A couple of weeks ago I migrated a React project using redux to <a target="_blank" rel="noopener noreferrer" href="https://redux-toolkit.js.org/">Redux Toolkit</a>.
Redux has always been my preferred state management tool for React (which doesn&#x27;t mean I &quot;love&quot; it), but one of its main problems is the amount of boilerplate code it requires.
Because of that I had a couple of helper functions, but there was still some things not properly handled, like proper type inference on reducers and actions (I use TypeScript).
A bit by accident (I think it was on Twitter), I found out about Redux Toolkit (RTK from now on), an official library from the redux team, which provides some opinionated helpers to reduce the amount of boilerplate, and I decided to give it a chance.
These are my impressions.
<h3>TL;DR</h3>
I first started to use it in a proof of concept, and I quickly loved it.
RTK does not &quot;replace&quot; any of the redux principles, so you can progressively adopt it, by replacing small bits of code. My plan was to use it just for some reducers, but I liked it so much that I ended up migrating the whole project in a series of consecutive pull requests.
<h3>A bit of context</h3>
Before the migration, I used to follow the <a target="_blank" rel="noopener noreferrer" href="https://github.com/erikras/ducks-modular-redux">ducks modular pattern</a>, which promotes putting all your action types, action creators and the reducer handling those actions on the same module.
This fits pretty well with RTK, as in most of the cases you will use <a target="_blank" rel="noopener noreferrer" href="https://redux-toolkit.js.org/api/createSlice">slices</a>, and end up exporting the reducer and action creators returned by it, from the same module.
If you use <a target="_blank" rel="noopener noreferrer" href="https://redux-toolkit.js.org/api/createAsyncThunk">async thunks</a>, they usually need to be used by that same slice, so it&#x27;s not a bad idea to export them also from there.
<h3>Why I love it</h3>
There are a couple of reasons for which I loved RTK right away. The main one is that they make an impressive work on making type inference in TypeScript work properly.
<ul>
<li>Slices detect the type of the initial state in order to know what is expected to be passed as <code>state</code> to reducers.</li>
<li>You can infer the type of your store from what is generated by <code>configureStore</code> or <code>combineReducers</code> functions, allowing you to also properly type the <code>dispatch</code> function.</li>
<li>Action creators are generated with the proper param types based on the reducers provided to <code>createSlice</code> or the callback provided to <code>createAsyncThunk</code>.</li>
<li><code>createSlice</code> can consume all the action creators generated by <code>createAsyncThunk</code>, and infer the proper payload for each one of them.</li>
</ul>
Thanks to this, the code is super predictable and potential bugs can be caught sooner.
<blockquote>
More details on how to use RTK with typescript https://redux-toolkit.js.org/usage/usage-with-typescript
</blockquote>
But this is not the only benefit. Other reasons that make it very nice to work with are:
<ul>
<li>It keeps the same concepts from Redux, but removing the boilerplate or your own helper functions if you have them (yes, I didn&#x27;t want to keep all those <code>switch</code> statements everywhere).</li>
<li>It is mostly transparent for the components dispatching the actions or consuming the state.</li>
<li>All the tests covering your reducers/actions/components should keep passing, as the outcome from RTK is the same as if you built everything manually.</li>
<li>It introduces a couple of extra helpers, like <a target="_blank" rel="noopener noreferrer" href="https://redux-toolkit.js.org/rtk-query/overview">RTK query</a> or the <a target="_blank" rel="noopener noreferrer" href="https://redux-toolkit.js.org/api/createListenerMiddleware">listener middleware</a>, which is very useful to dispatch actions as a result of other actions being dispatched.</li>
</ul>
<h3>Challenges I found</h3>
Of course, adopting a new tool always comes with some challenges.
The main &quot;problem&quot; I found is that RTK is a bit opinionated, but that&#x27;s precisely how they manage to reduce boilerplate code, and it&#x27;s also &quot;advertised&quot; in their website, so not a big surprise.
I usually prefer to be in control of everything (configuration over convention), but that&#x27;s just a personal preference, and sometimes it&#x27;s ok to accept a bit of magic for the sake of simplicity.
This caused the next challenges for my particular case:
<ul>
<li>
RTK assumes your project follows a strict <a target="_blank" rel="noopener noreferrer" href="https://redux.js.org/tutorials/fundamentals/part-7-standard-patterns#flux-standard-actions">Flux pattern</a>, meaning your actions always have the properties <code>type</code>, <code>payload</code> and optionally <code>meta</code> and <code>error</code>.
I was not following this pattern (my actions had the payload &quot;mixed&quot; on the first object level), so I had to adapt all my actions first, fix the tests, then migrate the reducer to RTK and make sure the tests kept passing.
<div></div>
</li>
<li>
Immutability is no longer promoted in reducers. Instead, the library requires <a target="_blank" rel="noopener noreferrer" href="https://immerjs.github.io/immer/">immer</a> internally, and using proxies they automatically ensure immutability, even if you &quot;mutate&quot; the state.
This feels a bit counterintuitive at first, because redux always promoted an immutable approach, and the use of immer is not obvious.
However, they have <a target="_blank" rel="noopener noreferrer" href="https://redux-toolkit.js.org/usage/immer-reducers#why-immer-is-built-in">good reasons</a> to do it, and if your reducers return a state object, you can still use the immutable approach. It&#x27;s up to you.
<div></div>
</li>
<li>
Action creators resulting of <code>createAsyncThunk</code> can only have one argument.
This is because the async thunk callback receives a second argument containing the <code>dispatch</code> and <code>getState</code> functions, as well as a couple of other goodies, enforcing this convention to avoid human errors.
<div></div>
</li>
<li>
You cannot dispatch other actions after the callback passed to <code>createAsyncThunk</code>, because it has to return the payload and RTK dispatches the action for you.
However, this is probably something you usually should not be doing, and in my case, I resolved it with the <a target="_blank" rel="noopener noreferrer" href="https://redux-toolkit.js.org/api/createListenerMiddleware">listener middleware</a>, listening for the second action in order to dispatch the second one.
</li>
</ul>
<h3>Conclusion</h3>
My conclusion is that RTK is definitely worth it if you use redux.
It makes code simpler, more predictable and better typed. However, it still requires knowing how redux works internally and what are its patterns, in order to understand why the library exposes the helpers it exposes and why.
For existing redux projects, you can migrate progressively, addressing the challenges bit by bit and not all at once, which simplifies its adoption.
For new projects in which you want to use redux, I would recommend starting right away with RTK.

