Rediger

Del via


ASP.NET Core Blazor performance best practices

Note

This isn't the latest version of this article. For the current release, see the .NET 8 version of this article.

Warning

This version of ASP.NET Core is no longer supported. For more information, see .NET and .NET Core Support Policy. For the current release, see the .NET 8 version of this article.

Important

This information relates to a pre-release product that may be substantially modified before it's commercially released. Microsoft makes no warranties, express or implied, with respect to the information provided here.

For the current release, see the .NET 8 version of this article.

Blazor is optimized for high performance in most realistic application UI scenarios. However, the best performance depends on developers adopting the correct patterns and features.

Note

The code examples in this article adopt nullable reference types (NRTs) and .NET compiler null-state static analysis, which are supported in ASP.NET Core in .NET 6 or later.

Optimize rendering speed

Optimize rendering speed to minimize rendering workload and improve UI responsiveness, which can yield a ten-fold or higher improvement in UI rendering speed.

Avoid unnecessary rendering of component subtrees

You might be able to remove the majority of a parent component's rendering cost by skipping the rerendering of child component subtrees when an event occurs. You should only be concerned about skipping the rerendering subtrees that are particularly expensive to render and are causing UI lag.

At runtime, components exist in a hierarchy. A root component (the first component loaded) has child components. In turn, the root's children have their own child components, and so on. When an event occurs, such as a user selecting a button, the following process determines which components to rerender:

  1. The event is dispatched to the component that rendered the event's handler. After executing the event handler, the component is rerendered.
  2. When a component is rerendered, it supplies a new copy of parameter values to each of its child components.
  3. After a new set of parameter values is received, each component decides whether to rerender. Components rerender if the parameter values may have changed, for example, if they're mutable objects.

The last two steps of the preceding sequence continue recursively down the component hierarchy. In many cases, the entire subtree is rerendered. Events targeting high-level components can cause expensive rerendering because every component below the high-level component must rerender.

To prevent rendering recursion into a particular subtree, use either of the following approaches:

  • Ensure that child component parameters are of primitive immutable types, such as string, int, bool, DateTime, and other similar types. The built-in logic for detecting changes automatically skips rerendering if the primitive immutable parameter values haven't changed. If you render a child component with <Customer CustomerId="item.CustomerId" />, where CustomerId is an int type, then the Customer component isn't rerendered unless item.CustomerId changes.
  • Override ShouldRender:
    • To accept nonprimitive parameter values, such as complex custom model types, event callbacks, or RenderFragment values.
    • If authoring a UI-only component that doesn't change after the initial render, regardless of parameter value changes.

The following airline flight search tool example uses private fields to track the necessary information to detect changes. The previous inbound flight identifier (prevInboundFlightId) and previous outbound flight identifier (prevOutboundFlightId) track information for the next potential component update. If either of the flight identifiers change when the component's parameters are set in OnParametersSet, the component is rerendered because shouldRender is set to true. If shouldRender evaluates to false after checking the flight identifiers, an expensive rerender is avoided:

@code {
    private int prevInboundFlightId = 0;
    private int prevOutboundFlightId = 0;
    private bool shouldRender;

    [Parameter]
    public FlightInfo? InboundFlight { get; set; }

    [Parameter]
    public FlightInfo? OutboundFlight { get; set; }

    protected override void OnParametersSet()
    {
        shouldRender = InboundFlight?.FlightId != prevInboundFlightId
            || OutboundFlight?.FlightId != prevOutboundFlightId;

        prevInboundFlightId = InboundFlight?.FlightId ?? 0;
        prevOutboundFlightId = OutboundFlight?.FlightId ?? 0;
    }

    protected override bool ShouldRender() => shouldRender;
}

An event handler can also set shouldRender to true. For most components, determining rerendering at the level of individual event handlers usually isn't necessary.

For more information, see the following resources:

Virtualization

When rendering large amounts of UI within a loop, for example, a list or grid with thousands of entries, the sheer quantity of rendering operations can lead to a lag in UI rendering. Given that the user can only see a small number of elements at once without scrolling, it's often wasteful to spend time rendering elements that aren't currently visible.

Blazor provides the Virtualize<TItem> component to create the appearance and scroll behaviors of an arbitrarily-large list while only rendering the list items that are within the current scroll viewport. For example, a component can render a list with 100,000 entries but only pay the rendering cost of 20 items that are visible.

