attempts to run the test rig with eventloopdelay measure and other feedback

Add a new performance test suite to simulate terminal scrolling and diagnose reported UI lag (indicated by eventloopDelay spikes) and high CPU usage. Make event loop monitoring opt-in to avoid overhead in production.

- Make EventLoopMonitor opt-in via GEMINI_EVENT_LOOP_MONITOR_ENABLED.
- Support passing custom environment variables to the child process in TestRig.
- Add CPU utilization percentage to the PerfTestHarness report.
- Inject child process event loop metrics into performance snapshots.
- Add a matrix of scrolling tests across different buffer configurations.

packages/test-utils/src/perf-test-harness.ts

@@ -367,13 +367,17 @@ export class PerfTestHarness {
 
       // Show CPU breakdown
       const cpuMs = `${(result.median.cpuTotalUs / 1000).toFixed(1)} ms`;
+      const utilization = (
+        (result.median.cpuTotalUs / 1000 / result.median.wallClockMs) *
+        100
+      ).toFixed(1);
       lines.push(
-        `  CPU: ${cpuMs} (user: ${formatUs(result.median.cpuUserUs)}, system: ${formatUs(result.median.cpuSystemUs)})`,
+        `  CPU: ${cpuMs} (${utilization}%) (user: ${formatUs(result.median.cpuUserUs)}, system: ${formatUs(result.median.cpuSystemUs)})`,
       );
 
       if (result.median.eventLoopDelayMaxMs > 0) {
         lines.push(
-          `  Event loop (runner): p50=${result.median.eventLoopDelayP50Ms.toFixed(1)}ms p95=${result.median.eventLoopDelayP95Ms.toFixed(1)}ms max=${result.median.eventLoopDelayMaxMs.toFixed(1)}ms`,
+          `  Event loop: p50=${result.median.eventLoopDelayP50Ms.toFixed(1)}ms p95=${result.median.eventLoopDelayP95Ms.toFixed(1)}ms max=${result.median.eventLoopDelayMaxMs.toFixed(1)}ms`,
         );
       }
 

packages/test-utils/src/test-rig.ts

@@ -1336,6 +1336,10 @@ export class TestRig {
     return logs;
   }
 
+  readTelemetryLogs(): ParsedLog[] {
+    return this._readAndParseTelemetryLog();
+  }
+
   readToolLogs() {
     // For Podman, first check if telemetry file exists and has content
     // If not, fall back to parsing from stdout

perf-tests/debug_session.ts

@@ -0,0 +1,63 @@
+/**
+ * @license
+ * Copyright 2026 Google LLC
+ * SPDX-License-Identifier: Apache-2.0
+ */
+import { Storage } from '../packages/core/src/config/storage';
+import { SessionSelector } from '../packages/cli/src/utils/sessionUtils';
+import { join } from 'node:path';
+import { mkdirSync, writeFileSync } from 'node:fs';
+
+async function main() {
+  const testDir = '/private/tmp/perf-scroll-up-true-true';
+  mkdirSync(testDir, { recursive: true });
+
+  const storage = new Storage(testDir);
+  await storage.initialize();
+
+  console.log('Project Temp Dir:', storage.getProjectTempDir());
+
+  const chatsDir = join(storage.getProjectTempDir(), 'chats');
+  mkdirSync(chatsDir, { recursive: true });
+
+  const sessionFile = join(chatsDir, 'session-2026-04-10T00-00-test.jsonl');
+  const metadata = {
+    sessionId: 'test-session-id',
+    projectHash: 'test-project-hash',
+    startTime: new Date().toISOString(),
+    lastUpdated: new Date().toISOString(),
+    kind: 'main',
+    hasUserOrAssistantMessage: true,
+    messageCount: 1,
+    summary: 'Test Session',
+  };
+
+  const msg = {
+    id: 'msg-0',
+    timestamp: new Date().toISOString(),
+    type: 'user',
+    content: 'Message content 0',
+  };
+
+  writeFileSync(
+    sessionFile,
+    JSON.stringify(metadata) + '\n' + JSON.stringify(msg) + '\n',
+  );
+  console.log('Created file:', sessionFile);
+
+  const selector = new SessionSelector(storage);
+  const sessions = await selector.listSessions();
+  console.log(
+    'Sessions found:',
+    sessions.map((s) => s.id),
+  );
+
+  try {
+    const resolved = await selector.resolveSession('test-session-id');
+    console.log('Resolved session:', resolved.sessionData.sessionId);
+  } catch (e) {
+    console.error('Failed to resolve:', e);
+  }
+}
+
+main();