A couple of weeks ago I migrated a React project using redux to Redux Toolkit. Redux has always been my preferred state management tool for React (which doesn't mean I "love" it), but one of its main problems is the amount of boilerplate code it requires.

Because of that I had a couple of helper…

My experience migrating to Redux Toolkit (RTK)

In 2019, GitHub published their own solution to run automated workflows called GitHub Actions, which allowed those hosting their code in GitHub, to be able to define and run their CI/CD pipelines in the same platform. When it was released, one of the main pain points to use it was that defining…

How to reduce duplication in your GitHub Actions workflows

Recently ReactJS v18 was released. I started to look into its improvements, and checked which of my projects could benefit from them.
The main react-based project I maintain at the moment of writing this article is <a target="_blank" rel="noopener noreferrer" href="https://github.com/shlinkio/shlink-web-client">shlink-web-client</a>, so I naturally created a new branch and started the process.
The update was quite smooth, as most of the other dependencies I was using from the react ecosystem were already updated to support v18. All but one, <a target="_blank" rel="noopener noreferrer" href="https://enzymejs.github.io/enzyme/">enzyme</a>, the testing framework I was using to unit-test react components.
To be precise, I was able to continue running my tests, but some of them (those using <code>mount</code> instead of <code>shallow</code>) were still rendering components with the &quot;old&quot; React v17 approach, causing them to behave as in that version and a warning being printed in the console.
<h3>First fix attempt</h3>
The way enzyme integrates with the react version you are using is through an adapter library. The last official one supports only React v16, but someone released an <a target="_blank" rel="noopener noreferrer" href="https://github.com/wojtekmaj/enzyme-adapter-react-17">unofficial adapter for v17</a> which, with close to 650k weekly downloads, quickly became the way to go.
I checked if there was one for React 18, but sadly there wasn&#x27;t.
Also, the author of the adapter for v17 posted an article indicating <a target="_blank" rel="noopener noreferrer" href="https://dev.to/wojtekmaj/enzyme-is-dead-now-what-ekl">enzyme is dead</a>, he was not going to release one for v18, and publishing one for v17 was a mistake.
With those facts in mind, I started to face what was probably going to be the moment to migrate to <a target="_blank" rel="noopener noreferrer" href="https://testing-library.com/docs/react-testing-library/intro">react testing library</a>.
<h3>Migration strategy</h3>
I decided to follow a migration strategy similar to the one suggested in the article above, but with a couple of modifications:
<ul>
<li>I update to React 18 anyway, as tests still pass.</li>
<li>Migrate right away all tests using <code>mount</code>, as those are the ones behaving differently.</li>
<li>New tests should be written directly with react testing library (RTL from now on).</li>
<li>When an existing test needs to be changed, consider migrating it to RTL, and do it if there&#x27;s time.</li>
<li>Introduce some PR from time to time just migrating a handful of tests.</li>
<li>Once everything is migrated, get rid of everything related with enzyme.</li>
</ul>
<h3>First contact and impressions</h3>
I was a bit sceptical with RTL, as it was focused more on integration tests, instead of unit tests, which meant I had to completely change the mindset to write tests.
It also didn&#x27;t support shallow rendering, meaning that you will end up sometimes rendering a bigger component tree when testing container-like components, potentially causing side effects that could affect your test.
On the other hand there&#x27;s more and more people stating that <a target="_blank" rel="noopener noreferrer" href="https://kentcdodds.com/blog/why-i-never-use-shallow-rendering">shallow rendering is an anti-pattern</a>, and that&#x27;s probably the reason of enzyme getting &quot;discontinued&quot;.
Also, RTL has become the <a target="_blank" rel="noopener noreferrer" href="https://reactjs.org/docs/testing.html#tools">officially recommended framework</a> to test react components, it has become very popular, and it has even joined a wider group of testing libraries under the <a target="_blank" rel="noopener noreferrer" href="https://testing-library.com/"><code>@testing-library</code></a> organization.
The thing is that I started to use it, and I was quickly surprised with how it felt to work with it, and all the benefits it was bringing to my tests.
<h3>Benefits I have experienced</h3>
Once I started migrating the first tests I noticed some benefits:
<ul>
<li>Changing my mindset was actually easier than I thought. The docs are very well written, so it&#x27;s relatively easy to start using RTL.</li>
