我是一个实用主义者，所以我会试着从成本和收益的角度来看待这件事。

请注意，对于我给出的任何缺点，我都认识到它们是可以解决的。这就是为什么我不把任何事情看得非黑即白，而是成本和收益。

优势

更容易的状态跟踪-不需要使用cookie，表单提交，本地存储，会话存储等来记住2个页面加载之间的状态。 每个页面(页眉、页脚、logo、版权横幅等)上的样板内容在每次典型的浏览器会话中只加载一次。 切换“页面”时没有开销延迟。

缺点

Performance monitoring - hands tied: Most browser-level performance monitoring solutions I have seen focus exclusively on page load time only, like time to first byte, time to build DOM, network round trip for the HTML, onload event, etc. Updating the page post-load via AJAX would not be measured. There are solutions which let you instrument your code to record explicit measures, like when clicking a link, start a timer, then end a timer after rendering the AJAX results, and send that feedback. New Relic, for example, supports this functionality. By using a SPA, you have tied yourself to only a few possible tools. Security / penetration testing - hands tied: Automated security scans can have difficulty discovering links when your entire page is built dynamically by a SPA framework. There are probably solutions to this, but again, you've limited yourself. Bundling: It is easy to get into a situation when you are downloading all of the code needed for the entire web site on the initial page load, which can perform terribly for low-bandwidth connections. You can bundle your JavaScript and CSS files to try to load in more natural chunks as you go, but now you need to maintain that mapping and watch for unintended files to get pulled in via unrealized dependencies (just happened to me). Again, solvable, but with a cost. Big bang refactoring: If you want to make a major architectural change, like say, switch from one framework to another, to minimize risk, it's desirable to make incremental changes. That is, start using the new, migrate on some basis, like per-page, per-feature, etc., then drop the old after. With traditional multi-page app, you could switch one page from Angular to React, then switch another page in the next sprint. With a SPA, it's all or nothing. If you want to change, you have to change the entire application in one go. Complexity of navigation: Tooling exists to help maintain navigational context in SPA's, like history.js, Angular 2, most of which rely on either the URL framework (#) or the newer history API. If every page was a separate page, you don't need any of that. Complexity of figuring out code: We naturally think of web sites as pages. A multi-page app usually partitions code by page, which aids maintainability.

我再次承认，这些问题中的每一个都是可以解决的，只是需要付出一定的代价。 但总有一天，你把所有的时间都花在了解决问题上，而这些问题本来是可以避免的。这又回到了益处以及它们对你的重要性上。

对于谷歌、亚马逊等公司来说，他们的服务器在24/7模式下以最大容量运行，减少流量意味着真正的金钱——更少的硬件、更少的能源、更少的维护。将cpu使用从服务器转移到客户端是有回报的，spa大放异彩。到目前为止，利大于弊。 所以，SPA与否很大程度上取决于用例。

只是为了提到另一个可能不那么明显(对于web开发人员)的spa用例: 我目前正在寻找一种在嵌入式系统中实现gui的方法，基于浏览器的架构似乎对我很有吸引力。传统上，嵌入式系统中的ui没有太多的可能性——Java、Qt、wx等或适当的商业框架。几年前，Adobe试图通过flash进入市场，但似乎并不成功。

如今，由于“嵌入式系统”与几年前的大型机一样强大，基于浏览器的UI通过REST连接到控制单元是一种可能的解决方案。这样做的好处是，为用户界面提供了大量的工具，而且是免费的。(例如Qt需要20-30美元/销售单位的特许权使用费加上3000-4000美元/开发者)

对于这样的架构，SPA提供了许多优势——例如，桌面应用程序开发人员更熟悉的开发方法，减少服务器访问(通常在汽车行业，UI和系统混乱是分开的硬件，其中系统部分有一个RT操作系统)。

由于唯一的客户端是内置的浏览器，上面提到的js可用性、服务器端日志记录、安全性等缺点都不再重要。

让我们来看看最流行的SPA网站之一，GMail。

1. SPA非常适合响应性非常强的网站:

Server-side rendering is not as hard as it used to be with simple techniques like keeping a #hash in the URL, or more recently HTML5 pushState. With this approach the exact state of the web app is embedded in the page URL. As in GMail every time you open a mail a special hash tag is added to the URL. If copied and pasted to other browser window can open the exact same mail (provided they can authenticate). This approach maps directly to a more traditional query string, the difference is merely in the execution. With HTML5 pushState() you can eliminate the #hash and use completely classic URLs which can resolve on the server on the first request and then load via ajax on subsequent requests.

2. 使用SPA，我们不需要对服务器使用额外的查询来下载页面。

The number of pages user downloads during visit to my web site?? really how many mails some reads when he/she opens his/her mail account. I read >50 at one go. now the structure of the mails is almost the same. if you will use a server side rendering scheme the server would then render it on every request(typical case). - security concern - you should/ should not keep separate pages for the admins/login that entirely depends upon the structure of you site take paytm.com for example also making a web site SPA does not mean that you open all the endpoints for all the users I mean I use forms auth with my spa web site. - in the probably most used SPA framework Angular JS the dev can load the entire html temple from the web site so that can be done depending on the users authentication level. pre loading html for all the auth types isn't SPA.