perf-tests/perf-scrolling.test.ts

@@ -0,0 +1,257 @@
+/**
+ * @license
+ * Copyright 2026 Google LLC
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import { describe, it, beforeAll, afterAll, expect } from 'vitest';
+import { TestRig, PerfTestHarness } from '@google/gemini-cli-test-utils';
+import { join, dirname } from 'node:path';
+import { fileURLToPath } from 'node:url';
+
+const __dirname = dirname(fileURLToPath(import.meta.url));
+const BASELINES_PATH = join(__dirname, 'baselines.json');
+const TOLERANCE_PERCENT = 15;
+
+const SAMPLE_COUNT = 1;
+const WARMUP_COUNT = 0;
+
+describe.each([
+  { alt: true, term: true },
+  { alt: false, term: false },
+  { alt: true, term: false },
+  { alt: false, term: true },
+])('Scrolling Performance Tests (alt=$alt, term=$term)', ({ alt, term }) => {
+  let harness: PerfTestHarness;
+
+  beforeAll(() => {
+    harness = new PerfTestHarness({
+      baselinesPath: BASELINES_PATH,
+      defaultTolerancePercent: TOLERANCE_PERCENT,
+      sampleCount: SAMPLE_COUNT,
+      warmupCount: WARMUP_COUNT,
+    });
+  });
+
+  afterAll(async () => {
+    await harness.generateReport();
+  }, 30000);
+
+  async function createRealHistory(rig: TestRig, count: number) {
+    const run = await rig.runInteractive({
+      args: [],
+      env: {
+        HOME: rig.homeDir!,
+        GEMINI_API_KEY: 'test-api-key',
+      },
+    });
+
+    await run.expectText('type your message');
+
+    for (let i = 0; i < count; i++) {
+      // Send the /compress command directly
+      await run.sendText('/compress');
+      await new Promise((r) => setTimeout(r, 100));
+      // First enter to select from autocomplete
+      await run.sendText('\r');
+      await new Promise((r) => setTimeout(r, 100));
+      // Second enter to execute
+      await run.sendText('\r');
+      // Wait for the output to confirm execution
+      await run.expectText('nothing to compress.');
+    }
+
+    // Exit gracefully by sending SIGTERM to ensure telemetry is flushed
+    await run.kill();
+    await run.expectExit();
+  }
+
+  function readMetrics(
+    rig: TestRig,
+  ): { p50: number; p95: number; max: number }[] {
+    const logs = rig.readTelemetryLogs();
+    const metrics: { p50: number; p95: number; max: number }[] = [];
+
+    for (const log of logs) {
+      if (log.scopeMetrics) {
+        for (const sm of log.scopeMetrics) {
+          for (const m of sm.metrics) {
+            if (m.descriptor.name === 'gemini_cli.event_loop.delay') {
+              const p50 = m.dataPoints.find(
+                (dp: { attributes: { percentile: string }; value: number }) =>
+                  dp.attributes.percentile === 'p50',
+              )?.value;
+              const p95 = m.dataPoints.find(
+                (dp: { attributes: { percentile: string }; value: number }) =>
+                  dp.attributes.percentile === 'p95',
+              )?.value;
+              const max = m.dataPoints.find(
+                (dp: { attributes: { percentile: string }; value: number }) =>
+                  dp.attributes.percentile === 'max',
+              )?.value;
+
+              if (p50 !== undefined || p95 !== undefined || max !== undefined) {
+                metrics.push({ p50: p50 ?? 0, p95: p95 ?? 0, max: max ?? 0 });
+              }
+            }
+          }
+        }
+      }
+    }
+    return metrics;
+  }
+
+  it('straight-scroll-up: scrolling all the way up', async () => {
+    await harness.runScenario(
+      `straight-scroll-up-alt-${alt}-term-${term}`,
+      async () => {
+        const rig = new TestRig();
+        try {
+          rig.setup(`perf-scroll-up-${alt}-${term}`, {
+            settings: {
+              ui: {
+                useAlternateBuffer: alt,
+                terminalBuffer: term,
+              },
+            },
+          });
+
+          await createRealHistory(rig, 10);
+
+          const run = await rig.runInteractive({
+            args: ['--resume', '1'],
+            env: {
+              GEMINI_EVENT_LOOP_MONITOR_ENABLED: 'true',
+              GEMINI_MEMORY_MONITOR_INTERVAL: '500',
+              HOME: rig.homeDir!,
+              GEMINI_API_KEY: 'test-api-key',
+            },