<li>My tests got simpler, and required less of those nasty global mocks that jest allows (I&#x27;m looking at you, <a target="_blank" rel="noopener noreferrer" href="https://jestjs.io/es-ES/docs/mock-functions#mocking-modules"><code>jest.mock</code></a>).</li>
<li>Tests were way less coupled with implementation details. I moved from testing the component props, to test what the user would see on the screen.</li>
<li>Related with previous point, I started to write my tests by looking at the screen instead of the component&#x27;s implementation.</li>
<li>Changes in source code led to fewer tests requiring to be changed.</li>
<li>A progressive migration was possible, where some tests still use enzyme and others are already using RTL.</li>
</ul>
And overall, it felt good and I have already dedicated a couple of PRs just to migrate more tests.
<h3>Challenges I faced</h3>
But of course, migrating to a new tool always comes with its own challenges, and this was not an exception. I&#x27;ll explain the main ones I have faced so far and how I tackled them:
<ul>
<li>
RTL lets you do some things in a couple of different ways, and the docs may not be super clear about what&#x27;s the recommended option.
For this I recommend reading <a target="_blank" rel="noopener noreferrer" href="https://acel.me/DCMb1">this article</a> from Kent C. Dodds, RTL&#x27;s author, which explains some common mistakes when using the library.
</li>
<li>
At first, I struggled a bit with asynchronous side effects and state changes in components, which can easily end up printing a warning like <code>Warning: An update to MyComponent inside a test was not wrapped in act</code>.
The warning also explains how you are supposed to solve this by wrapping your code in <code>act()</code>, but this warning comes from React, not RTL, and what it is not telling you is that RTL already wraps all needed operations in <code>act()</code>.
What this error usually mean is that you are not waiting for something to happen. <a target="_blank" rel="noopener noreferrer" href="https://acel.me/jnUdT">This other article</a>, also from Kent C. Dodds, explains everything you need to know about the things that can lead to this warning, and how to solve them.
</li>
<li>
Since you start testing from the user point of view, some use cases can be challenging to test without coupling with implementation details. For example, when you have a component which dynamically renders different images or icons, I used to directly check the component properties.
For this I found the best solution was to use <a target="_blank" rel="noopener noreferrer" href="https://jestjs.io/docs/snapshot-testing">jest snapshots</a>. I didn&#x27;t want to manually check which SVG was being rendered in the DOM, just that different ones were being used in every case.
<pre class="language-tsx"><code class="language-tsx">// I went from this...
test(&#x27;...&#x27;, () =&gt; {
 const wrapper = shallow(&lt;MyComp dynamicValue=&quot;foo&quot; /&gt;);
 expect(wrapper.find(FontAwesomeIcon).prop(&#x27;icon&#x27;)).toEqual(someIcon);
})

// To this
test(&#x27;...&#x27;, () =&gt; {
 const { container } = render(&lt;MyComp dynamicValue=&quot;foo&quot; /&gt;);
 expect(container.firstChild).toMatchSnapshot();
})
</code></pre>
</li>
<li>
Something very similar happens if you use some library that renders canvas. There&#x27;s no way to test that from a DOM point of view, which is what RTL does.
In my case I was using <a target="_blank" rel="noopener noreferrer" href="https://www.chartjs.org/">Chart.js</a> to render some charts, so I used <a target="_blank" rel="noopener noreferrer" href="https://github.com/hustcc/jest-canvas-mock">jest-canvas-mock</a>, which exposes a method on the canvas to see which are all the events invoked from the library.
Then I used snapshots again to verify they had the expected &quot;shape&quot;.
<pre class="language-tsx"><code class="language-tsx">// I went from this...
test(&#x27;...&#x27;, () =&gt; {
 const wrapper = shallow(&lt;MyCompWithCanvas /&gt;);
 expect(wrapper.find(Chart).prop(&#x27;data&#x27;).datasets).toEqual(...);
})

// To this
test(&#x27;...&#x27;, () =&gt; {
 const { container } = render(&lt;MyCompWithCanvas /&gt;);
 const events = container.querySelector(&#x27;canvas&#x27;)?.getContext(&#x27;2d&#x27;)?.__getEvents();

 expect(events).toBeTruthy();
 expect(events).toMatchSnapshot();
})
</code></pre>
</li>
</ul>
<h3>Conclusion</h3>
For better or worst, enzyme should no longer be considered an option. If you are starting a new project or are using it on small ones, consider migrating to RTL as soon as possible.
If you are using it in bigger projects, following the process described in this article will probably help you.
In any case, you will get surprised. React testing library is a great tool and provides many benefits over enzyme.
Its main issue is that it&#x27;s not a drop-in replacement for enzyme, so you will have to change a bit your mindset.

