fix: detect workbench recovery authority fanout

2026-07-08 19:32:56 +02:00
parent bd4722c9a8
commit 2b0778f21f
9 changed files with 509 additions and 13 deletions
@@ -65,6 +65,27 @@ const browserFreezePolicy = {
  },
 };

+async function runObserveAnalyzer(files: Record<string, string>): Promise<Record<string, any>> {
+  const stateDir = await mkdtemp(join(tmpdir(), "unidesk-web-observe-analyzer-"));
+  const analyzerPath = join(stateDir, "analyze.mjs");
+  await writeFile(analyzerPath, nodeWebObserveAnalyzerSource(), { mode: 0o700 });
+  for (const [name, content] of Object.entries(files)) {
+    await writeFile(join(stateDir, name), content.endsWith("\n") ? content : content + "\n");
+  }
+  const result = spawnSync("bun", [analyzerPath, stateDir], {
+    cwd: join(import.meta.dir, "../../.."),
+    env: {
+      ...process.env,
+      UNIDESK_WEB_OBSERVE_ANALYZE_TAIL_SAMPLES: "0",
+      UNIDESK_WEB_OBSERVE_ALERT_THRESHOLDS_JSON: JSON.stringify(alertThresholds),
+      UNIDESK_WEB_OBSERVE_BROWSER_FREEZE_POLICY_JSON: JSON.stringify(browserFreezePolicy),
+    },
+    encoding: "utf8",
+  });
+  assert.equal(result.status, 0, result.stderr || result.stdout);
+  return JSON.parse(await readFile(join(stateDir, "analysis", "report.json"), "utf8"));
+}
+
 test("observe analyzer classifies stale completed session rail when a newer turn is running", async () => {
  const stateDir = await mkdtemp(join(tmpdir(), "unidesk-web-observe-analyzer-"));
  const analyzerPath = join(stateDir, "analyze.mjs");
@@ -163,3 +184,105 @@ test("observe analyzer classifies stale completed session rail when a newer turn
  assert.equal(report.sampleMetrics.workbenchTurnStateTriad.summary.invalidRowCount, 1);
  assert.equal(report.sampleMetrics.workbenchTurnStateTriad.summary.cardFinalResponseMismatchCount, 0);
 }, 20_000);
+
+test("observe analyzer flags automatic recovery legacy fanout authority", async () => {
+  const samples = [
+    JSON.stringify({
+      seq: 1,
+      ts: "2026-07-01T15:21:25.000Z",
+      path: "/workbench/sessions/ses_auto",
+      url: "https://hwlab.example.test/workbench/sessions/ses_auto",
+      pageRole: "control",
+      pageId: "control-test",
+      routeSessionId: "ses_auto",
+      activeSessionId: "ses_auto",
+      turns: []
+    })
+  ].join("\n") + "\n";
+  const network = [
+    JSON.stringify({
+      ts: "2026-07-01T15:21:26.000Z",
+      type: "response",
+      method: "POST",
+      status: 202,
+      url: "https://hwlab.example.test/v1/web-performance",
+      webPerformancePayload: {
+        parseStatus: "parsed",
+        schemaVersion: "hwlab-web-performance-v2",
+        events: [{
+          kind: "workbench_ui_event",
+          eventType: "runtime_diagnostic",
+          module: "workbench-stream-transport",
+          diagnosticCode: "workbench_recovery_action",
+          recoveryAction: "hydrate-trace-events",
+          rootCause: "sse-gap",
+          eventCount: 0
+        }]
+      }
+    }),
+    JSON.stringify({ ts: "2026-07-01T15:21:27.000Z", type: "response", method: "GET", status: 200, url: "https://hwlab.example.test/v1/workbench/turns/trc_auto", bodySummary: { pathKind: "workbench-turn", terminalEvidenceCount: 1, traceIds: ["trc_auto"] } }),
+    JSON.stringify({ ts: "2026-07-01T15:21:28.000Z", type: "response", method: "GET", status: 200, url: "https://hwlab.example.test/v1/workbench/sessions/ses_auto/messages", bodySummary: { pathKind: "workbench-session-messages", terminalEvidenceCount: 1, sessionIds: ["ses_auto"] } }),
