blob: da7474b1ec136ebfd8176069000a62b6d11de91d [file] [log] [blame]
// Copyright 2019 The Fuchsia Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
import 'dart:convert';
import 'dart:io';
import 'package:args/args.dart';
import 'package:image/image.dart';
import 'package:logging/logging.dart';
import 'package:sl4f/sl4f.dart' as sl4f;
import 'package:test/test.dart';
int _ignoreAlpha(int pixel) => pixel & 0x00ffffff;
/// Returns true if [image] is all black.
bool _isAllBlack(Image image) => => _ignoreAlpha(pixel) == 0);
/// How many times to check the screen.
const _tries = 10;
/// How long to wait between screen checks.
const _delay = Duration(seconds: 10);
/// The path to the catapult converter binary. It must exist in the package.
const _catapultConverterPath = 'runtime_deps/catapult_converter';
/// We don't know if the system's clock is restarted from zero at reboot. So
/// we try to estimate. We assume that if the clock shows more than this long
/// since power-on, that reboot does not reset the clock. If it shows less
/// than this, we assume that reboot resets the clock.
const _maxElapsedSincePowerOn = Duration(minutes: 3);
/// An estimate of how long it took between reboot and timekeeper.cmx startup.
/// Used if the device's real time clock is not reset to zero on reboot.
const _rebootToTimekeeper = Duration(seconds: 15);
/// Wait at most this long for reboot to ensure all programs of interest have
/// hit their measurement checkpoints.
const _maxTestRuntime = Duration(seconds: 90);
/// Collects uptime information.
class Uptime {
// Elapsed time since last reboot.
Duration sinceReboot;
// Elapsed time from power-on until the time measurement was taken. Could be
// different than duration since reboot if the system monotonic clock is not
// reset to zero.
Duration sincePowerOn;
// Low estimate for when reboot measurement was taken, from host side, wall clock.
DateTime lowRebootWallClock;
// High estimate for when reboot measurement was taken, from host side, wall clock.
DateTime highRebootWallClock;
Uptime(this.lowRebootWallClock, this.sinceReboot, this.highRebootWallClock,
/// Returns an estimate of how long it was from Unix epoch to reboot.
Duration sinceEpoch() {
final Duration diff = highRebootWallClock.difference(lowRebootWallClock);
final Duration halfDiff = Duration(microseconds: diff.inMicroseconds ~/ 2);
final DateTime average = lowRebootWallClock.add(halfDiff);
final rebootWallClock =
return rebootWallClock - sinceReboot;
/// Elapsed time from power on until last reboot.
Duration get rebootSincePowerOn => sincePowerOn - sinceReboot;
void main(List<String> args) {
..level = Level.ALL
..onRecord.listen((rec) => print('[${rec.level}]: ${rec.message}'));
final parser = ArgParser()
help: 'If set, attempts to start basemgr, which is required for taking '
'a successful screenshot in some system configurations',
negatable: true,
defaultsTo: true);
final argResults = parser.parse(args);
final bool startBasemgr = argResults['start_basemgr'];
sl4f.Sl4f sl4fDriver;
sl4f.Scenic scenicDriver;
sl4f.Modular basemgrController;
final performance = sl4f.Performance(sl4fDriver);
final log = Logger('screen_is_not_black_test');
setUp(() async {
sl4fDriver = sl4f.Sl4f.fromEnvironment();
await sl4fDriver.startServer();
scenicDriver = sl4f.Scenic(sl4fDriver);
basemgrController = sl4f.Modular(sl4fDriver);
tearDown(() async {
await sl4fDriver.stopServer();
// Finds the duration passed between the instant system timer showed zero and
// the instant timekeeper was started.
// Uptime is tricky. The system's uptime as reported by the monotonic clock
// apparently doesn't always get reset to zero on a reboot. So we may get a
// monotonic clock reading that is showing a reading based on a zero reference
// point unrelated to a recent reboot.
// Here's a proposed pragmatic way out. Seeing as timekeeper is started
// fairly early in a device's lifetime, we try to detect how long ago
// timekeeper was started with reference to the zero point of the monotonic
// clock. If this time was too far in the past, assume that the duration
// timer was not reset. If it was reasonably recently with respect to the
// zero reference point (see _maxElapsedSincePowerOn), assume that the clock
// was indeed starting at zero. Otherwise adopt an arbitrary reference point
// which places startup of timekeeper.cmx about 15 seconds after reboot.
// While this makes relative time measurements meaningless, we can still
// compare successive runs and measure absolute time changes.