Recently ReactJS v18 was released. I started to look into its improvements, and checked which of my projects could benefit from them. The main react-based project I maintain at the moment of writing this article is shlink-web-client, so I naturally created a new branch and started the process.

The…

My experience migrating from enzyme to react testing library

Capturing code coverage for a test suite is a very useful way to know which parts of your source code are actually getting executed by tests. This is useful not only to know if you need to add more tests to your project (in case the coverage is too low), but also, if you want to apply technics like…

Capturing remote code coverage in API tests with PHPUnit

In the last couple of days I have been playing around with the idea of combining GitHub Actions and GitHub Pages to be able to dynamically build preview environments every time a pull request is opened against a repository hosting a client-side web app.
The idea is that you can quickly preview the changes made by some contributor, without having to clone their branch and running everything locally, something specially useful for open source projects.
In this article I will explain how to set everything up, and what are some considerations and problems you may face during the process.
<h3>Enable github pages</h3>
The first thing you need to do is enable your project to publish pages on some branch. I recommend creating a new branch for this, let&#x27;s say <code>preview-env</code>, and then enabling publishing on it.
In order to do that go to your repository and then Settings -&gt; Pages, look for the Source section, and you should see something like this:
<img src="/assets/img/preview-envs/github-pages-before-enabling.png" alt="GitHub pages before enabling"/>
Click in the dropdown and select your new branch, <code>preview-env</code>. Then click Save.
<img src="/assets/img/preview-envs/github-pages-after-enabling.png" alt="GitHub pages after enabling"/>
After saving you will see a message at the top of this section telling you what&#x27;s the URL in which the pages are published. It always follows the same pattern. If your organization is <code>my-org</code> and your repository is <code>my-repo</code>, the url will be <code>https://my-org.github.io/my-repo/</code>.
<h3>Configure the workflow</h3>
Now we need to configure a workflow that will &quot;deploy&quot; the preview environment for the branch every time a new pull request is created or updated:
Let&#x27;s imagine we have a React web app that is built with node.js. The configuration could look like this.
<pre class="language-yml"><code class="language-yml"># .github/workflows/deploy-preview.yml
name: Deploy preview

on:
 pull_request: null

jobs:
 deploy:
 runs-on: ubuntu-20.04
 continue-on-error: true
 steps:
 - name: Checkout code
 uses: actions/checkout@v2
 - name: Use node.js 14.15
 uses: actions/setup-node@v1
 with:
 node-version: 14.15
 - name: Generate slug
 id: generate_slug
 run: echo &quot;##[set-output name=slug;]$(echo ${GITHUB_HEAD_REF#refs/heads/} | sed -r &#x27;s/[~\^]+//g&#x27; | sed -r &#x27;s/[^a-zA-Z0-9]+/-/g&#x27; | sed -r &#x27;s/^-+\|-+$//g&#x27; | tr A-Z a-z)&quot;
 - name: Build
 run: npm ci &amp;&amp; npm run build
 - name: Deploy
 uses: JamesIves/github-pages-deploy-action@4.1.1
 with:
 branch: preview-env
 folder: build
 target-folder: ${{ steps.generate_slug.outputs.slug }}
 - name: Publish env
 uses: marocchino/sticky-pull-request-comment@v2
 with:
 header: Preview environment
 message: |
 ## Preview environment
 https://my-org.github.io/my-repo/${{ steps.generate_slug.outputs.slug }}/
</code></pre>
<blockquote>
If you are not familiar with GitHub Actions yet, take a look at the <a target="_blank" rel="noopener noreferrer" href="https://docs.github.com/en/actions">official documentation</a>.
</blockquote>
This workflow will follow the next steps:
<ul>
<li>Checkout the code from the repository.</li>
<li>Set-up node.js 14.15</li>
<li>Based on the branch name, generate a slug that we will use to have separated deployment folders for every pull request we receive. The slug is exposed to next steps as an output.</li>
<li>Build the project. This step will obviously depend on the needs of your project.</li>
<li>Using <a target="_blank" rel="noopener noreferrer" href="https://github.com/JamesIves/github-pages-deploy-action">JamesIves/github-pages-deploy-action</a>, deploy the contents we previously built in the GitHub page, under a sub-folder matching the slug we generated a few steps above, and under the branch <code>preview-env</code> (remember we enabled GitHub pages for this branch).</li>
<li>Finally, using <a target="_blank" rel="noopener noreferrer" href="https://github.com/marocchino/sticky-pull-request-comment">marocchino/sticky-pull-request-comment</a>, post a comment in the pull request with a link to the preview environment, so that you can just click on it and see the deployment.</li>
</ul>
The steps may change depending on your needs and how you build your project. For example, I have an extra step to update a property in the <code>package.json</code> that is then used to allow serving the app in a sub-path instead of the root of the domain.
You may also not build your site with node.js, as you could be using any of the multiple existing static site generators out there, but you get the idea.
Finally, you have probably noticed the config includes <code>continue-on-error: true</code>. This will allow the pull request to be merged even if this process fails for some reason.
<h3>Considerations when using forks</h3>
While the configuration above will work perfectly when creating pull requests inside the same repository, there are some limitations when the pull request comes from a fork (which is the usual in OSS projects).
For security reasons, the <code>GITHUB_TOKEN</code> injected in those cases doesn&#x27;t have write permissions, which will make the &quot;deploy&quot; step fail.
Also, you cannot define your own token with more permissions and inject it as a secret, as the secrets are not exposed to workflows running on forks.
However, with a couple small config changes, we can work around these limitations:
<h4>Change event</h4>
The first thing we need to do is use the <code>pull_request_target</code> event instead of the <code>pull_request</code> one.
<pre class="language-diff"><code class="language-diff">on:
- pull_request: null
+ pull_request_target: null
</code></pre>
According to the documentation, this is equivalent to the <code>pull_request</code> event, but it runs in the scope of the base repository.
Thanks to this, the <code>GITHUB_TOKEN</code> will come with write permissions, and we will have access to secrets if needed.
<h4>Checkout the fork</h4>
The second thing we are going to change is that we now need to specify the repository to checkout.
In order to do that, we need to add two inputs to the &quot;Checkout code&quot; step, like this:
<pre class="language-diff"><code class="language-diff"> steps:
 - name: Checkout code
 uses: actions/checkout@v2
+ with:
+ repository: ${{ github.event.pull_request.head.repo.full_name }}
+ ref: ${{ github.event.pull_request.head.ref }}
</code></pre>
This will ensure we checkout from the fork repository, instead of the base one, which is the default, as it&#x27;s the one where the <code>pull_request_target</code> event is scoped to.
And that&#x27;s it. With these two changes you can now deploy preview environments for pull requests coming from forks.
<blockquote>
IMPORTANT! The limitations of the <code>pull_request</code> event are there for a reason. People could potentially get access to your secrets or manipulate your code. Proceed at your own risk.
</blockquote>
<h3>Other considerations</h3>
Other than the limitations for forks, there are a few other minor things to take into consideration.
<ul>
<li>Changes in GitHub Pages do not reflect immediately. It may take up to a couple of minutes for the deployment to be available, but considering you will usually not review a PR right after it has been created, this should not be a big issue.</li>
<li>As you have seen, every time a pull request is created, the deployment will generate a new folder in your GitHub Pages branch. This may result in a lot of old folders to pile-up, so you may want to set up some way to clean them up periodically or after the pull request has been merged.</li>
<li>Due to the fact that GitHub Pages only allows static pages to be served, this process cannot be used with apps that require some kind of back-end or server-side logic.</li>
<li>If your project is already using GitHub Pages for something else (serve the project&#x27;s website or docs), that may conflict with this, preventing you from having this kind of preview envs.</li>
</ul>

