fix(web-probe): stabilize browser transport sampling (#717)
Co-authored-by: Codex <codex@noreply.local>
This commit is contained in:
Lyon
@@ -22,6 +22,7 @@ const screenshotIntervalMs = positiveInteger(process.env.UNIDESK_WEB_OBSERVE_SCR
|
||||
const maxSamples = positiveInteger(process.env.UNIDESK_WEB_OBSERVE_MAX_SAMPLES, 0);
|
||||
const observerRefreshIntervalMs = positiveInteger(process.env.UNIDESK_WEB_OBSERVE_OBSERVER_REFRESH_INTERVAL_MS, 180000);
|
||||
const viewport = parseViewport(process.env.UNIDESK_WEB_OBSERVE_VIEWPORT || "1440x900");
|
||||
const browserProxyMode = parseBrowserProxyMode(process.env.UNIDESK_WEB_OBSERVE_BROWSER_PROXY_MODE || "auto");
|
||||
const playwrightProxy = proxyConfigFromEnv(baseUrl);
|
||||
const chromiumLaunchOptions = chromiumLaunchOptionsForProxy(playwrightProxy);
|
||||
const pageId = "control-" + randomBytes(4).toString("hex");
|
||||
@@ -512,9 +513,9 @@ async function authenticate(browserContext, authPage) {
|
||||
|
||||
async function pageAuthLogin(authPage, loginUrl) {
|
||||
if (!authPage) throw new Error("auth page is not ready");
|
||||
await authPage.goto(loginUrl, { waitUntil: "domcontentloaded", timeout: 12000 });
|
||||
await authPage.goto(new URL("/assets/favicon.svg", baseUrl).toString(), { waitUntil: "domcontentloaded", timeout: 12000 });
|
||||
return authPage.evaluate(async (input) => {
|
||||
const response = await fetch("/auth/login", {
|
||||
const response = await fetch(input.loginUrl, {
|
||||
method: "POST",
|
||||
headers: { accept: "application/json", "content-type": "application/json" },
|
||||
body: JSON.stringify({ username: input.username, password: input.password }),
|
||||
@@ -526,7 +527,7 @@ async function pageAuthLogin(authPage, loginUrl) {
|
||||
status: response.status,
|
||||
statusText: response.statusText || "",
|
||||
};
|
||||
}, { username, password });
|
||||
}, { username, password, loginUrl });
|
||||
}
|
||||
|
||||
function publicAuth(value) {
|
||||
@@ -560,7 +561,7 @@ function isRetryableAuthStatus(status) {
|
||||
|
||||
function isRetryableAuthError(error) {
|
||||
const message = error && error.message ? String(error.message) : String(error || "");
|
||||
return /AbortError|EAI_AGAIN|ETIMEDOUT|ECONNRESET|ECONNREFUSED|ECONNABORTED|socket hang up|ERR_NETWORK_CHANGED|fetch failed|network|timeout|aborted/iu.test(message);
|
||||
return /AbortError|EAI_AGAIN|ETIMEDOUT|ECONNRESET|ECONNREFUSED|ECONNABORTED|socket hang up|ERR_NETWORK_CHANGED|fetch failed|failed to fetch|network|timeout|aborted/iu.test(message);
|
||||
}
|
||||
|
||||
function authFailureMessage(failure) {
|
||||
@@ -573,6 +574,7 @@ function authFailureMessage(failure) {
|
||||
}
|
||||
|
||||
function proxyConfigFromEnv(targetBaseUrl) {
|
||||
if (browserProxyMode === "direct") return null;
|
||||
let target;
|
||||
try {
|
||||
target = new URL(targetBaseUrl);
|
||||
@@ -628,6 +630,7 @@ function publicNetwork(proxy) {
|
||||
},
|
||||
browser: {
|
||||
proxyMode: proxy === null ? "direct-no-proxy-server" : "explicit-playwright-proxy",
|
||||
requestedProxyMode: browserProxyMode,
|
||||
proxyEnvCleared: true,
|
||||
valuesRedacted: true,
|
||||
},
|
||||
@@ -635,6 +638,11 @@ function publicNetwork(proxy) {
|
||||
};
|
||||
}
|
||||
|
||||
function parseBrowserProxyMode(raw) {
|
||||
if (raw === "auto" || raw === "direct") return raw;
|
||||
return "auto";
|
||||
}
|
||||
|
||||
function publicProxyServer(raw) {
|
||||
try {
|
||||
const parsed = new URL(String(raw || ""));
|
||||
@@ -2383,10 +2391,10 @@ function buildFindings(samples, control, network, errors, sampleMetrics, promptN
|
||||
const sessionRailTitleSummary = sampleMetrics?.sessionRailTitles?.summary || {};
|
||||
