Flutter 性能分析


It’s been said that “a fast app is great, but a smooth app is even better.” If your app isn’t rendering smoothly, how do you fix it? Where do you begin? This guide shows you where to start, steps to take, and tools that can help.


Diagnosing performance problems

分析应用的性能问题需要打开性能监控图层 (performance overlay) 来观察 UI 和 GPU 线程。在此之前,要确保是在 分析模式 下运行,而且当前设备不是虚拟机。使用用户可能采用的最慢设备来获取最佳结果。

To diagnose an app with performance problems, you’ll enable the performance overlay to look at the UI and raster threads. (The raster thread was previously known as the GPU thread.) Before you begin, you want to make sure that you’re running in profile mode, and that you’re not using an emulator. For best results, you might choose the slowest device that your users might use.


Connect to a physical device

几乎全部的 Flutter 应用性能调试都应该在真实的 Android 或者 iOS 设备上以 分析模式 进行。通常来说,调试模式或者是模拟器上运行的应用的性能指标和发布模式的表现并不相同。 应该考虑在用户使用的最慢的设备上检查性能。

Almost all performance debugging for Flutter applications should be conducted on a physical Android or iOS device, with your Flutter application running in profile mode. Using debug mode, or running apps on simulators or emulators, is generally not indicative of the final behavior of release mode builds. You should consider checking performance on the slowest device that your users might reasonably use.


Run in profile mode

除了一些调试性能问题所必须的额外方法, Flutter 的分析模式和发布模式的编译和运行基本相同。例如,分析模式为分析工具提供了追踪信息。

Flutter’s profile mode compiles and launches your application almost identically to release mode, but with just enough additional functionality to allow debugging performance problems. For example, profile mode provides tracing information to the profiling tools.


Launch the app in profile mode as follows:

  • 在 Android Studio 和 IntelliJ 使用 Run > Flutter Run main.dart in Profile Mode 选项

    In Android Studio and IntelliJ, use the Run > Flutter Run main.dart in Profile Mode menu item.

  • 在 VS Code 中,打开 launch.json 文件,设置 flutterMode 属性为 profile(当分析完成后,改回 release 或者 debug):

    In VS Code, open your launch.json file, and set the flutterMode property to profile (when done profiling, change it back to release or debug):

    "configurations": [
        "name": "Flutter",
        "request": "launch",
        "type": "dart",
        "flutterMode": "profile"
  • 命令行使用 --profile 参数运行

    From the command line, use the --profile flag:

    $ flutter run --profile

关于不同模式的更多信息,请参考文档: Flutter 的构建模式选择

For more information on the different modes, see Flutter’s build modes.

下面我们会从打开 DevTools、查看性能图层开始讲述。

You’ll begin by opening DevTools and viewing the performance overlay, as discussed in the next section.

运行 DevTools

Launch DevTools

Dart DevTool 提供诸如性能分析、堆测试以及显示代码覆盖率等功能。 DevTool 的 [Timeline] 界面可以让开发者逐帧分析应用的 UI 性能。

DevTools provides features like profiling, examining the heap, displaying code coverage, enabling the performance overlay, and a step-by-step debugger. DevTools’ [Timeline] allows you to investigate the UI performance of your application on a frame-by-frame basis.

一旦你的应用程序在分析模式下运行,即 运行 DevTools

Once your app is running in profile mode, launch DevTools.


The performance overlay

性能图层用两张图表显示应用的耗时信息。如果 UI 产生了卡顿(跳帧),这些图表可以帮助分析原因。图表在当前应用的最上层展示,但并不是用普通的 widget 方式绘制的—Flutter 引擎自身绘制了该图层来尽可能减少对性能的影响。每一张图表都代表当前线程的最近 300 帧表现。

The performance overlay displays statistics in two graphs that show where time is being spent in your app. If the UI is janky (skipping frames), these graphs help you figure out why. The graphs display on top of your running app, but they aren’t drawn like a normal widget—the Flutter engine itself paints the overlay and only minimally impacts performance. Each graph represents the last 300 frames for that thread.

本节阐述如何打开性能图层并用其来分析应用中卡顿的原因。下面的截图展示了 Flutter Gallery 样例的性能图层:

This section describes how to enable the performance overlay and use it to diagnose the cause of jank in your application. The following screenshot shows the performance overlay running on the Flutter Gallery example:

Screenshot of overlay showing zero jank

raster 线程的性能情况在上面,UI 线程显示在下面。

Performance overlay showing the raster thread (top), and UI thread (bottom).
The vertical green bars represent the current frame.