For more information, see ASP.NET Core Razor component virtualization.

Create lightweight, optimized components

Most Razor components don't require aggressive optimization efforts because most components don't repeat in the UI and don't rerender at high frequency. For example, routable components with an @page directive and components used to render high-level pieces of the UI, such as dialogs or forms, most likely appear only one at a time and only rerender in response to a user gesture. These components don't usually create high rendering workload, so you can freely use any combination of framework features without much concern about rendering performance.

However, there are common scenarios where components are repeated at scale and often result in poor UI performance:

  • Large nested forms with hundreds of individual elements, such as inputs or labels.
  • Grids with hundreds of rows or thousands of cells.
  • Scatter plots with millions of data points.

If modelling each element, cell, or data point as a separate component instance, there are often so many of them that their rendering performance becomes critical. This section provides advice on making such components lightweight so that the UI remains fast and responsive.

Avoid thousands of component instances

Each component is a separate island that can render independently of its parents and children. By choosing how to split the UI into a hierarchy of components, you are taking control over the granularity of UI rendering. This can result in either good or poor performance.

By splitting the UI into separate components, you can have smaller portions of the UI rerender when events occur. In a table with many rows that have a button in each row, you may be able to have only that single row rerender by using a child component instead of the whole page or table. However, each component requires additional memory and CPU overhead to deal with its independent state and rendering lifecycle.

In a test performed by the ASP.NET Core product unit engineers, a rendering overhead of around 0.06 ms per component instance was seen in a Blazor WebAssembly app. The test app rendered a simple component that accepts three parameters. Internally, the overhead is largely due to retrieving per-component state from dictionaries and passing and receiving parameters. By multiplication, you can see that adding 2,000 extra component instances would add 0.12 seconds to the rendering time and the UI would begin feeling slow to users.

It's possible to make components more lightweight so that you can have more of them. However, a more powerful technique is often to avoid having so many components to render. The following sections describe two approaches that you can take.

For more information on memory management, see Host and deploy ASP.NET Core server-side Blazor apps.

Inline child components into their parents

Consider the following portion of a parent component that renders child components in a loop:

<div class="chat">
    @foreach (var message in messages)
    {
        <ChatMessageDisplay Message="message" />
    }
</div>

ChatMessageDisplay.razor:

<div class="chat-message">
    <span class="author">@Message.Author</span>
    <span class="text">@Message.Text</span>
</div>

@code {
    [Parameter]
    public ChatMessage? Message { get; set; }
}

The preceding example performs well if thousands of messages aren't shown at once. To show thousands of messages at once, consider not factoring out the separate ChatMessageDisplay component. Instead, inline the child component into the parent. The following approach avoids the per-component overhead of rendering so many child components at the cost of losing the ability to rerender each child component's markup independently:

<div class="chat">
    @foreach (var message in messages)
    {
        <div class="chat-message">
            <span class="author">@message.Author</span>
            <span class="text">@message.Text</span>
        </div>
    }
</div>
Define reusable RenderFragments in code

You might be factoring out child components purely as a way of reusing rendering logic. If that's the case, you can create reusable rendering logic without implementing additional components. In any component's @code block, define a RenderFragment. Render the fragment from any location as many times as needed:

@RenderWelcomeInfo

<p>Render the welcome content a second time:</p>

@RenderWelcomeInfo

@code {
    private RenderFragment RenderWelcomeInfo = @<p>Welcome to your new app!</p>;
}

To make RenderTreeBuilder code reusable across multiple components, declare the RenderFragment public and static:

public static RenderFragment SayHello = @<h1>Hello!</h1>;

SayHello in the preceding example can be invoked from an unrelated component. This technique is useful for building libraries of reusable markup snippets that render without per-component overhead.

RenderFragment delegates can accept parameters. The following component passes the message (message) to the RenderFragment delegate:

<div class="chat">
    @foreach (var message in messages)
    {
        @ChatMessageDisplay(message)
    }
</div>

@code {
    private RenderFragment<ChatMessage> ChatMessageDisplay = message =>
        @<div class="chat-message">
            <span class="author">@message.Author</span>
            <span class="text">@message.Text</span>
        </div>;
}

The preceding approach reuses rendering logic without per-component overhead. However, the approach doesn't permit refreshing the subtree of the UI independently, nor does it have the ability to skip rendering the subtree of the UI when its parent renders because there's no component boundary. Assignment to a RenderFragment delegate is only supported in Razor component files (.razor), and event callbacks aren't supported.