3.还有其他的优势吗?别再听别人说了。

现在，您可以放心地假设客户机将拥有支持javascript的浏览器。 网站只有一个入口。正如我之前提到的，状态维护是可能的，你可以有任意数量的入口点，但你应该有一个确定的入口点。 即使在SPA用户只看到他有适当的权利。你不需要一次注射所有的东西。加载不同的html模板和javascript async也是一个有效的SPA的一部分。

我能想到的优点是:

rendering html obviously takes some resources now every user visiting you site is doing this. also not only rendering major logics are now done client side instead of server side. date time issues - I just give the client UTC time is a pre set format and don't even care about the time zones I let javascript handle it. this is great advantage to where I had to guess time zones based on location derived from users IP. to me state is more nicely maintained in an SPA because once you have set a variable you know it will be there. this gives a feel of developing an app rather than a web page. this helps a lot typically in making sites like foodpanda, flipkart, amazon. because if you are not using client side state you are using expensive sessions. websites surely are extremely responsive - I'll take an extreme example for this try making a calculator in a non SPA website(I know its weird).

来自评论的更新

It doesn't seem like anyone mentioned about sockets and long-polling. If you log out from another client say mobile app, then your browser should also log out. If you don't use SPA, you have to re-create the socket connection every time there is a redirect. This should also work with any updates in data like notifications, profile update etc An alternate perspective: Aside from your website, will your project involve a native mobile app? If yes, you are most likely going to be feeding raw data to that native app from a server (ie JSON) and doing client-side processing to render it, correct? So with this assertion, you're ALREADY doing a client-side rendering model. Now the question becomes, why shouldn't you use the same model for the website-version of your project? Kind of a no-brainer. Then the question becomes whether you want to render server-side pages only for SEO benefits and convenience of shareable/bookmarkable URLs

我是一个实用主义者，所以我会试着从成本和收益的角度来看待这件事。

请注意，对于我给出的任何缺点，我都认识到它们是可以解决的。这就是为什么我不把任何事情看得非黑即白，而是成本和收益。

优势

更容易的状态跟踪-不需要使用cookie，表单提交，本地存储，会话存储等来记住2个页面加载之间的状态。 每个页面(页眉、页脚、logo、版权横幅等)上的样板内容在每次典型的浏览器会话中只加载一次。 切换“页面”时没有开销延迟。

缺点

Performance monitoring - hands tied: Most browser-level performance monitoring solutions I have seen focus exclusively on page load time only, like time to first byte, time to build DOM, network round trip for the HTML, onload event, etc. Updating the page post-load via AJAX would not be measured. There are solutions which let you instrument your code to record explicit measures, like when clicking a link, start a timer, then end a timer after rendering the AJAX results, and send that feedback. New Relic, for example, supports this functionality. By using a SPA, you have tied yourself to only a few possible tools. Security / penetration testing - hands tied: Automated security scans can have difficulty discovering links when your entire page is built dynamically by a SPA framework. There are probably solutions to this, but again, you've limited yourself. Bundling: It is easy to get into a situation when you are downloading all of the code needed for the entire web site on the initial page load, which can perform terribly for low-bandwidth connections. You can bundle your JavaScript and CSS files to try to load in more natural chunks as you go, but now you need to maintain that mapping and watch for unintended files to get pulled in via unrealized dependencies (just happened to me). Again, solvable, but with a cost. Big bang refactoring: If you want to make a major architectural change, like say, switch from one framework to another, to minimize risk, it's desirable to make incremental changes. That is, start using the new, migrate on some basis, like per-page, per-feature, etc., then drop the old after. With traditional multi-page app, you could switch one page from Angular to React, then switch another page in the next sprint. With a SPA, it's all or nothing. If you want to change, you have to change the entire application in one go. Complexity of navigation: Tooling exists to help maintain navigational context in SPA's, like history.js, Angular 2, most of which rely on either the URL framework (#) or the newer history API. If every page was a separate page, you don't need any of that. Complexity of figuring out code: We naturally think of web sites as pages. A multi-page app usually partitions code by page, which aids maintainability.

我再次承认，这些问题中的每一个都是可以解决的，只是需要付出一定的代价。 但总有一天，你把所有的时间都花在了解决问题上，而这些问题本来是可以避免的。这又回到了益处以及它们对你的重要性上。

如果没有首先定义如何在服务器端解决安全性和API稳定性问题，就不要考虑使用SPA。然后，您将看到使用SPA的一些真正优点。具体来说，如果您使用实现OAUTH 2.0的rest式服务器来实现安全性，那么您将实现两个基本的关注点分离，从而降低开发和维护成本。

这将把会话(以及它的安全性)转移到SPA上，从而减轻服务器的所有开销。 您的API变得既稳定又易于扩展。

前面曾暗示过，但未明确;如果您的目标是部署Android和Apple应用程序，那么编写一个JavaScript SPA，它由本地调用包装，以在浏览器(Android或Apple)中托管屏幕，从而消除了同时维护Apple代码库和Android代码库的需要。

我知道这是一个老问题，但我想补充单页应用程序的另一个缺点:

If you build an API that returns results in a data language (such as XML or JSON) rather than a formatting language (like HTML), you are enabling greater application interoperability, for example, in business-to-business (B2B) applications. Such interoperability has great benefits but does allow people to write software to "mine" (or steal) your data. This particular disadvantage is common to all APIs that use a data language, and not to SPAs in general (indeed, an SPA that asks the server for pre-rendered HTML avoids this, but at the expense of poor model/view separation). This risk exposed by this disadvantage can be mitigated by various means, such as request limiting and connection blocking, etc.