Interpreting the graphs

最顶部(标志了 “GPU”)的图形表示 raster 线程所花费的时间,底部的图表显示了 UI 线程所花费的时间。横跨图表中的白线代表了 16 ms 内沿竖轴的增量;如果这些线在图表中都没有超过它的话,说明你的运行帧率低于 60 Hz。而横轴则表示帧。只有当你的应用绘制时这个图表才会更新,所以如果它空闲的话,图表就不会动。

The top graph (marked “GPU”) shows the time spent by the raster thread, the bottom one graph shows the time spent by the UI thread. The white lines across the graphs show 16ms increments along the vertical axis; if the graph ever goes over one of these lines then you are running at less than 60Hz. The horizontal axis represents frames. The graph is only updated when your application paints, so if it’s idle the graph stops moving.

这个浮层只应在 分析模式 中使用,因为在 调试模式 下有意牺牲了性能来换取昂贵的断言以帮助开发,所以这时候的结果会有误导性。

The overlay should always be viewed in profile mode, since debug mode performance is intentionally sacrificed in exchange for expensive asserts that are intended to aid development, and thus the results are misleading.

每一帧都应该在 1/60 秒(大约 16 ms)内创建并显示。如果有一帧超时(任意图像)而无法显示,就导致了卡顿,图表之一就会展示出来一个红色竖条。如果是在 UI 图表出现了红色竖条,则表明 Dart 代码消耗了大量资源。而如果红色竖条是在 GPU 图表出现的,意味着场景太复杂导致无法快速渲染。

Each frame should be created and displayed within 1/60th of a second (approximately 16ms). A frame exceeding this limit (in either graph) fails to display, resulting in jank, and a vertical red bar appears in one or both of the graphs. If a red bar appears in the UI graph, the Dart code is too expensive If a red vertical bar appears in the GPU graph, the scene is too complicated to render quickly.

Screenshot of performance overlay showing jank with red bars

当两张图表都是红色时,就要开始对 UI 线程 (Dart VM) 进行诊断了。

The vertical red bars indicate that the current frame is expensive to both render and paint.
When both graphs display red, start by diagnosing the UI thread.

Flutter 的线程

Flutter’s threads

Flutter 使用多个线程来完成其必要的工作,图层中仅展示了其中两个线程。您写的所有 Dart 代码都在 UI 线程上运行。尽管您没有直接访问其他线程的权限,但是您对 UI 线程的操作会对其他线程产生性能影响。

Flutter uses several threads to do its work, though only two of the threads are shown in the overlay. All of your Dart code runs on the UI thread. Although you have no direct access to any other thread, your actions on the UI thread have performance consequences on other threads.


Platform thread

平台线程实际上就是主线程。Plugin 的代码将会在这里运行。想要了解更多信息,请参阅 Android 的 MainThread 以及 iOS 的 UIKit 文档。

The platform’s main thread. Plugin code runs here. For more information, see the UIKit documentation for iOS, or the MainThread documentation for Android. This thread is not shown in the performance overlay.

UI 线程

UI thread

UI 线程在 Dart VM 中执行 Dart 代码。该线程包括开发者写下的代码和 Flutter 框架根据应用行为生成的代码。当应用创建和展示场景的时候,UI 线程首先建立一个 图层树(layer tree) ,一个包含设备无关的渲染命令的轻量对象,并将图层树发送到 GPU 线程来渲染到设备上。 不要阻塞这个线程! 在性能图层的最低栏展示该线程。

The UI thread executes Dart code in the Dart VM. This thread includes code that you wrote, and code executed by Flutter’s framework on your app’s behalf. When your app creates and displays a scene, the UI thread creates a layer tree, a lightweight object containing device-agnostic painting commands, and sends the layer tree to the raster thread to be rendered on the device. Don’t block this thread! Shown in the bottom row of the performance overlay.

Raster 线程(以前叫 GPU 线程)

Raster thread (previously known as the GPU thread)

raster 线程拿到 layer tree,并将它交给 GPU(图形处理单元)。你无法直接与 GPU 线程或其数据通信,但如果该线程变慢,一定是开发者 Dart 代码中的某处导致的。图形库 Skia 在该线程运行,并在性能图层的最顶栏显示该线程。这个线程之前被叫做「GPU 线程」,因为它为 GPU 进行栅格化,但我们重新将它命名为「raster 线程」,这是因为许多开发者错误的(但是能理解)认为该线程运行在 GPU 单元。

The raster thread takes the layer tree and displays it by talking to the GPU (graphic processing unit). You cannot directly access the raster thread or its data but, if this thread is slow, it’s a result of something you’ve done in the Dart code. Skia, the graphics library, runs on this thread. Shown in the top row of the performance overlay. This thread was previously known as the “GPU thread” because it rasterizes for the GPU. But it is running on the CPU. We renamed it to “raster thread” because many developers wrongly (but understandably) assumed the thread runs on the GPU unit.


I/O thread

执行昂贵的操作(常见的有 I/O)以避免阻塞 UI 或者 raster 线程。这个线程将不会显示在 performance overlay 上。

Performs expensive tasks (mostly I/O) that would otherwise block either the UI or raster threads. This thread is not shown in the performance overlay.