+          });
+
+          await run.expectText('type your message');
+          // We expect to see the history text when resuming
+          await run.expectText('nothing to compress.', 30000);
+
+          const snapshot = await harness.measure(
+            'straight-scroll',
+            async () => {
+              // Send PageUp 10 times to go to top
+              for (let i = 0; i < 10; i++) {
+                run.ptyProcess.write('\x1b[5~');
+                await new Promise((r) => setImmediate(r));
+              }
+              await run.expectText('type your message');
+            },
+          );
+
+          await run.type('/exit');
+          await run.sendText('\r');
+          await run.sendText('\r');
+          await run.expectExit();
+          run.ptyProcess.kill();
+
+          const metrics = readMetrics(rig);
+
+          expect(metrics.length).toBeGreaterThan(0);
+
+          if (metrics.length > 0) {
+            snapshot.eventLoopDelayP50Ms = Math.max(
+              ...metrics.map((m) => m.p50),
+            );
+            snapshot.eventLoopDelayP95Ms = Math.max(
+              ...metrics.map((m) => m.p95),
+            );
+            snapshot.eventLoopDelayMaxMs = Math.max(
+              ...metrics.map((m) => m.max),
+            );
+          }
+
+          expect(snapshot.eventLoopDelayMaxMs).toBeLessThan(250);
+
+          return snapshot;
+        } finally {
+          rig.cleanup();
+          delete process.env['GEMINI_EVENT_LOOP_MONITOR_ENABLED'];
+          delete process.env['GEMINI_MEMORY_MONITOR_INTERVAL'];
+        }
+      },
+    );
+  }, 180000);
+
+  it('jitter-scroll: random scroll ups and downs', async () => {
+    await harness.runScenario(
+      `jitter-scroll-alt-${alt}-term-${term}`,
+      async () => {
+        const rig = new TestRig();
+        try {
+          rig.setup(`perf-jitter-scroll-${alt}-${term}`, {
+            settings: {
+              ui: {
+                useAlternateBuffer: alt,
+                terminalBuffer: term,
+              },
+            },
+          });
+
+          await createRealHistory(rig, 10);
+
+          const run = await rig.runInteractive({
+            args: ['--resume', '1'],
+            env: {
+              GEMINI_EVENT_LOOP_MONITOR_ENABLED: 'true',
+              GEMINI_MEMORY_MONITOR_INTERVAL: '500',
+              HOME: rig.homeDir!,
+              GEMINI_API_KEY: 'test-api-key',
+            },
+          });
+
+          await run.expectText('type your message');
+          await run.expectText('nothing to compress.', 30000);
+
+          const snapshot = await harness.measure('jitter-scroll', async () => {
+            // Simulate jitter: 3 up, 1 down. Do it 10 times to ensure we go all the way up.
+            for (let i = 0; i < 10; i++) {
+              for (let j = 0; j < 3; j++) {
+                run.ptyProcess.write('\x1b[5~'); // PageUp
+                await new Promise((r) => setImmediate(r));
+              }
+              run.ptyProcess.write('\x1b[6~'); // PageDown
+              await new Promise((r) => setImmediate(r));
+            }
+            await run.expectText('type your message');
+          });
+
+          await run.type('/exit');
+          await run.sendText('\r');
+          await run.sendText('\r');
+          await run.expectExit();
+          run.ptyProcess.kill();
+
+          const metrics = readMetrics(rig);
+
+          expect(metrics.length).toBeGreaterThan(0);
+
+          if (metrics.length > 0) {
+            snapshot.eventLoopDelayP50Ms = Math.max(
+              ...metrics.map((m) => m.p50),
+            );
+            snapshot.eventLoopDelayP95Ms = Math.max(
+              ...metrics.map((m) => m.p95),
+            );
+            snapshot.eventLoopDelayMaxMs = Math.max(
+              ...metrics.map((m) => m.max),
+            );
+          }
+
+          expect(snapshot.eventLoopDelayMaxMs).toBeLessThan(250);
+
+          return snapshot;
+        } finally {
+          rig.cleanup();
+          delete process.env['GEMINI_EVENT_LOOP_MONITOR_ENABLED'];
+          delete process.env['GEMINI_MEMORY_MONITOR_INTERVAL'];
+        }
+      },
+    );
+  }, 180000);
+});