if (Number(sessionRailTitleSummary.majorityFallbackSampleCount ?? 0) > 0) findings.push({ id: "session-rail-title-fallback-majority", severity: "red", summary: "more than half of visible session list rows used fallback Session ses_... titles", count: sessionRailTitleSummary.majorityFallbackSampleCount, maxFallbackRatio: sessionRailTitleSummary.maxFallbackRatio, maxFallbackTitleCount: sessionRailTitleSummary.maxFallbackTitleCount, samples: sampleMetrics.sessionRailTitles.samples.slice(0, 20), examples: sampleMetrics.sessionRailTitles.examples.slice(0, 20) });
|
||||
if ((runtimeAlerts?.summary?.httpErrorCount ?? 0) > 0) findings.push({ id: "runtime-http-errors", severity: "amber", summary: "natural page requests returned HTTP error status during observation", count: runtimeAlerts.summary.httpErrorCount, groups: runtimeAlerts.networkHttpErrorsByPath.slice(0, 12) });
|
||||
if ((runtimeAlerts?.summary?.requestFailedCount ?? 0) > 0) findings.push({ id: "runtime-requestfailed", severity: "amber", summary: "browser requestfailed events were captured during observation", count: runtimeAlerts.summary.requestFailedCount, groups: runtimeAlerts.networkRequestFailedByPath.slice(0, 12) });
|
||||
if ((runtimeAlerts?.summary?.significantRequestFailedCount ?? runtimeAlerts?.summary?.requestFailedCount ?? 0) > 0) findings.push({ id: "runtime-requestfailed", severity: "amber", summary: "browser requestfailed events were captured during observation", count: runtimeAlerts.summary.significantRequestFailedCount ?? runtimeAlerts.summary.requestFailedCount, groups: (runtimeAlerts.networkSignificantRequestFailedByPath ?? runtimeAlerts.networkRequestFailedByPath).slice(0, 12) });
|
||||
if ((runtimeAlerts?.summary?.domDiagnosticSampleCount ?? 0) > 0) findings.push({ id: "runtime-dom-diagnostics", severity: "amber", summary: "diagnostic/error/warning-like text was visible in sampled DOM", count: runtimeAlerts.summary.domDiagnosticSampleCount, groupCount: runtimeAlerts.summary.domDiagnosticGroupCount ?? 0, groups: runtimeAlerts.domDiagnosticsByText.slice(0, 12), samples: runtimeAlerts.domDiagnostics.slice(0, 12) });
|
||||
if ((runtimeAlerts?.summary?.executionErrorCount ?? 0) > 0) findings.push({ id: "runtime-execution-errors", severity: "red", summary: "Workbench rendered execution failure/error rows during observation", count: runtimeAlerts.summary.executionErrorCount, groups: runtimeAlerts.runtimeExecutionErrorsByCode.slice(0, 12) });
|
||||
if ((runtimeAlerts?.summary?.consoleAlertCount ?? 0) > 0) findings.push({ id: "runtime-console-alerts", severity: "amber", summary: "browser console warning/error entries were captured during observation", count: runtimeAlerts.summary.consoleAlertCount, groups: runtimeAlerts.consoleAlertsByPath.slice(0, 12) });
|
||||
if ((runtimeAlerts?.summary?.significantConsoleAlertCount ?? runtimeAlerts?.summary?.consoleAlertCount ?? 0) > 0) findings.push({ id: "runtime-console-alerts", severity: "amber", summary: "browser console warning/error entries were captured during observation", count: runtimeAlerts.summary.significantConsoleAlertCount ?? runtimeAlerts.summary.consoleAlertCount, groups: (runtimeAlerts.significantConsoleAlertsByPath ?? runtimeAlerts.consoleAlertsByPath).slice(0, 12) });
|
||||
const crossPageDiffs = detectCrossPageProjectionDiffs(samples);
|
||||
const crossPageProjectionDiffs = crossPageDiffs.filter((item) => item.diffKind !== "trace-visibility");
|
||||
const crossPageTraceVisibilityDiffs = crossPageDiffs.filter((item) => item.diffKind === "trace-visibility");
|
||||
@@ -2946,6 +2954,7 @@ function buildRuntimeAlerts(samples, control, network, consoleEvents, errors) {
|
||||
const requestFailed = naturalNetwork
|
||||
.filter((item) => item?.type === "requestfailed")
|
||||
.map((item) => networkAlertEvent(item, promptTimes));
|
||||
const significantRequestFailed = requestFailed.filter((item) => !isBenignLongLivedStreamClosureAlert(item));
|
||||
const domDiagnostics = [];
|
||||
const executionErrors = [];
|
||||
const baselineExecutionErrors = [];
|
||||