你可以在 GitHub wiki 上的框架结构 (The Framework architecture) 一文中了解更多信息和一些视频内容,另外你可以在我们的社区中查看文章 The Layer Cake

For links to more information and videos, see The Framework architecture on the GitHub wiki, and the community article, The Layer Cake.


Displaying the performance overlay


You can toggle display of the performance overlay as follows:

  • 使用 Flutter Inspector

    Using the Flutter inspector

  • 从命令行启动

    From the command line

  • 写入代码


使用 Flutter inspector

Using the Flutter inspector

打开 PerformanceOverlay widget 最简单的方法是 IDE 中 Flutter 插件提供的 Flutter inspector,你可以在 开发者工具使用 Flutter inspector 工具 中找到。只需单击 Performance Overlay 按钮,即可在正在运行的应用程序上切换图层。

The easiest way to enable the PerformanceOverlay widget is from the Flutter inspector, which is available in the Inspector view in DevTools. Simply click the Performance Overlay button to toggle the overlay on your running app.


From the command line

使用 p 参数触发性能图层。

Toggle the performance overlay using the P key from the command line.



To enable the overlay programmatically, see Performance overlay, a section in the Debugging Flutter apps programmatically page.

定位 UI 图表中的问题

Identifying problems in the UI graph

如果性能图层的 UI 图表显示红色,就要从分析 Dart VM 开始着手了,即使 GPU 图表同样显示红色。

If the performance overlay shows red in the UI graph, start by profiling the Dart VM, even if the GPU graph also shows red.

定位 GPU 图表中的问题

Identifying problems in the GPU graph

有些情况下界面的图层树构造起来虽然容易,但在 raster 线程下渲染却很耗时。这种情况发生时,UI 图表没有红色,但 GPU 图表会显示红色。这时需要找出代码中导致渲染缓慢的原因。特定类型的负载对 GPU 来说会更加复杂。可能包括不必要的对 saveLayer 的调用,许多对象间的复杂操作,还可能是特定情形下的裁剪或者阴影。

Sometimes a scene results in a layer tree that is easy to construct, but expensive to render on the raster thread. When this happens, the UI graph has no red, but the GPU graph shows red. In this case, you’ll need to figure out what your code is doing that is causing rendering code to be slow. Specific kinds of workloads are more difficult for the GPU. They might involve unnecessary calls to saveLayer, intersecting opacities with multiple objects, and clips or shadows in specific situations.

如果推断的原因是动画中的卡顿的话,可以点击 Flutter inspector 中的 Slow Animations 按钮,来使动画速度减慢 5x。如果你想从更多方面控制动画速度,你可以参考 programmatically

If you suspect that the source of the slowness is during an animation, click the Slow Animations button in the Flutter inspector to slow animations down by 5x. If you want more control on the speed, you can also do this programmatically.

卡顿是第一帧发生的还是贯穿整个动画过程呢?如果是整个动画过程的话,会是裁剪导致的吗?也许有可以替代裁剪的方法来绘制场景。比如说,不透明图层的长方形中用尖角来取代圆角裁剪。如果是一个静态场景的淡入、旋转或者其他操作,可以尝试使用重绘边界 ([RepaintBoundary][])。

Is the slowness on the first frame, or on the whole animation? If it’s the whole animation, is clipping causing the slow down? Maybe there’s an alternative way of drawing the scene that doesn’t use clipping. For example, overlay opaque corners onto a square instead of clipping to a rounded rectangle. If it’s a static scene that’s being faded, rotated, or otherwise manipulated, a RepaintBoundary might help.


Checking for offscreen layers

保存图层 (saveLayer) 方法是 Flutter 框架中最重量的操作之一。更新屏幕时这个方法很有用,但它可能使应用变慢,如果不是必须的话,应该避免使用这个方法。即便没有显式地调用 saveLayer,也可能在其他操作中间接调用了该方法。可以使用棋盘画面以外的层 ([PerformanceOverlayLayer.checkerboardOffscreenLayers][]) 开关来检查场景是否使用了 saveLayer

The saveLayer method is one of the most expensive methods in the Flutter framework. It’s useful when applying post-processing to the scene, but it can slow your app and should be avoided if you don’t need it. Even if you don’t call saveLayer explicitly, implicit calls might happen on your behalf. You can check whether your scene is using saveLayer with the PerformanceOverlayLayer.checkerboardOffscreenLayers switch.