In the last couple of days I have been playing around with the idea of combining GitHub Actions and GitHub Pages to be able to dynamically build preview environments every time a pull request is opened against a repository hosting a client-side web app. The idea is that you can quickly preview the…

Setting-up a preview environment for your client-side apps with GitHub Actions and GitHub Pages

Asynchronous and non-blocking runtimes are pretty usual in many programming languages, as well as long-lived web apps that stay in memory and are capable of dispatching multiple HTTP requests without having to be fully bootstrapped every time. This has not been traditionally the case with PHP apps.
However, many projects are starting to get some adoption, that bring this long-lived approach to the PHP world.
The problem with this is that PHP developers are not used to this, and they tend to continue acting as they have always done. Also, libraries might have made some assumptions, making them not work as expected when used under this context.
Some of these projects are <a target="_blank" rel="noopener noreferrer" href="https://www.swoole.co.uk">swoole</a>, <a target="_blank" rel="noopener noreferrer" href="https://reactphp.org/">ReactPHP</a> or <a target="_blank" rel="noopener noreferrer" href="https://amphp.org/">Amphp</a>, to give some examples.
In this article, I will be focusing on the first one, and explain how to approach some of the pain points I have found after using it for a year and a half on an <a target="_blank" rel="noopener noreferrer" href="https://github.com/shlinkio/shlink">existing PHP project</a>.
<h3>Introducing Swoole</h3>
The main difference swoole has compared to other similar projects is that it&#x27;s written as a native PHP extension, and needs to be installed with <code>pecl</code>.
This is obviously a bit cumbersome, but the benefit is that, since it&#x27;s written in C++, it has a better handling of memory allocation, which is very important to achieve what it tries to do.
What swoole does is running a main process which bootstraps the PHP application once, and then keeps it in memory so that it can continue dispatching requests without having to load resources every time. This makes it really fast.
Those HTTP requests are handled by a set of child processes called &quot;web workers&quot;. Swoole forks the contents in memory for the main process to each one of the workers, so each worker has a &quot;copy&quot; of the application in memory.
Each worker is independent, but each one of them can only handle one request at a time, so you have to increase the number of workers if you expect a lot of concurrent traffic.
Other than web workers, swoole also has another set of child processes called &quot;task workers&quot;, which can be used to delegate long tasks at runtime to prevent web workers from getting blocked.
Here you can see a diagram of swoole&#x27;s architecture.
<img src="https://www.swoole.co.uk/images/how-swoole-works.png" alt="How Swoole works"/>
<blockquote>
More information can be found in <a target="_blank" rel="noopener noreferrer" href="https://www.swoole.co.uk/how-it-works">Swoole&#x27;s website</a>.
</blockquote>
<h3>Problems to solve</h3>
Now that we know how swoole works, let&#x27;s see some of the challanges I have faced in th past, which are usually not that ovbious when you are used to think in terms of &quot;blocking&quot; and &quot;short-lived&quot;.
<h4>Database connections are kept open</h4>
One of the first things you find out is that, since the application stays in memory, any database connection that has been opened by a worker (either web or task worker), will be kept open and reused in subsequent executions or HTTP requests served by the app.
This, which is in theory good, has a side effect, which is not obvious during development: the opened connection will eventually expire.
Since most of the popular database abstraction layers in PHP are coming from the times in which PHP was just a bootstrap-on-every-request technology, they never cared too much about expired database connections, as they would be opened again on next request.
However, when using swoole and the like, you have to implement some mechanism which takes care of reconnecting or gracefully recovering when this happens.
<blockquote>
I wrote an article explaining how I solved this on a middleware-based app served with swoole and using doctrine: <a href="/2019/11/04/how-to-properly-handle-a-doctrine-entity-manager-on-an-expressive-application-served-with-swoole/">How to properly handle a doctrine entity manager on an expressive application served with swoole</a>
</blockquote>
Obviously, swoole (and also others) provides some tools to work with databases (like this <a target="_blank" rel="noopener noreferrer" href="https://www.swoole.co.uk/docs/modules/swoole-async-mysql-client">MySQL client</a>), which I&#x27;m sure do not have this problem (although, I haven&#x27;t personally tried this), but they are more limited than other well adopted libraries.
<h4>In-memory caches don&#x27;t behave as expected</h4>
It is a frequent practice to use local in-memory caches, like APCu, to improve applications performance in production.
Usually APCu is faster than other options, like redis or memcached, so it&#x27;s the choice when you only have one instance of your app running (meaning, you don&#x27;t have a cluster behind a load balancer).
However, even if you are running your app on just one server with swoole, because of the web worker system explained earlier, you actually have several copies of your app in memory, and each one of them will have a separated and not shared cache when using APCu.