For a non-static field, method, or property that can't be referenced by a field initializer, such as TitleTemplate in the following example, use a property instead of a field for the RenderFragment:

protected RenderFragment DisplayTitle =>
    @<div>
        @TitleTemplate
    </div>;

Don't receive too many parameters

If a component repeats extremely often, for example, hundreds or thousands of times, the overhead of passing and receiving each parameter builds up.

It's rare that too many parameters severely restricts performance, but it can be a factor. For a TableCell component that renders 4,000 times within a grid, each parameter passed to the component adds around 15 ms to the total rendering cost. Passing ten parameters requires around 150 ms and causes a UI rendering lag.

To reduce parameter load, bundle multiple parameters in a custom class. For example, a table cell component might accept a common object. In the following example, Data is different for every cell, but Options is common across all cell instances:

@typeparam TItem

...

@code {
    [Parameter]
    public TItem? Data { get; set; }
    
    [Parameter]
    public GridOptions? Options { get; set; }
}

However, consider that it might be an improvement not to have a table cell component, as shown in the preceding example, and instead inline its logic into the parent component.

Note

When multiple approaches are available for improving performance, benchmarking the approaches is usually required to determine which approach yields the best results.

For more information on generic type parameters (@typeparam), see the following resources:

Ensure cascading parameters are fixed

The CascadingValue component has an optional IsFixed parameter:

  • If IsFixed is false (the default), every recipient of the cascaded value sets up a subscription to receive change notifications. Each [CascadingParameter] is substantially more expensive than a regular [Parameter] due to the subscription tracking.
  • If IsFixed is true (for example, <CascadingValue Value="someValue" IsFixed="true">), recipients receive the initial value but don't set up a subscription to receive updates. Each [CascadingParameter] is lightweight and no more expensive than a regular [Parameter].

Setting IsFixed to true improves performance if there are a large number of other components that receive the cascaded value. Wherever possible, set IsFixed to true on cascaded values. You can set IsFixed to true when the supplied value doesn't change over time.

Where a component passes this as a cascaded value, IsFixed can also be set to true:

<CascadingValue Value="this" IsFixed="true">
    <SomeOtherComponents>
</CascadingValue>

For more information, see ASP.NET Core Blazor cascading values and parameters.

Avoid attribute splatting with CaptureUnmatchedValues

Components can elect to receive "unmatched" parameter values using the CaptureUnmatchedValues flag:

<div @attributes="OtherAttributes">...</div>

@code {
    [Parameter(CaptureUnmatchedValues = true)]
    public IDictionary<string, object>? OtherAttributes { get; set; }
}

This approach allows passing arbitrary additional attributes to the element. However, this approach is expensive because the renderer must:

  • Match all of the supplied parameters against the set of known parameters to build a dictionary.
  • Keep track of how multiple copies of the same attribute overwrite each other.

Use CaptureUnmatchedValues where component rendering performance isn't critical, such as components that aren't repeated frequently. For components that render at scale, such as each item in a large list or in the cells of a grid, try to avoid attribute splatting.

For more information, see ASP.NET Core Blazor attribute splatting and arbitrary parameters.

Implement SetParametersAsync manually

A significant source of per-component rendering overhead is writing incoming parameter values to [Parameter] properties. The renderer uses reflection to write the parameter values, which can lead to poor performance at scale.

In some extreme cases, you may wish to avoid the reflection and implement your own parameter-setting logic manually. This may be applicable when:

  • A component renders extremely often, for example, when there are hundreds or thousands of copies of the component in the UI.
  • A component accepts many parameters.
  • You find that the overhead of receiving parameters has an observable impact on UI responsiveness.

