<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Sebastian Thoss on Medium]]></title>
        <description><![CDATA[Stories by Sebastian Thoss on Medium]]></description>
        <link>https://medium.com/@dj_thossi?source=rss-efbce6210af8------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*VjKyZNfEuAKZP54EwjUhsQ.jpeg</url>
            <title>Stories by Sebastian Thoss on Medium</title>
            <link>https://medium.com/@dj_thossi?source=rss-efbce6210af8------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Thu, 19 Nov 2026 17:43:36 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@dj_thossi/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[SmokeTest in development]]></title>
            <link>https://stories.celebrate.company/smoketest-in-development-ff23159dfe25?source=rss-efbce6210af8------2</link>
            <guid isPermaLink="false">https://medium.com/p/ff23159dfe25</guid>
            <category><![CDATA[seo]]></category>
            <category><![CDATA[development]]></category>
            <category><![CDATA[tech]]></category>
            <dc:creator><![CDATA[Sebastian Thoss]]></dc:creator>
            <pubDate>Tue, 14 Nov 2017 15:54:22 GMT</pubDate>
            <atom:updated>2018-08-22T14:16:03.857Z</atom:updated>
            <content:encoded><![CDATA[<p>Today I want to tell you about what SmokeTests in software development are, and how they help us at <a href="https://www.kartenmacherei.de/">kartenmacherei</a> to identify problems early.</p><h4>What we use SmokeTests for.</h4><p>We realized that after deploying <a href="https://tech.kartenmacherei.de/dont-fear-the-walking-dead-6baca1a96f6f">Fury</a> we’d need to ensure that we were still able serve all of our URLs. We wanted to not just test if the HTML was stored as a key-value pair, but that also the customer facing URLs were working. We decided on SmokeTests to verify over 20,000 URLs, checking against the following criteria:</p><ul><li>Status Code is 200</li><li>Pages are served extremely fast</li><li>Content is served from the correct machine</li><li>In the case of a redirect, that it redirects to the same machine</li></ul><p>In addition we tested not just URLs shipped from Fury, but also from our legacy shop system. We decided that this is necessary because of our <a href="https://tech.kartenmacherei.de/dont-fear-the-walking-dead-6baca1a96f6f">special Request handling</a> we’ve set up.</p><p>We introduce SmokeTests as the last step during our <a href="https://tech.kartenmacherei.de/how-to-deploy-a-black-horse-2f81ed65cd47">deployment</a>. We only activate the freshly deployed server when ALL our smoke tested URLs are responding correctly.</p><h4>Where SmokeTests actually come from.</h4><p>Smoke testing is a common technique to find holes in drainage and piping. In order to find a leak, you block all exits in the area (e.g. with sandbags). Than you blow a lot of smoke into the pipe at one entrance. After a while the smoke will find its way through the pipe and out at the leak. Now you know where you have to search in order to get to the leak. We converted this principle to be applicable for kartenmacherei’s website.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*O_WFXboV47xWz8ELiL64VQ.jpeg" /></figure><h4>Which problems become visible through SmokeTests.</h4><p>I just checked the numbers. We’ve run SmokeTests more than 3600 times and tested more than 73,1 million URLs since we launched Fury. Yes it fails sometimes, but this is mostly due to 3 main reasons:</p><ol><li>Someone forgot to redirect old URLs after renaming a product. This is actually the most common case.</li><li>When we worked on redirects, we learned that we should have made sure to test that the redirect URL stays in the same Domain. Otherwise this can cause strange side effects and random false positives.</li><li>Rarely some DNS lookup fails for no reason. At least that is what the output tells us. We never managed to properly reproduce it. Increasing the DNS cache didn’t solve it either.</li></ol><p>Testing the missing redirects after renaming products is an additional benefit of SmokeTests. We realised afterwards that with SmokeTests we can even be sure that very old URLs stay working, because our customers bookmark a lot of links to our website and share them with their family and friends.</p><h4>What’s the biggest disadvantage of our SmokeTests.</h4><p>We maintain our list of URLs manually. This is by design, otherwise there would be a high chance that the test would erroneously report “green” for pages that don’t exist. Updating this list is a quite a pain point because we just want to add additional URLs. We haven’t yet updated the list. In addition, our current list of URLs don’t have any additional GET parameters. So we’re not testing if there are possible side effects when using parameters — like for pre selecting format or color.</p><h4>How do SmokeTests work.</h4><p>In my next post I want to go with you into more detail about how SmokeTests work and what we had to do to make them fast, too.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ff23159dfe25" width="1" height="1" alt=""><hr><p><a href="https://stories.celebrate.company/smoketest-in-development-ff23159dfe25">SmokeTest in development</a> was originally published in <a href="https://stories.celebrate.company/">celebrate company</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to deploy a black horse]]></title>