This can lead to inconsistencies which are hard to debug, so you need to have it in mind.
Also, since the information that is being loaded is kept in memory anyway, you will not notice such a big improvement by using APCu or other in-memory caches.
<blockquote>
Notice that this problem does not happen when using distributed/centralized caching solutions, such as redis or memcached, as in that case, the caching technology is shared by all workers.
</blockquote>
<h4>No need to optimize file including/requiring</h4>
Many projects have complex configuration systems which are spread into several files. While this make it more maintainable, it also has a performance impact when all those files need to be loaded on every request.
Because of that, there are libraries like <a target="_blank" rel="noopener noreferrer" href="https://github.com/laminas/laminas-config-aggregator">laminas-config-aggregator</a> that can dynamically load config from many sources and, if some conditions are met, they can merge the result in a single bigger file that can be used in later requests.
In the case of swoole, as all the files will be kept in memory once loaded for the first time, there&#x27;s no need for this kind of optimization.
<h4>Class autoloading happens once</h4>
This point is very related with the previous one.
Composer offers a lot of optimization options so that class autoloading is faster in production. While you usually will still want to use some of those so that the first time a class is being loaded it is still reasonably fast, any subsequent hits will just use the class which is already in memory.
One of the composer flags you will probably don&#x27;t want use with swoole is <code>--apcu-autoloader</code>, because of the reasons mentioned two points above this one.
<h4>Services should be truly stateless</h4>
This is something that you should always do anyway, but it&#x27;s specially important when a service instance is kept in memory between HTTP requests.
Sometimes it&#x27;s easy to end up having some small piece of internal state on a service, where we save something as a side effect of a method call, to end up using it when another one of the public methods is invoked.
While this could seem harmless when the whole app is bootstrapped on every request, it can have very dangerous side effects when the app is served with swoole, because the instance will persist until another request is dispatched by the same web worker.
Just imagine you saved some user data on a service, and it ends up being served to another user on the next request. That won&#x27;t look good.
Because of this, make sure your services are truly stateless, and if for any reason that&#x27;s not possible, at least remember to have some mechanism in place that flushes the data just before finishing the request.
<h4>Hot reloading for dev envs is not that good</h4>
One thing we usually take for granted is that we can modify a PHP file and just by making a new request, the new code will be executed. However, when the code is kept in memory, it&#x27;s not that straightforward.
When you are developing a project which is served with swoole, you will have to restart the server every time to get your changes applied.
Swoole has a mechanism to do a light server reload, which is faster. You can find the docs in <a target="_blank" rel="noopener noreferrer" href="https://www.swoole.co.uk/docs/modules/swoole-server-reload">it&#x27;s website</a>.
The problem with this is that you need to implement the mechanism to invoke that when a file is changed, or use some library that does it for you.
Also, I have noticed that even calling that and seeing that the server is reloaded, does not always have the expected effect, and you have to still end up doing a hard reload. It&#x27;s something I don&#x27;t fully understand and I still need to investigate a bit further.
<h3>Conclusion</h3>
As it always happens when new technologies appear, it takes some time to get used to them, but I strongly suggest you to give a try to swoole or any other non-blocking runtime, because it makes your apps super fast.
Just remember to have all of the above in mind ;-)
That said, here you can find some libraries that will make your life easier when trying to serve existing projects with swoole, without having to couple with it and being able to continue using your favourite framework:
<ul>
<li>IDE helper: <a target="_blank" rel="noopener noreferrer" href="https://github.com/swoft-cloud/swoole-ide-helper">https://github.com/swoft-cloud/swoole-ide-helper</a></li>
<li>Mezzio (formerly Zend Expressive): <a target="_blank" rel="noopener noreferrer" href="https://github.com/mezzio/mezzio-swoole">https://github.com/mezzio/mezzio-swoole</a></li>
<li>Symfony: <a target="_blank" rel="noopener noreferrer" href="https://github.com/k911/swoole-bundle">https://github.com/k911/swoole-bundle</a></li>
<li>Laravel: <a target="_blank" rel="noopener noreferrer" href="https://github.com/swooletw/laravel-swoole">https://github.com/swooletw/laravel-swoole</a></li>
</ul>
Also, here you can find some more articles talking about this topic:
<ul>
<li><a target="_blank" rel="noopener noreferrer" href="https://www.zend.com/blog/why-you-should-use-asynchronous-php">https://www.zend.com/blog/why-you-should-use-asynchronous-php</a></li>
<li><a target="_blank" rel="noopener noreferrer" href="https://samsonasik.wordpress.com/2020/04/05/using-swoole-in-mezzio-application-with-sdebug/">https://samsonasik.wordpress.com/2020/04/05/using-swoole-in-mezzio-application-with-sdebug/</a></li>
</ul>