In extreme cases, you can override the component's virtual SetParametersAsync method and implement your own component-specific logic. The following example deliberately avoids dictionary lookups:

@code {
    [Parameter]
    public int MessageId { get; set; }

    [Parameter]
    public string? Text { get; set; }

    [Parameter]
    public EventCallback<string> TextChanged { get; set; }

    [Parameter]
    public Theme CurrentTheme { get; set; }

    public override Task SetParametersAsync(ParameterView parameters)
    {
        foreach (var parameter in parameters)
        {
            switch (parameter.Name)
            {
                case nameof(MessageId):
                    MessageId = (int)parameter.Value;
                    break;
                case nameof(Text):
                    Text = (string)parameter.Value;
                    break;
                case nameof(TextChanged):
                    TextChanged = (EventCallback<string>)parameter.Value;
                    break;
                case nameof(CurrentTheme):
                    CurrentTheme = (Theme)parameter.Value;
                    break;
                default:
                    throw new ArgumentException($"Unknown parameter: {parameter.Name}");
            }
        }

        return base.SetParametersAsync(ParameterView.Empty);
    }
}

In the preceding code, returning the base class SetParametersAsync runs the normal lifecycle method without assigning parameters again.

As you can see in the preceding code, overriding SetParametersAsync and supplying custom logic is complicated and laborious, so we don't generally recommend adopting this approach. In extreme cases, it can improve rendering performance by 20-25%, but you should only consider this approach in the extreme scenarios listed earlier in this section.

Don't trigger events too rapidly

Some browser events fire extremely frequently. For example, onmousemove and onscroll can fire tens or hundreds of times per second. In most cases, you don't need to perform UI updates this frequently. If events are triggered too rapidly, you may harm UI responsiveness or consume excessive CPU time.

Rather than use native events that rapidly fire, consider the use of JS interop to register a callback that fires less frequently. For example, the following component displays the position of the mouse but only updates at most once every 500 ms:

@implements IDisposable
@inject IJSRuntime JS

<h1>@message</h1>

<div @ref="mouseMoveElement" style="border:1px dashed red;height:200px;">
    Move mouse here
</div>

@code {
    private ElementReference mouseMoveElement;
    private DotNetObjectReference<MyComponent>? selfReference;
    private string message = "Move the mouse in the box";

    [JSInvokable]
    public void HandleMouseMove(int x, int y)
    {
        message = $"Mouse move at {x}, {y}";
        StateHasChanged();
    }

    protected override async Task OnAfterRenderAsync(bool firstRender)
    {
        if (firstRender)
        {
            selfReference = DotNetObjectReference.Create(this);
            var minInterval = 500;

            await JS.InvokeVoidAsync("onThrottledMouseMove", 
                mouseMoveElement, selfReference, minInterval);
        }
    }

    public void Dispose() => selfReference?.Dispose();
}

The corresponding JavaScript code registers the DOM event listener for mouse movement. In this example, the event listener uses Lodash's throttle function to limit the rate of invocations:

<script src="https://cdnjs.cloudflare.com/ajax/libs/lodash.js/4.17.20/lodash.min.js"></script>
<script>
  function onThrottledMouseMove(elem, component, interval) {
    elem.addEventListener('mousemove', _.throttle(e => {
      component.invokeMethodAsync('HandleMouseMove', e.offsetX, e.offsetY);
    }, interval));
  }
</script>

Avoid rerendering after handling events without state changes

Components inherit from ComponentBase, which automatically invokes StateHasChanged after the component's event handlers are invoked. In some cases, it might be unnecessary or undesirable to trigger a rerender after an event handler is invoked. For example, an event handler might not modify component state. In these scenarios, the app can leverage the IHandleEvent interface to control the behavior of Blazor's event handling.

Note

The approach in this section doesn't flow exceptions to error boundaries. For more information and demonstration code that supports error boundaries by calling ComponentBase.DispatchExceptionAsync, see AsNonRenderingEventHandler + ErrorBoundary = unexpected behavior (dotnet/aspnetcore #54543).

To prevent rerenders for all of a component's event handlers, implement IHandleEvent and provide a IHandleEvent.HandleEventAsync task that invokes the event handler without calling StateHasChanged.

In the following example, no event handler added to the component triggers a rerender, so HandleSelect doesn't result in a rerender when invoked.

HandleSelect1.razor:

@page "/handle-select-1"
@using Microsoft.Extensions.Logging
@implements IHandleEvent
@inject ILogger<HandleSelect1> Logger

<p>
    Last render DateTime: @dt
</p>

<button @onclick="HandleSelect">
    Select me (Avoids Rerender)
</button>

@code {
    private DateTime dt = DateTime.Now;

    private void HandleSelect()
    {
        dt = DateTime.Now;

        Logger.LogInformation("This event handler doesn't trigger a rerender.");
    }

    Task IHandleEvent.HandleEventAsync(
        EventCallbackWorkItem callback, object? arg) => callback.InvokeAsync(arg);
}

In addition to preventing rerenders after event handlers fire in a component in a global fashion, it's possible to prevent rerenders after a single event handler by employing the following utility method.