+    JSON.stringify({ ts: "2026-07-01T15:21:29.000Z", type: "response", method: "GET", status: 200, url: "https://hwlab.example.test/v1/workbench/traces/trc_auto/events", bodySummary: { pathKind: "workbench-trace-events", traceEventLikeCount: 3, traceIds: ["trc_auto"] } }),
+    JSON.stringify({ ts: "2026-07-01T15:21:30.000Z", type: "response", method: "GET", status: 200, url: "https://hwlab.example.test/v1/workbench/sync?sessionId=ses_auto&traceId=trc_auto&since=7", bodySummary: { pathKind: "workbench-sync-replay", eventCount: 1, cursorOutboxSeq: 12, realtimeAuthority: "workbench-realtime-authority-v2", terminalSeal: true } })
+  ].join("\n") + "\n";
+
+  const report = await runObserveAnalyzer({ "samples.jsonl": samples, "network.jsonl": network });
+  const finding = report.findings.find((item: Record<string, unknown>) => item.id === "workbench-automatic-recovery-fanout-authority");
+
+  assert.equal(finding?.rootCause, "workbench_automatic_recovery_legacy_fanout");
+  assert.equal(finding?.recoveryAuthority.summary.forbiddenAutomaticRecoveryCount, 1);
+  assert.equal(finding?.recoveryAuthority.summary.legacyFanoutWindowCount, 1);
+  assert.equal(finding?.recoveryAuthority.summary.syncReplayEvidenceCount, 1);
+  assert.equal(report.runtimeAlerts.webPerformanceRuntimeDiagnostics[0].recoveryAction, "hydrate-trace-events");
+}, 20_000);
+
+test("observe analyzer allows explicit detail trace reads when sync authority evidence exists", async () => {
+  const samples = JSON.stringify({
+    seq: 1,
+    ts: "2026-07-01T15:31:25.000Z",
+    path: "/workbench/sessions/ses_detail",
+    url: "https://hwlab.example.test/workbench/sessions/ses_detail",
+    pageRole: "control",
+    pageId: "control-test",
+    routeSessionId: "ses_detail",
+    activeSessionId: "ses_detail",
+    turns: []
+  }) + "\n";
+  const network = [
+    JSON.stringify({ ts: "2026-07-01T15:31:26.000Z", type: "response", method: "GET", status: 200, url: "https://hwlab.example.test/v1/workbench/traces/trc_detail/events", bodySummary: { pathKind: "workbench-trace-events", traceEventLikeCount: 2, traceIds: ["trc_detail"], authority: "trace-detail-only", detailProjection: true } }),
+    JSON.stringify({
+      ts: "2026-07-01T15:31:27.000Z",
+      type: "response",
+      method: "POST",
+      status: 202,
+      url: "https://hwlab.example.test/v1/web-performance",
+      webPerformancePayload: {
+        parseStatus: "parsed",
+        schemaVersion: "hwlab-web-performance-v2",
+        events: [{
+          kind: "workbench_ui_event",
+          eventType: "runtime_diagnostic",
+          module: "workbench-sync-replay",
+          diagnosticCode: "workbench_sync_replay_applied",
+          rootCause: "sse-gap",
+          eventCount: 1,
+          sinceOutboxSeq: 7,
+          syncCursorOutboxSeq: 12,
+          realtimeAuthority: "workbench-realtime-authority-v2",
+          entityFamilyCount: 1,
+          entityFamilies: "turn",
+          maxEntityVersion: 3,
+          projectionRevision: "rev_detail",
+          terminalSeal: true,
+          detailProjection: false
+        }]
+      }
+    })
+  ].join("\n") + "\n";
+
+  const report = await runObserveAnalyzer({ "samples.jsonl": samples, "network.jsonl": network });
+  const finding = report.findings.find((item: Record<string, unknown>) => item.id === "workbench-automatic-recovery-fanout-authority");
+
+  assert.equal(finding, undefined);
+  assert.equal(report.runtimeAlerts.webPerformanceRuntimeDiagnostics[0].syncCursorOutboxSeq, 12);
+}, 20_000);