Asynchronous and non-blocking runtimes are pretty usual in many programming languages, as well as long-lived web apps that stay in memory and are capable of dispatching multiple HTTP requests without having to be fully bootstrapped every time. This has not been traditionally the case with PHP apps.…

Considerations when working with async PHP runtimes like swoole

Some of you probably know that I have a pet project in which I like to work from time to time. This project is a self-hosted URL shortener called <a target="_blank" rel="noopener noreferrer" href="https://shlink.io">Shlink</a>.
Shlink is built using expressive as the base framework for the HTTP-dispatching task. A while ago, an expressive module was released to officially support serving expressive apps with <a target="_blank" rel="noopener noreferrer" href="https://www.swoole.co.uk/">swoole</a>, and I decided to include it on Shlink.
<h3>How does swoole work?</h3>
Swoole is an asynchronous non-blocking I/O framework which works in a similar way as node.js, but for PHP apps.
It has been conceived in a way that the app stays in memory between requests, removing the bootstrapping step and making apps to be served much faster.
Swoole can serve as many simultaneous requests as you tell it to. For this it makes use of workers. Each worker can serve one request at a time, and every worker has a separate instance of the app loaded in memory (this is something I <a target="_blank" rel="noopener noreferrer" href="https://twitter.com/dominikzogg/status/1189403552399134720">learned recently</a>).
Because of this, you have to sometimes change a bit the way you design your code. For example, if a service needs to be stateful (it shouldn&#x27;t, but shit happens), you will have to remember that the state will persist between requests, and you will want to somehow reset it to avoid unexpected side effects.
<h3>How does this affect the EntityManager?</h3>
One of the main services that keeps this internal state in Shlink is the doctrine <a target="_blank" rel="noopener noreferrer" href="https://www.doctrine-project.org/projects/doctrine-orm/en/current/tutorials/getting-started.html#obtaining-the-entitymanager">EntityManager</a>.
Due to the fact that it was originally designed to be used on a classical web server + CGI context, assuming it would be recreated on every request, there are a few considerations to take into account:
<ul>
<li>It implements the unit of work pattern. It internally keeps track of all the entities that have been created/changed until you decide to flush it.</li>
<li>If something fails, the <code>EntityManager</code> can get closed, which makes it unusable after that moment.</li>
<li>If the database connection expires, the <code>EntityManager</code> will not gracefully recover. Instead, it will fail, since it&#x27;s not aware of the connection not being usable, getting yourself in previous point.</li>
</ul>
<h3>How to workaround it</h3>
I learned about these issues by try and error, by using it on the app and getting the failures. Now, I want to share how I workaround those issues.
The first one I faced was the one making the app unusable after the <code>EntityManager</code> was closed. In order to solve this I created a decorated instance (naming, an object implementing <code>EntityManagerInterface</code> which in turn was wrapping an actual <code>EntityManager</code>), which was making sure the wrapped instance was transparently recreated if at some point it was closed.
However, this caused some side effects, and I finally realized that it was better to make the <code>EntityManager</code> to have to be reopened from the outside, and only once per request (this is closer to how it would behave on a classical set-up).
That&#x27;s how I came up with the <code>ReopeningEntityManager</code> Shlink is currently using (as of v1.20.0):
<pre class="language-php"><code class="language-php">&lt;?php

declare(strict_types=1);

namespace Shlinkio\Shlink\Common\Doctrine;