Add the following EventUtil class to a Blazor app. The static actions and functions at the top of the EventUtil class provide handlers that cover several combinations of arguments and return types that Blazor uses when handling events.

EventUtil.cs:

using System;
using System.Threading.Tasks;
using Microsoft.AspNetCore.Components;

public static class EventUtil
{
    public static Action AsNonRenderingEventHandler(Action callback)
        => new SyncReceiver(callback).Invoke;
    public static Action<TValue> AsNonRenderingEventHandler<TValue>(
            Action<TValue> callback)
        => new SyncReceiver<TValue>(callback).Invoke;
    public static Func<Task> AsNonRenderingEventHandler(Func<Task> callback)
        => new AsyncReceiver(callback).Invoke;
    public static Func<TValue, Task> AsNonRenderingEventHandler<TValue>(
            Func<TValue, Task> callback)
        => new AsyncReceiver<TValue>(callback).Invoke;

    private record SyncReceiver(Action callback) 
        : ReceiverBase { public void Invoke() => callback(); }
    private record SyncReceiver<T>(Action<T> callback) 
        : ReceiverBase { public void Invoke(T arg) => callback(arg); }
    private record AsyncReceiver(Func<Task> callback) 
        : ReceiverBase { public Task Invoke() => callback(); }
    private record AsyncReceiver<T>(Func<T, Task> callback) 
        : ReceiverBase { public Task Invoke(T arg) => callback(arg); }

    private record ReceiverBase : IHandleEvent
    {
        public Task HandleEventAsync(EventCallbackWorkItem item, object arg) => 
            item.InvokeAsync(arg);
    }
}

Call EventUtil.AsNonRenderingEventHandler to call an event handler that doesn't trigger a render when invoked.

In the following example:

  • Selecting the first button, which calls HandleClick1, triggers a rerender.
  • Selecting the second button, which calls HandleClick2, doesn't trigger a rerender.
  • Selecting the third button, which calls HandleClick3, doesn't trigger a rerender and uses event arguments (MouseEventArgs).

HandleSelect2.razor:

@page "/handle-select-2"
@using Microsoft.Extensions.Logging
@inject ILogger<HandleSelect2> Logger

<p>
    Last render DateTime: @dt
</p>

<button @onclick="HandleClick1">
    Select me (Rerenders)
</button>

<button @onclick="EventUtil.AsNonRenderingEventHandler(HandleClick2)">
    Select me (Avoids Rerender)
</button>

<button @onclick="EventUtil.AsNonRenderingEventHandler<MouseEventArgs>(HandleClick3)">
    Select me (Avoids Rerender and uses <code>MouseEventArgs</code>)
</button>

@code {
    private DateTime dt = DateTime.Now;

    private void HandleClick1()
    {
        dt = DateTime.Now;

        Logger.LogInformation("This event handler triggers a rerender.");
    }

    private void HandleClick2()
    {
        dt = DateTime.Now;

        Logger.LogInformation("This event handler doesn't trigger a rerender.");
    }
    
    private void HandleClick3(MouseEventArgs args)
    {
        dt = DateTime.Now;

        Logger.LogInformation(
            "This event handler doesn't trigger a rerender. " +
            "Mouse coordinates: {ScreenX}:{ScreenY}", 
            args.ScreenX, args.ScreenY);
    }
}

In addition to implementing the IHandleEvent interface, leveraging the other best practices described in this article can also help reduce unwanted renders after events are handled. For example, overriding ShouldRender in child components of the target component can be used to control rerendering.

Avoid recreating delegates for many repeated elements or components

Blazor's recreation of lambda expression delegates for elements or components in a loop can lead to poor performance.

The following component shown in the event handling article renders a set of buttons. Each button assigns a delegate to its @onclick event, which is fine if there aren't many buttons to render.

