fix(accessibility): scan fixes from the 2 Nov stability report

docs/TOOLS.md

@@ -340,11 +340,11 @@ searchCapability  →  describeCapability  →  invokeCapability
 
 | Tool | What it does |
 |---|---|
-| `startAccessibilityScan` | Start a web accessibility scan of a page URL and retrieve a local CSV report path. Optional settings: `wcagVersion` (WCAG 2.0–2.2 at A/AA/AAA, or RGAA 4.1), `bestPractices`, `needsReview`, `advancedRules` and `experimentalRules`; the answer states the settings the scan ran with. Mobile apps are scanned through an App Automate run instead (`setupBrowserStackAppAutomateTests` with `accessibility: true`). |
+| `startAccessibilityScan` | Start a web accessibility scan of a page URL and retrieve a local CSV report path. Optional settings: `wcagVersion` (WCAG 2.0–2.2 at A/AA/AAA, or RGAA 4.1), `bestPractices`, `needsReview`, `advancedRules` and `experimentalRules`; the answer states the settings the scan ran with. To scan the same page again (for example after a fix), pass the earlier scan's `scanId`: that starts a new run of that scan with its saved settings, which `fetchAccessibilityIssues` can compare with `compareWith: "previous"`. Mobile apps are scanned through an App Automate run instead (`setupBrowserStackAppAutomateTests` with `accessibility: true`). |
 | `accessibilityExpert` | Ask the A11y Expert (WCAG 2.0/2.1/2.2, mobile/web usability, best practices). |
 | `createAccessibilityAuthConfig` | Create an authentication configuration (form-based or basic) for accessibility scans behind a login. |
 | `getAccessibilityAuthConfig` | Retrieve an existing accessibility authentication configuration by ID. |
-| `fetchAccessibilityIssues` | Fetch accessibility issues, with pagination, from a Website Scanner run, an Automated Tests build (web) or App Accessibility build (mobile app, by the build id the SDK prints), a Workflow Analyzer report, an App Accessibility manual report (`appReportId`), or a report's dashboard link (a Website Scanner link is read by the run's build id, however old the scan; a Workflow Analyzer link by the report id it carries). Once the tests or scan have finished, waits while the report is still processing; a run still in progress is read as it is, with a note. For an app build with no scans, lists the likely causes. |
+| `fetchAccessibilityIssues` | Fetch accessibility issues, with pagination, from a Website Scanner run, an Automated Tests build (web) or App Accessibility build (mobile app, by the build id the SDK prints), a Workflow Analyzer report, an App Accessibility manual report (`appReportId`), or a report's dashboard link (a Website Scanner link is read by the run's build id, however old the scan; a Workflow Analyzer link by the report id it carries). Once the tests or scan have finished, a report that is still processing is not waited on: the answer says to call again in a minute or two; a run still in progress is read as it is, with a note. For an app build with no scans, lists the likely causes. |
 | `list_app_accessibility_builds` | List an App Accessibility project's builds newest first, with each build's id, status, app and issue counts - use this to find the latest run, a build by name, or one person's builds. |
 | `list_app_accessibility_projects` | List the account's App Accessibility projects newest first, or find one by name - use this to get the project id that lists its builds. |
 | `list_manual_app_accessibility_reports` | List manual App Accessibility reports (App Live sessions scanned with Workflow Analyzer) newest first, with each report's id, app, status and issue counts - use this to find a report QA shared or the latest manual scan of an app. |

src/lib/apiClient.ts

@@ -63,6 +63,47 @@ class ApiResponse<T = any> {
   }
 }
 
+/**
+ * How long a connection gives each of a host's addresses before trying the next. Node's own
+ * default is 250 ms: on a slow route the IPv4 attempt is dropped that early and the IPv6 one
+ * fails where there is no IPv6 route, so the call fails with "connect ETIMEDOUT <IPv4>;
+ * connect ENETUNREACH <IPv6>" within a second. The hosted server saw exactly that.
+ */
+export const CONNECT_ATTEMPT_TIMEOUT_MS = 2000;
+
+// Shared, so connections are reused as they are on Node's own global agent.
+const directAgent = new https.Agent({
+  keepAlive: true,
+  autoSelectFamilyAttemptTimeout: CONNECT_ATTEMPT_TIMEOUT_MS,
+});
+
+/**
+ * A connect failure is retried only when it came within this long. A slower one means every
+ * address already timed out, so the host is down: failing now keeps loops (scan-list walks,
+ * polling) from stalling twice as long on it.
+ */
+export const CONNECT_RETRY_WINDOW_MS = 5000;
+
+/** Failures to open the connection at all: nothing was sent, so a request can be repeated. */
+const CONNECT_FAILURE =
+  /\bconnect (ETIMEDOUT|ENETUNREACH|EHOSTUNREACH|ECONNREFUSED)\b/;
+const CONNECT_FAILURE_CODES = new Set([
+  "ETIMEDOUT",
+  "ENETUNREACH",
+  "EHOSTUNREACH",
+  "ECONNREFUSED",
+]);
+
+export function isConnectFailure(error: any): boolean {
+  if (!error || error.response) return false;
+  const causes = [error, error.cause, ...(error.errors ?? [])].filter(Boolean);
+  return causes.some(
+    (cause: any) =>
+      (cause.syscall === "connect" && CONNECT_FAILURE_CODES.has(cause.code)) ||
+      CONNECT_FAILURE.test(String(cause.message ?? "")),
+  );
+}
+
 // Utility to create HTTPS agent if needed (proxy/CA)
 function getAxiosAgent(): AxiosRequestConfig["httpsAgent"] | undefined {
   const proxyHost = config.browserstackLocalOptions.proxyHost;
@@ -88,10 +129,11 @@ function getAxiosAgent(): AxiosRequestConfig["httpsAgent"] | undefined {
     // CA only
     return new https.Agent({
       ca: fs.readFileSync(caCertPath),
+      autoSelectFamilyAttemptTimeout: CONNECT_ATTEMPT_TIMEOUT_MS,
     });
   }
   // Default agent (no proxy, no CA)
-  return undefined;
+  return directAgent;
 }
 
 class ApiClient {
@@ -148,7 +190,20 @@ class ApiClient {
     try {
       this.validateUrl(url, config);
 
-      const res = await fn(this.axiosAgent);
+      let res: AxiosResponse<T>;
+      const started = Date.now();
+      try {
+        res = await fn(this.axiosAgent);
+      } catch (error: any) {
+        // A connection that never opened sent nothing, so one more try is safe even for a
+        // POST; a second failure is reported as before. Only a fast failure is retried.
+        if (
+          !isConnectFailure(error) ||
+          Date.now() - started >= CONNECT_RETRY_WINDOW_MS
+        )
+          throw error;
+        res = await fn(this.axiosAgent);
+      }
       return new ApiResponse<T>(res);
     } catch (error: any) {
       if (error.response && !raise_error) {