Future<Uptime> findRebootTimestamp(sl4f.Inspect inspect) async {
final lowRebootWallClock =;
final healthRoot = await inspect.snapshotRoot('timekeeper.cmx');
final highRebootWallClock =;
if (healthRoot == null) {'System uptime was not found, using zero values instead.');
return Uptime(,,,;
final uptimeNanos = healthRoot['current']['system_uptime_monotonic_nanos'];
final uptime = Duration(microseconds: uptimeNanos ~/ 1e3);
var sinceReboot = uptime;
if (sinceReboot > _maxElapsedSincePowerOn) {
// If the system monotonic clock measures duration since power-on,
// instead of duration since reboot, compute the adjusted uptime.
final start = Duration(
microseconds: healthRoot['start_time_monotonic_nanos'] ~/ 1e3);
sinceReboot = sinceReboot - start + _rebootToTimekeeper;
return Uptime(lowRebootWallClock, sinceReboot, highRebootWallClock, uptime);
List<sl4f.TestCaseResults> record(String programName, String nodeName,
dynamic root, String metricName, String renamedMetric, Duration reboot) {
final node = root[nodeName];
if (node == null) {
return [];
final metricValue = (node[metricName] ?? 0) ~/ 1e3;
final duration = Duration(microseconds: metricValue);
List<sl4f.TestCaseResults> results = [
'$renamedMetric/$programName', sl4f.Unit.milliseconds, [
duration.inMilliseconds.toDouble() - reboot.inMilliseconds.toDouble()
return results;
// Records all results under 'node'. The content of 'node' is assumed to
// consist only of timestamps.
// The optional renaming parameter is used to rename the final data point,
// and thereby aggregate several metrics into one.
List<sl4f.TestCaseResults> recordAll(String programName, String nodeName,
dynamic root, String renamedMetrics, Duration reboot,
{Map<Pattern, String> renaming}) {
final node = root[nodeName];
if (node == null) {
return [];
List<sl4f.TestCaseResults> result = [];
node.forEach((metric, measuredNanos) {
final Duration vd = Duration(microseconds: measuredNanos ~/ 1e3);
final Duration record = vd - reboot;
var metricName = '$renamedMetrics/$programName/$metric';
void replace(pattern, replacement) {
metricName = metricName.replaceFirst(pattern, replacement);
result.add(sl4f.TestCaseResults(metricName, sl4f.Unit.milliseconds,
return result;
/// Exports the reboot timing results to a fuchsiaperf file. Requires that
/// the `catapult_converter` binary is added to the runtime_deps, see the
/// file in this directory for the mechanics.
Future<void> exportTimings(String testSuite, Duration rebootDuration) async {
var rebootResults = [
// The host-side measurement of reboot duration.
sl4f.TestCaseResults('Reboot', sl4f.Unit.milliseconds,
final inspect = sl4f.Inspect(sl4fDriver);
final Uptime uptime = await findRebootTimestamp(inspect);
// Collect all health nodes and other metrics relevant for the boot process
// from the running programs. Some programs, like 'timekeeper' and `kcounter_inspect`
// are more interesting since they have metrics useful for analyzing the timing of the
// boot process.
final diagnostics = await inspect.snapshotAll();
for (final inspectResult in diagnostics) {
final label = inspectResult['moniker'];
final payload = inspectResult['payload'];
if (payload == null || payload['root'] == null) {
'Null payload/root for $label. Metadata: ${inspectResult['metadata']}');
final rootNode = payload['root'];
// For all nodes that have it, add health node information.
) +
// Record all durations under the node named 'fuchsia.inspect.Timestamps'.
renaming: {
RegExp(r'\/session-[0-9]+'): '/session-NNN',
List<Map<String, dynamic>> results = [];
for (final result in rebootResults) {
results.add(result.toJson(testSuite: testSuite));
File fuchsiaPerfFile = await sl4f.Dump().writeAsString(
'screen_is_not_black_test', 'fuchsiaperf.json', json.encode(results));
await performance.convertResults(
_catapultConverterPath, fuchsiaPerfFile, Platform.environment);'Catapult file created.');
test('the startup screen is not black', () async {
// Reboot Fuchsia so that we are testing the initial startup state.
// This is necessary because other tests (currently
// garnet_input_latency_benchmarks_test) kill Scenic, which would cause this
// test to fail. We use this as an opportunity to record the time taken
// to reboot as a performance test result.
var rebootDuration = await sl4fDriver.reboot();
try {
if (startBasemgr) {
// This call, and the 'shutdown' below are no-ops in configurations where
// modular is already running.
await basemgrController.boot();
} else {'The test was started with --no-start_basemgr, '
'so the test will not ensure that basemgr.cmx is running; '
'screenshotting should be provided through some other means.');
for (var attempt = 0; attempt < _tries; attempt++) {
try {
final screen = await scenicDriver.takeScreenshot(dumpName: 'screen');
if (!_isAllBlack(screen)) {
final pause = _maxTestRuntime - rebootDuration;
print('Saw a screen that is not black; waiting for $pause now.');
await Future.delayed(
pause, () => exportTimings('fuchsia.boot', rebootDuration));
} on sl4f.JsonRpcException {
print('Error taking screenshot; Scenic might not be ready yet.');
await Future.delayed(_delay);
fail('Screen was all black.');
} finally {
if (startBasemgr) {
await basemgrController.shutdown(forceShutdownBasemgr: false);
// This is a large test that waits for the DUT to come up and to start
// rendering something.
timeout: Timeout.none);