EventHandlerExample5.razor:

@page "/event-handler-example-5"

<h1>@heading</h1>

@for (var i = 1; i < 4; i++)
{
    var buttonNumber = i;

    <p>
        <button @onclick="@(e => UpdateHeading(e, buttonNumber))">
            Button #@i
        </button>
    </p>
}

@code {
    private string heading = "Select a button to learn its position";

    private void UpdateHeading(MouseEventArgs e, int buttonNumber)
    {
        heading = $"Selected #{buttonNumber} at {e.ClientX}:{e.ClientY}";
    }
}
@page "/event-handler-example-5"

<h1>@heading</h1>

@for (var i = 1; i < 4; i++)
{
    var buttonNumber = i;

    <p>
        <button @onclick="@(e => UpdateHeading(e, buttonNumber))">
            Button #@i
        </button>
    </p>
}

@code {
    private string heading = "Select a button to learn its position";

    private void UpdateHeading(MouseEventArgs e, int buttonNumber)
    {
        heading = $"Selected #{buttonNumber} at {e.ClientX}:{e.ClientY}";
    }
}

If a large number of buttons are rendered using the preceding approach, rendering speed is adversely impacted leading to a poor user experience. To render a large number of buttons with a callback for click events, the following example uses a collection of button objects that assign each button's @onclick delegate to an Action. The following approach doesn't require Blazor to rebuild all of the button delegates each time the buttons are rendered:

LambdaEventPerformance.razor:

@page "/lambda-event-performance"

<h1>@heading</h1>

@foreach (var button in Buttons)
{
    <p>
        <button @key="button.Id" @onclick="button.Action">
            Button #@button.Id
        </button>
    </p>
}

@code {
    private string heading = "Select a button to learn its position";

    private List<Button> Buttons { get; set; } = new();

    protected override void OnInitialized()
    {
        for (var i = 0; i < 100; i++)
        {
            var button = new Button();

            button.Id = Guid.NewGuid().ToString();

            button.Action = (e) =>
            {
                UpdateHeading(button, e);
            };

            Buttons.Add(button);
        }
    }

    private void UpdateHeading(Button button, MouseEventArgs e)
    {
        heading = $"Selected #{button.Id} at {e.ClientX}:{e.ClientY}";
    }

    private class Button
    {
        public string? Id { get; set; }
        public Action<MouseEventArgs> Action { get; set; } = e => { };
    }
}

Optimize JavaScript interop speed

Calls between .NET and JavaScript require additional overhead because:

  • Calls are asynchronous.
  • Parameters and return values are JSON-serialized to provide an easy-to-understand conversion mechanism between .NET and JavaScript types.

Additionally for server-side Blazor apps, these calls are passed across the network.

Avoid excessively fine-grained calls

Since each call involves some overhead, it can be valuable to reduce the number of calls. Consider the following code, which stores a collection of items in the browser's localStorage:

private async Task StoreAllInLocalStorage(IEnumerable<TodoItem> items)
{
    foreach (var item in items)
    {
        await JS.InvokeVoidAsync("localStorage.setItem", item.Id, 
            JsonSerializer.Serialize(item));
    }
}

The preceding example makes a separate JS interop call for each item. Instead, the following approach reduces the JS interop to a single call:

private async Task StoreAllInLocalStorage(IEnumerable<TodoItem> items)
{
    await JS.InvokeVoidAsync("storeAllInLocalStorage", items);
}

The corresponding JavaScript function stores the whole collection of items on the client:

function storeAllInLocalStorage(items) {
  items.forEach(item => {
    localStorage.setItem(item.id, JSON.stringify(item));
  });
}

For Blazor WebAssembly apps, rolling individual JS interop calls into a single call usually only improves performance significantly if the component makes a large number of JS interop calls.

Consider the use of synchronous calls

Call JavaScript from .NET

This section only applies to client-side components.

JS interop calls are asynchronous, regardless of whether the called code is synchronous or asynchronous. Calls are asynchronous to ensure that components are compatible across server-side and client-side render modes. On the server, all JS interop calls must be asynchronous because they're sent over a network connection.

If you know for certain that your component only runs on WebAssembly, you can choose to make synchronous JS interop calls. This has slightly less overhead than making asynchronous calls and can result in fewer render cycles because there's no intermediate state while awaiting results.

To make a synchronous call from .NET to JavaScript in a client-side component, cast IJSRuntime to IJSInProcessRuntime to make the JS interop call:

@inject IJSRuntime JS

...