@@ -3049,6 +3058,7 @@ function buildRuntimeAlerts(samples, control, network, consoleEvents, errors) {
|
||||
const consoleAlerts = consoleEvents
|
||||
.filter((item) => /error|warning|warn|assert/iu.test(String(item?.type || "")) || isDiagnosticText(item?.text))
|
||||
.map((item) => consoleAlertEvent(item, promptTimes));
|
||||
const significantConsoleAlerts = consoleAlerts.filter((item) => !isBenignLongLivedStreamClosureAlert(item));
|
||||
const pageErrors = errors.map((item) => ({
|
||||
ts: item.ts ?? null,
|
||||
promptIndex: promptIndexForTs(promptTimes, item.ts),
|
||||
@@ -3061,20 +3071,26 @@ function buildRuntimeAlerts(samples, control, network, consoleEvents, errors) {
|
||||
summary: {
|
||||
httpErrorCount: httpErrors.length,
|
||||
requestFailedCount: requestFailed.length,
|
||||
significantRequestFailedCount: significantRequestFailed.length,
|
||||
benignLongLivedStreamClosureCount: requestFailed.length - significantRequestFailed.length,
|
||||
domDiagnosticSampleCount: domDiagnostics.length,
|
||||
domDiagnosticGroupCount: groupDomDiagnostics(domDiagnostics).length,
|
||||
executionErrorCount: executionErrors.length,
|
||||
baselineExecutionErrorCount: baselineExecutionErrors.length,
|
||||
consoleAlertCount: consoleAlerts.length,
|
||||
significantConsoleAlertCount: significantConsoleAlerts.length,
|
||||
pageErrorCount: pageErrors.length,
|
||||
networkErrorGroupCount: groupNetworkAlerts(httpErrors).length,
|
||||
requestFailedGroupCount: groupNetworkAlerts(requestFailed).length,
|
||||
significantRequestFailedGroupCount: groupNetworkAlerts(significantRequestFailed).length,
|
||||
executionErrorGroupCount: groupExecutionErrors(executionErrors).length,
|
||||
baselineExecutionErrorGroupCount: groupExecutionErrors(baselineExecutionErrors).length,
|
||||
consoleAlertGroupCount: groupConsoleAlerts(consoleAlerts).length
|
||||
consoleAlertGroupCount: groupConsoleAlerts(consoleAlerts).length,
|
||||
significantConsoleAlertGroupCount: groupConsoleAlerts(significantConsoleAlerts).length
|
||||
},
|
||||
networkHttpErrorsByPath: groupNetworkAlerts(httpErrors),
|
||||
networkRequestFailedByPath: groupNetworkAlerts(requestFailed),
|
||||
networkSignificantRequestFailedByPath: groupNetworkAlerts(significantRequestFailed),
|
||||
domDiagnostics: domDiagnostics.slice(-80),
|
||||
domDiagnosticsByText: groupDomDiagnostics(domDiagnostics),
|
||||
domDiagnosticsByFingerprint: groupDomDiagnostics(domDiagnostics).slice(0, 80),
|
||||
@@ -3084,6 +3100,8 @@ function buildRuntimeAlerts(samples, control, network, consoleEvents, errors) {
|
||||
baselineRuntimeExecutionErrorsByCode: groupExecutionErrors(baselineExecutionErrors),
|
||||
consoleAlerts: consoleAlerts.slice(0, 80),
|
||||
consoleAlertsByPath: groupConsoleAlerts(consoleAlerts),
|
||||
significantConsoleAlerts: significantConsoleAlerts.slice(0, 80),
|
||||
significantConsoleAlertsByPath: groupConsoleAlerts(significantConsoleAlerts),
|
||||
pageErrors: pageErrors.slice(0, 40)
|
||||
};
|
||||
}
|
||||
@@ -3313,6 +3331,13 @@ function groupConsoleAlerts(events) {
|
||||
return Array.from(groups.values()).sort((a, b) => b.count - a.count || String(a.urlPath).localeCompare(String(b.urlPath)));
|
||||
}
|
||||
|
||||
function isBenignLongLivedStreamClosureAlert(event) {
|
||||
if (event?.urlPath !== "/v1/workbench/events") return false;
|
||||
if (event.status !== null && event.status !== undefined && Number(event.status) > 0) return false;
|
||||
const text = String(event.failureKind || event.errorPreview || event.preview || "");
|
||||
return /ERR_NETWORK_CHANGED|ERR_ABORTED|net::ERR_NETWORK_CHANGED|net::ERR_ABORTED|aborted|network changed/iu.test(text);
|
||||
}
|
||||
|
||||
function networkAlertEvent(item, promptTimes) {
|
||||
const failureText = item.failureKind ?? item.failure ?? item.errorText ?? null;
|
||||
return {
|
||||
|
||||
Reference in New Issue
Block a user