often stem from unresponsive requests. Adjust your RemoteWebDriver timeout settings to account for parallel grid delays:

  ChromeOptions options = new ChromeOptions();
  // Add your desired capabilities
  
  // Configure custom HTTP client with longer timeouts
  HttpClient.Factory httpClientFactory = HttpClient.Factory.create()
      .withConnectTimeout(Duration.ofSeconds(15))
      .withReadTimeout(Duration.ofSeconds(15));
  
  HttpCommandExecutor executor = new HttpCommandExecutor(URI.create("http://hub:4444/wd/hub"), httpClientFactory);
  RemoteWebDriver driver = new RemoteWebDriver(executor, options);
  
  // Set standard driver timeouts
  driver.manage().timeouts().implicitlyWait(Duration.ofSeconds(10));
  driver.manage().timeouts().pageLoadTimeout(Duration.ofSeconds(30));
  driver.manage().timeouts().scriptTimeout(Duration.ofSeconds(10));
  

• Optimize Gauge Parallel Execution
  Don’t overwhelm your Grid with more concurrent tests than it can handle:

  • In env/default/default.properties, set max_concurrency to match your Grid’s capacity (e.g., if you have 2 nodes each supporting 4 sessions, set it to 8):

    parallel=true
    max_concurrency=8
    

  • Ensure every test scenario cleans up its WebDriver instance in an AfterScenario hook, even if the test fails. Leaking drivers will clog the Grid and cause random failures.

• Upgrade Selenium Docker Images
  The 4.3.0-20220706 version is quite old and has known stability issues with parallel runs. Upgrade to a recent stable release (e.g., 4.15.0-20231108) to pick up bug fixes for Netty and Grid session management:

  docker run -d -p 4444:4444 selenium/hub:4.15.0-20231108
  docker run -d --shm-size=2g --link selenium-hub:hub selenium/node-chrome:4.15.0-20231108
  

• Add Retry Logic for Flaky Operations
  Since failures are random, wrap high-risk operations (clicks, element finds, switches) in retry logic. You can use Gauge’s built-in retry annotation or custom code:

  • Gauge annotation example:

    @Retry(count = 2)
    fun selectAffiliateFromReport() {
        // Your step logic here
    }
    

  • Custom retry wrapper for element actions:

    fun checkOrUncheckCheckbox(element: RemoteWebElement, shouldCheck: Boolean) {
        val maxAttempts = 3
        var attempt = 0
        while (attempt < maxAttempts) {
            try {
                if (element.isSelected != shouldCheck) {
                    element.click()
                }
                break
            } catch (e: Exception) {
                attempt++
                if (attempt == maxAttempts) throw e
                Thread.sleep(1000) // Short wait before retrying
            }
        }
    }
    

• Rule Out Network Issues
  Random Netty failures can also come from unstable network connections between your test runner and Grid:

  • Test connectivity with ping <hub-ip> or traceroute <hub-ip> to check for latency or packet loss
  • If using Docker bridge network, try host network mode (if your environment allows) to eliminate port mapping overhead:

    docker run -d --network host selenium/hub:4.3.0-20220706
    docker run -d --network host --shm-size=2g selenium/node-chrome:4.3.0-20220706
    

内容的提问来源于stack exchange，提问作者mbdev