打开开关之后,运行应用并检查是否有图像的轮廓闪烁。如果有新的帧渲染的话,容器就会闪烁。举个例子,也许有一组对象的透明度要使用 saveLayer 来渲染。在这种情况下,相比通过 widget 树中高层次的父 widget 操作,单独对每个 widget 来应用透明度可能性能会更好。其他可能大量消耗资源的操作也同理,比如裁剪或者阴影。

Once the switch is enabled, run the app and look for any images that are outlined with a flickering box. The box flickers from frame to frame if a new frame is being rendered. For example, perhaps you have a group of objects with opacities that are rendered using saveLayer. In this case, it’s probably more performant to apply an opacity to each individual widget, rather than a parent widget higher up in the widget tree. The same goes for other potentially expensive operations, such as clipping or shadows.

当遇到对 saveLayer 的调用时,先问问自己:

When you encounter calls to saveLayer, ask yourself these questions:

  • 应用是否需要这个效果?

    Does the app need this effect?

  • 可以减少调用么?

    Can any of these calls be eliminated?

  • 可以对单独元素操作而不是一组元素么?

    Can I apply the same effect to an individual element instead of a group?


Checking for non-cached images

使用重绘边界 ([RepaintBoundary][]) 来缓存图片是个好主意,当需要的时候。

Caching an image with RepaintBoundary is good, when it makes sense.

从资源的角度看,最重量级的操作之一是用图像文件来渲染纹理。首先,需要从持久存储中取出压缩图像,然后解压缩到宿主存储中(GPU 存储),再传输到设备存储器中 (RAM) 。

One of the most expensive operations, from a resource perspective, is rendering a texture using an image file. First, the compressed image is fetched from persistent storage. The image is decompressed into host memory (GPU memory), and transferred to device memory (RAM).

也就是说,图像的 I/O 操作是重量级的。缓存提供了复杂层次的快照,这样就可以方便地渲染到随后的帧中。 因为光栅缓存入口的构建需要大量资源,同时增加了 GPU 存储的负载,所以只在必须时才缓存图片。

In other words, image I/O can be expensive. The cache provides snapshots of complex hierarchies so they are easier to render in subsequent frames. Because raster cache entries are expensive to construct and take up loads of GPU memory, cache images only where absolutely necessary.

打开覆盖层性能棋盘格光栅缓存图像 (PerformanceOverlayLayer.checkerboardRasterCacheImages) 开关可以检查哪些图片被缓存了。

You can see which images are being cached by enabling the PerformanceOverlayLayer.checkerboardRasterCacheImages switch.


Run the app and look for images rendered with a randomly colored checkerboard, indicating that the image is cached. As you interact with the scene, the checkerboarded images should remain constant—you don’t want to see flickering, which would indicate that the cached image is being re-cached.

大多数情况下,开发者都希望在网格里看到的是静态图片,而不是非静态图片。如果静态图片没有被缓存,可以将其放到重绘边界 ([RepaintBoundary][]) widget 中来缓存。虽然引擎也可能忽略 repaint boundary,如果它认为图像还不够复杂的话。

In most cases, you want to see checkerboards on static images, but not on non-static images. If a static image isn’t cached, you can cache it by placing it into a RepaintBoundary widget. Though the engine might still ignore a repaint boundary if it thinks the image isn’t complex enough.

检视 widget 重建性能

Viewing the widget rebuild profiler

Flutter 框架的设计使得构建达不到 60 fps 流畅度的应用变得困难。通常情况下如果卡顿,就是因为每一帧被重建的 UI 比需求更多的简单 bug。 Widget rebuild profiler 可以帮助调试和修复这些问题引起的 bug。

The Flutter framework is designed to make it hard to create applications that are not 60fps and smooth. Often, if you have jank, it’s because there is a simple bug causing more of the UI to be rebuilt each frame than required. The Widget rebuild profiler helps you debug and fix performance problems due to these sorts of bugs.

可以检视 widget inspector 中当前屏幕和帧下的 widget 重建数量。了解细节,可以参考 在 Android Studio 或类 IntelliJ 里开发 Flutter 应用 中的 显示性能数据

You can view the widget rebuilt counts for the current screen and frame in the Flutter plugin for Android Studio and IntelliJ. For details on how to do this, see Show performance data.



可以通过编写评分测试来测量和追踪应用的性能。 Flutter Driver 库提供了对评分的支持。基于这套测试框架就可以生成以下几项的测试标准:

You can measure and track your app’s performance by writing benchmark tests. The Flutter Driver library provides support for benchmarking. Using this integration test framework, you can generate metrics to track the following:

  • 卡顿


  • 下载大小

    Download size

  • 电池性能

    Battery efficiency

  • 启动时间

    Startup time


Tracking these benchmarks allows you to be informed when a regression is introduced that adversely affects performance.

了解更多,请参考 测试 Flutter 应用 中的 集成测试 一节。

For more information, see Integration testing, a section in Testing Flutter apps.


Other resources

以下链接提供了关于 Flutter 工具的使用和 Flutter 调试的更多信息:

The following resources provide more information on using Flutter’s tools and debugging in Flutter: