fix(web-probe): stabilize browser transport sampling (#717)

Co-authored-by: Codex <codex@noreply.local>

scripts/src/hwlab-node-impl.ts

@@ -20,6 +20,7 @@ import type { RenderedCliResult } from "./output";
 type SecretAction = "status" | "ensure" | "cleanup-owned-postgres" | "cleanup-obsolete";
 type SecretPreset = "openfga" | "master-server-admin-api-key" | "bootstrap-admin" | "code-agent-provider" | "cloud-api-db" | "owned-postgres-cleanup" | "obsolete-secret-cleanup";
 type NodeRuntimeRenderLocation = "node-host" | "local";
+type WebProbeBrowserProxyMode = "auto" | "direct";
 
 interface NodeWebProbeRunOptions {
   action: "run";
@@ -48,6 +49,7 @@ interface NodeWebProbeScriptOptions {
   url: string;
   timeoutMs: number;
   viewport: string;
+  browserProxyMode: WebProbeBrowserProxyMode;
   commandTimeoutSeconds: number;
   scriptText: string;
   scriptSource: {
@@ -70,6 +72,7 @@ interface NodeWebProbeObserveOptions {
   url: string;
   targetPath: string;
   viewport: string;
+  browserProxyMode: WebProbeBrowserProxyMode;
   sampleIntervalMs: number;
   screenshotIntervalMs: number;
   observerRefreshIntervalMs: number;
@@ -6774,6 +6777,7 @@ function parseNodeWebProbeOptions(args: string[]): NodeWebProbeOptions {
       "--url",
       "--timeout-ms",
       "--viewport",
+      "--browser-proxy-mode",
       "--script-file",
       "--command-timeout-seconds",
     ]), new Set([]));
@@ -6792,6 +6796,7 @@ function parseNodeWebProbeOptions(args: string[]): NodeWebProbeOptions {
       url: optionValue(args, "--url") ?? spec.publicWebUrl,
       timeoutMs,
       viewport: optionValue(args, "--viewport") ?? "1440x900",
+      browserProxyMode: parseWebProbeBrowserProxyMode(optionValue(args, "--browser-proxy-mode")),
       commandTimeoutSeconds,
       scriptText,
       scriptSource: {
@@ -6899,6 +6904,7 @@ function parseNodeWebProbeObserveOptions(
     "--url",
     "--target-path",
     "--viewport",
+    "--browser-proxy-mode",
     "--sample-interval-ms",
     "--screenshot-interval-ms",
     "--observer-refresh-interval-ms",
@@ -6940,6 +6946,7 @@ function parseNodeWebProbeObserveOptions(
     url: optionValue(args, "--url") ?? spec.publicWebUrl,
     targetPath: optionValue(args, "--target-path") ?? "/workbench",
     viewport: optionValue(args, "--viewport") ?? "1440x900",
+    browserProxyMode: parseWebProbeBrowserProxyMode(optionValue(args, "--browser-proxy-mode")),
     sampleIntervalMs: positiveIntegerOption(args, "--sample-interval-ms", 5000, 600000),
     screenshotIntervalMs: positiveIntegerOption(args, "--screenshot-interval-ms", 300000, 86_400_000),
     observerRefreshIntervalMs: positiveIntegerOption(args, "--observer-refresh-interval-ms", 180000, 86_400_000),
@@ -6979,6 +6986,12 @@ function parseNodeWebProbeObserveCommandType(value: string): NodeWebProbeObserve
   throw new Error(`web-probe observe command --type must be login, preflight, goto, newSession, sendPrompt, selectProvider, clickSession, screenshot, mark, or stop; got ${value}`);
 }
 
+function parseWebProbeBrowserProxyMode(value: string | undefined): WebProbeBrowserProxyMode {
+  if (value === undefined || value === "auto") return "auto";
+  if (value === "direct") return "direct";
+  throw new Error(`web-probe --browser-proxy-mode must be auto or direct, got ${value}`);
+}
+
 function nodeWebProbeAutoCommandTimeoutSeconds(input: {
   timeoutMs: number;
   waitAfterSubmitMs: number;
@@ -7421,6 +7434,7 @@ function runNodeWebProbeObserveStart(
     `UNIDESK_WEB_OBSERVE_OBSERVER_REFRESH_INTERVAL_MS=${shellQuote(String(options.observerRefreshIntervalMs))}`,
     `UNIDESK_WEB_OBSERVE_MAX_SAMPLES=${shellQuote(String(options.maxSamples))}`,
     `UNIDESK_WEB_OBSERVE_VIEWPORT=${shellQuote(options.viewport)}`,
+    `UNIDESK_WEB_OBSERVE_BROWSER_PROXY_MODE=${shellQuote(options.browserProxyMode)}`,
   ].join(" ");
   const script = [
     "set -eu",
@@ -9505,23 +9519,24 @@ function runNodeWebProbeScript(
   const webProbeProxy = nodeWebProbeHostProxyEnv(spec);
   const script = nodeWebProbeScriptRemoteShell(options, secretSpec, material.password ?? "", webProbeProxy);
   const result = runTransWorkspaceStdinScript(options.node, spec.workspace, script, options.commandTimeoutSeconds);
+  const commandTimedOut = result.timedOut || result.exitCode === 124;
   const stdoutReport = parseJsonObject(result.stdout);
   const runPaths = webProbeScriptRunPathsFromStderr(result.stderr);
   const recoveredReport = stdoutReport === null ? readNodeWebProbeScriptReport(options, spec, runPaths.reportPath) : null;
-  const recoveredArtifacts = stdoutReport === null || result.timedOut ? readNodeWebProbeScriptArtifacts(options, spec, runPaths.runDir) : null;
+  const recoveredArtifacts = stdoutReport === null || commandTimedOut ? readNodeWebProbeScriptArtifacts(options, spec, runPaths.runDir) : null;
   const parsedReport = stdoutReport ?? recoveredReport?.report ?? null;
   const report = compactWebProbeScriptResult(parsedReport);
   const passed = result.exitCode === 0 && report?.ok === true;
   const summary = nullableRecord(report?.summary);
   const stdoutBytes = Buffer.byteLength(result.stdout, "utf8");
   const outputFailureKind = parsedReport === null
-    ? result.timedOut
+    ? commandTimedOut
       ? "web-probe-command-timeout"
       : stdoutBytes > 64 * 1024
       ? "web-probe-output-too-large"
       : "web-probe-report-parse-failed"
     : null;
-  const degradedReason = result.timedOut
+  const degradedReason = commandTimedOut
     ? "web-probe-command-timeout"
     : typeof summary?.degradedReason === "string"
     ? summary.degradedReason
@@ -9529,10 +9544,10 @@ function runNodeWebProbeScript(
       ? report.failureKind
       : outputFailureKind
         ? outputFailureKind
-      : result.timedOut
+      : commandTimedOut
         ? "web-probe-command-timeout"
         : null;
-  const failureKind = result.timedOut
+  const failureKind = commandTimedOut
     ? "web-probe-command-timeout"
     : typeof summary?.failureKind === "string"
     ? summary.failureKind
@@ -9541,7 +9556,7 @@ function runNodeWebProbeScript(
       : outputFailureKind;
   const effectiveSummary = summary !== null ? {
     ...summary,
-    transportTimedOut: result.timedOut,
+    transportTimedOut: commandTimedOut,
     recoveredFrom: stdoutReport !== null ? "stdout" : recoveredReport?.source ?? null,
   } : (outputFailureKind === null ? null : {
     ok: false,
@@ -9564,6 +9579,7 @@ function runNodeWebProbeScript(
     screenshots: recoveredArtifacts?.screenshots ?? [],
     artifacts: recoveredArtifacts?.artifacts ?? null,
     stdoutBytes,
+    exitCode: result.exitCode,
     stderrTail: result.stderr.trim().slice(-2000),
     valuesRedacted: true,
   });
@@ -9631,6 +9647,7 @@ function nodeWebProbeScriptRemoteShell(options: NodeWebProbeScriptOptions, secre
       "UNIDESK_WEB_PROBE_USER_SCRIPT=\"$user_script\"",
       `UNIDESK_WEB_PROBE_TIMEOUT_MS=${shellQuote(String(options.timeoutMs))}`,
       `UNIDESK_WEB_PROBE_VIEWPORT=${shellQuote(options.viewport)}`,
+      `UNIDESK_WEB_PROBE_BROWSER_PROXY_MODE=${shellQuote(options.browserProxyMode)}`,
       "node \"$runner\"",
     ].join(" "),
   ].join("\n");