@code {
    protected override void HandleSomeEvent()
    {
        var jsInProcess = (IJSInProcessRuntime)JS;
        var value = jsInProcess.Invoke<string>("javascriptFunctionIdentifier");
    }
}

When working with IJSObjectReference in ASP.NET Core 5.0 or later client-side components, you can use IJSInProcessObjectReference synchronously instead. IJSInProcessObjectReference implements IAsyncDisposable/IDisposable and should be disposed for garbage collection to prevent a memory leak, as the following example demonstrates:

@inject IJSRuntime JS
@implements IAsyncDisposable

...

@code {
    ...
    private IJSInProcessObjectReference? module;

    protected override async Task OnAfterRenderAsync(bool firstRender)
    {
        if (firstRender)
        {
            module = await JS.InvokeAsync<IJSInProcessObjectReference>("import", 
            "./scripts.js");
        }
    }

    ...

    async ValueTask IAsyncDisposable.DisposeAsync()
    {
        if (module is not null)
        {
            await module.DisposeAsync();
        }
    }
}

Call .NET from JavaScript

This section only applies to client-side components.

JS interop calls are asynchronous, regardless of whether the called code is synchronous or asynchronous. Calls are asynchronous to ensure that components are compatible across server-side and client-side render modes. On the server, all JS interop calls must be asynchronous because they're sent over a network connection.

If you know for certain that your component only runs on WebAssembly, you can choose to make synchronous JS interop calls. This has slightly less overhead than making asynchronous calls and can result in fewer render cycles because there's no intermediate state while awaiting results.

To make a synchronous call from JavaScript to .NET in a client-side component, use DotNet.invokeMethod instead of DotNet.invokeMethodAsync.

Synchronous calls work if:

  • The component is only rendered for execution on WebAssembly.
  • The called function returns a value synchronously. The function isn't an async method and doesn't return a .NET Task or JavaScript Promise.

This section only applies to client-side components.

JS interop calls are asynchronous, regardless of whether the called code is synchronous or asynchronous. Calls are asynchronous to ensure that components are compatible across server-side and client-side render modes. On the server, all JS interop calls must be asynchronous because they're sent over a network connection.

If you know for certain that your component only runs on WebAssembly, you can choose to make synchronous JS interop calls. This has slightly less overhead than making asynchronous calls and can result in fewer render cycles because there's no intermediate state while awaiting results.

To make a synchronous call from .NET to JavaScript in a client-side component, cast IJSRuntime to IJSInProcessRuntime to make the JS interop call:

@inject IJSRuntime JS

...

@code {
    protected override void HandleSomeEvent()
    {
        var jsInProcess = (IJSInProcessRuntime)JS;
        var value = jsInProcess.Invoke<string>("javascriptFunctionIdentifier");
    }
}

When working with IJSObjectReference in ASP.NET Core 5.0 or later client-side components, you can use IJSInProcessObjectReference synchronously instead. IJSInProcessObjectReference implements IAsyncDisposable/IDisposable and should be disposed for garbage collection to prevent a memory leak, as the following example demonstrates:

@inject IJSRuntime JS
@implements IAsyncDisposable

...

@code {
    ...
    private IJSInProcessObjectReference? module;

    protected override async Task OnAfterRenderAsync(bool firstRender)
    {
        if (firstRender)
        {
            module = await JS.InvokeAsync<IJSInProcessObjectReference>("import", 
            "./scripts.js");
        }
    }

    ...

    async ValueTask IAsyncDisposable.DisposeAsync()
    {
        if (module is not null)
        {
            await module.DisposeAsync();
        }
    }
}

Consider the use of unmarshalled calls

This section only applies to Blazor WebAssembly apps.

When running on Blazor WebAssembly, it's possible to make unmarshalled calls from .NET to JavaScript. These are synchronous calls that don't perform JSON serialization of arguments or return values. All aspects of memory management and translations between .NET and JavaScript representations are left up to the developer.

Warning

While using IJSUnmarshalledRuntime has the least overhead of the JS interop approaches, the JavaScript APIs required to interact with these APIs are currently undocumented and subject to breaking changes in future releases.

function jsInteropCall() {
  return BINDING.js_to_mono_obj("Hello world");
}
@inject IJSRuntime JS

@code {
    protected override void OnInitialized()
    {
        var unmarshalledJs = (IJSUnmarshalledRuntime)JS;
        var value = unmarshalledJs.InvokeUnmarshalled<string>("jsInteropCall");
    }
}

Use JavaScript [JSImport]/[JSExport] interop

JavaScript [JSImport]/[JSExport] interop for Blazor WebAssembly apps offers improved performance and stability over the JS interop API in framework releases prior to ASP.NET Core in .NET 7.