use Doctrine\ORM\Decorator\EntityManagerDecorator;

class ReopeningEntityManager extends EntityManagerDecorator
{
 /** @var callable */
 private $createEm;

 public function __construct(callable $createEm)
 {
 parent::__construct($createEm());
 $this-&gt;createEm = $createEm;
 }

 public function open(): void
 {
 if (! $this-&gt;wrapped-&gt;isOpen()) {
 $this-&gt;wrapped = ($this-&gt;createEm)();
 }
 }
}
</code></pre>
<blockquote>
You can find the actual code in <a target="_blank" rel="noopener noreferrer" href="https://github.com/shlinkio/shlink-common/blob/master/src/Doctrine/ReopeningEntityManager.php">shlink-common</a>.
</blockquote>
It just expects a factory that creates the actual <code>EntityManager</code>, and exposes an <code>open</code> public method which re-creates the wrapped instance if closed.
This method is called from a <a target="_blank" rel="noopener noreferrer" href="https://www.php-fig.org/psr/psr-15/">PSR-15</a> middleware which needs to be registered early in the middleware pipeline. It makes sure the method is called before the next middleware on the stack is invoked:
<pre class="language-php"><code class="language-php">&lt;?php

declare(strict_types=1);

namespace Shlinkio\Shlink\Common\Middleware;

use Psr\Http\Message\ResponseInterface;
use Psr\Http\Message\ServerRequestInterface;
use Psr\Http\Server\MiddlewareInterface;
use Psr\Http\Server\RequestHandlerInterface;
use Shlinkio\Shlink\Common\Doctrine\ReopeningEntityManager;

class CloseDbConnectionMiddleware implements MiddlewareInterface
{
 /** @var ReopeningEntityManager */
 private $em;

 public function __construct(ReopeningEntityManager $em)
 {
 $this-&gt;em = $em;
 }

 public function process(ServerRequestInterface $request, RequestHandlerInterface $handler): ResponseInterface
 {
 $this-&gt;em-&gt;open();

 try {
 return $handler-&gt;handle($request);
 } finally {
 $this-&gt;em-&gt;getConnection()-&gt;close();
 $this-&gt;em-&gt;clear();
 }
 }
}
</code></pre>
<blockquote>
Again, the code can be found in <a target="_blank" rel="noopener noreferrer" href="https://github.com/shlinkio/shlink-common/blob/master/src/Middleware/CloseDbConnectionMiddleware.php">shlink-common</a>
</blockquote>
This middleware also does two more things (this time, after calling the next middleware on the stack, and making sure it does it even if an exception is thrown).
<ul>
<li>It first closes the database connection. By doing it so, we make sure the <code>EntityManager</code> will reconnect the next time it needs to interact with the database, and therefore, we avoid the &quot;expired connection&quot; issue.</li>
<li>Then it clears the <code>EntityManager</code> itself. This avoids memory leaks, and also prevents that any orphan/non-flushed entity &quot;transcends&quot; to the next request, solving the first problem.</li>
</ul>
<h3>Further considerations</h3>
This approach has been working well so far, but it&#x27;s important to assume that the <code>EntityManager</code> was not really designed for this non-blocking kind of context.
You need to know that the worker making use of it will not be able to serve any request until the <code>EntityManager</code> finishes its job. Because of this you might want to increase the number of swoole workers.
Also, closing the DB connection on every request is not the most optimal approach. It would be perfect to have a connection pool that was able to recreate connections when expired, and to keep them open as long as possible for faster operations. However, the <code>EntityManager</code> was not designed for that, so it&#x27;s better to close it.
You also need a <code>callable</code> to pass to the <code>ReopeningEntityManager</code>. The way you solve this depends on your implementation and the dependencies you use.
In my case, in Shlink I make use of the <code>ServiceManager</code>, so I have a delegator factory which takes the actual factory and passes it to the <code>ReopeningEntityManager</code>:
<pre class="language-php"><code class="language-php">&lt;?php

declare(strict_types=1);

namespace Shlinkio\Shlink\Common\Doctrine;

use Psr\Container\ContainerInterface;

class ReopeningEntityManagerDelegator
{
 public function __invoke(ContainerInterface $container, string $name, callable $createEm): ReopeningEntityManager
 {
 return new ReopeningEntityManager($createEm);
 }
}
</code></pre>
<blockquote>
You can find the actual code in <a target="_blank" rel="noopener noreferrer" href="https://github.com/shlinkio/shlink-common/blob/master/src/Doctrine/ReopeningEntityManagerDelegator.php">shlink-common</a>.
</blockquote>
<h3>Conclusion</h3>
I wanted to write this article as a recap for myself too, since I had to ask and investigate many times, and I couldn&#x27;t find a single source of truth explaining everything.
I hope that, if you ended up here, this article served you well.

Some of you probably know that I have a pet project in which I like to work from time to time. This project is a self-hosted URL shortener called Shlink. Shlink is built using expressive as the base framework for the HTTP-dispatching task. A while ago, an expressive module was released to…

Alejandro Celaya | Blog

Software development, agile methodologies and open source software.

tag Archive — "swoole"

— 2020 —

— 2019 —