For more information, see JavaScript JSImport/JSExport interop with ASP.NET Core Blazor.

Ahead-of-time (AOT) compilation

Ahead-of-time (AOT) compilation compiles a Blazor app's .NET code directly into native WebAssembly for direct execution by the browser. AOT-compiled apps result in larger apps that take longer to download, but AOT-compiled apps usually provide better runtime performance, especially for apps that execute CPU-intensive tasks. For more information, see ASP.NET Core Blazor WebAssembly build tools and ahead-of-time (AOT) compilation.

Minimize app download size

Runtime relinking

For information on how runtime relinking minimizes an app's download size, see ASP.NET Core Blazor WebAssembly build tools and ahead-of-time (AOT) compilation.

Use System.Text.Json

Blazor's JS interop implementation relies on System.Text.Json, which is a high-performance JSON serialization library with low memory allocation. Using System.Text.Json shouldn't result in additional app payload size over adding one or more alternate JSON libraries.

For migration guidance, see How to migrate from Newtonsoft.Json to System.Text.Json.

Intermediate Language (IL) trimming

This section only applies to client-side Blazor scenarios.

Trimming unused assemblies from a Blazor WebAssembly app reduces the app's size by removing unused code in the app's binaries. For more information, see Configure the Trimmer for ASP.NET Core Blazor.

Linking a Blazor WebAssembly app reduces the app's size by trimming unused code in the app's binaries. The Intermediate Language (IL) Linker is only enabled when building in Release configuration. To benefit from this, publish the app for deployment using the dotnet publish command with the -c|--configuration option set to Release:

dotnet publish -c Release

Lazy load assemblies

This section only applies to client-side Blazor scenarios.

Load assemblies at runtime when the assemblies are required by a route. For more information, see Lazy load assemblies in ASP.NET Core Blazor WebAssembly.

Compression

This section only applies to Blazor WebAssembly apps.

When a Blazor WebAssembly app is published, the output is statically compressed during publish to reduce the app's size and remove the overhead for runtime compression. Blazor relies on the server to perform content negotiation and serve statically-compressed files.

After an app is deployed, verify that the app serves compressed files. Inspect the Network tab in a browser's developer tools and verify that the files are served with Content-Encoding: br (Brotli compression) or Content-Encoding: gz (Gzip compression). If the host isn't serving compressed files, follow the instructions in Host and deploy ASP.NET Core Blazor WebAssembly.

Disable unused features

This section only applies to client-side Blazor scenarios.

Blazor WebAssembly's runtime includes the following .NET features that can be disabled for a smaller payload size:

  • Blazor WebAssembly carries globalization resources required to display values, such as dates and currency, in the user's culture. If the app doesn't require localization, you may configure the app to support the invariant culture, which is based on the en-US culture.
  • Adopting invariant globalization only results in using non-localized timezone names. To trim timezone code and data from the app, apply the <InvariantTimezone> MSBuild property with a value of true in the app's project file:

    <PropertyGroup>
      <InvariantTimezone>true</InvariantTimezone>
    </PropertyGroup>
    

    Note

    <BlazorEnableTimeZoneSupport> overrides an earlier <InvariantTimezone> setting. We recommend removing the <BlazorEnableTimeZoneSupport> setting.

  • A data file is included to make timezone information correct. If the app doesn't require this feature, consider disabling it by setting the <BlazorEnableTimeZoneSupport> MSBuild property to false in the app's project file:

    <PropertyGroup>
      <BlazorEnableTimeZoneSupport>false</BlazorEnableTimeZoneSupport>
    </PropertyGroup>
    
  • Collation information is included to make APIs such as StringComparison.InvariantCultureIgnoreCase work correctly. If you're certain that the app doesn't require the collation data, consider disabling it by setting the BlazorWebAssemblyPreserveCollationData MSBuild property in the app's project file to false:

    <PropertyGroup>
      <BlazorWebAssemblyPreserveCollationData>false</BlazorWebAssemblyPreserveCollationData>
    </PropertyGroup>