            <link>https://stories.celebrate.company/how-to-deploy-a-black-horse-2f81ed65cd47?source=rss-efbce6210af8------2</link>
            <guid isPermaLink="false">https://medium.com/p/2f81ed65cd47</guid>
            <category><![CDATA[continuous-integration]]></category>
            <category><![CDATA[pipeline]]></category>
            <category><![CDATA[tech]]></category>
            <category><![CDATA[deployment]]></category>
            <dc:creator><![CDATA[Sebastian Thoss]]></dc:creator>
            <pubDate>Wed, 16 Nov 2016 14:19:58 GMT</pubDate>
            <atom:updated>2017-04-21T14:49:26.356Z</atom:updated>
            <content:encoded><![CDATA[<p>In my last post “<a href="https://tech.kartenmacherei.de/dont-fear-the-walking-dead-6baca1a96f6f#.q9fnukz72">Don’t fear the walking dead</a>”, I showed how we fought off our zombie by introducing a black horse named Fury. This time I’d like to tell you more about how we deploy this new component.</p><p>When we first thought about how we want to deploy this piece of code, it looked to be a difficult task because there are so many possible ways out there. First we set up some primary goals in order to narrow down the list of possible solutions.</p><ul><li>We want to deploy often</li><li>Deployment should be possible at any day, any time</li><li>A deployment should be fast</li><li>Deployments shouldn’t effect each other</li><li>The deployment process should be simple and easily understandable</li></ul><p>After having decided on these goals, we discussed several possible solutions. In the end we decided to go for an A/B deployment.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*KI7EyF8TFcXBzHQigKOWOA.png" /></figure><p>To get an A/B deployment set up and running it’s necessary to have 3 servers: 1 to route all requests and 2 servers to run the application on.</p><p>1st web-server (router) get’s all requests from the web and is proxying them through to only one of the application servers. This server is called the “active box” and can be either Server A or B. Both should be identical. Only one of them can be active, of which the active one can easily be changed in our top level web-server by running a single command. Let’s assume Server B is our current “active box”. This results in server A being idle. Now we can deploy our latest changes onto Server A without affecting “active box”.</p><p>To deploy now on server A we decided to assume this server in an unknown state. This will make each deployment step easier and more understandable. We came up with 4 steps we need to do in order to deploy Fury.</p><h3>1. Deploy code to server A</h3><p>This step makes sure that all source code is in the correct place. All environment parameters are set correctly and needed directories are linked.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*BNjKsaPCqHXnUrrMzW838A.png" /></figure><h3>2. Collect &amp; Export</h3><p>In order to fill up key-value storage (K/V) and search engine we now need to run the “Collect &amp; Export” script on server A. As described in our previous post, Collect &amp; Export will collect all needed information from a read slave of our legacy system’s database. Before exporting HTML and JSON into K/V and search engine we clear them both to make sure, we don’t have any artefacts left which could affect the application.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Mc5ChP5FXMWjq87lzU3rMg.png" /></figure><h3>3. Smoke Tests</h3><p>This step will simulate several thousand requests to server A to make sure that the most important pages are working correctly and fast.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*va1UXrQOsuvFrGdI8-92yw.png" /></figure><h3>4. Switch to A</h3><p>Last but not least, if all previous steps finished successfully and there was no problem, we tell the top level web-server (router) that server A is now active and all requests should be routed to this server.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7I_k5eVVui0NtMyOjLS9EA.png" /></figure><p>If one of the previous described steps fails, nothing happens. The deployment process just stops. With the next deployment, everything will start from scratch. This whole deployment process needs about 15 minutes to finish. It’s so fast that we’re able to deploy several times per hour. Even if there are no changes at all.</p><p>After setting up this deployment architecture and running it for several months, I can tell I still love it. It is extremely reliable and strict. It will even fail if someone renamed and forgot to put a redirect in place for one of the pages covered by Smoke Tests. This is the usual reason for breaking a deployment. The only downside I can think of is that the output of our deployment process is too technical. Non IT colleagues might be overwhelmed with the mass of output and it is difficult for them to find which URLs broke during testing. Something we could work on in future.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2f81ed65cd47" width="1" height="1" alt=""><hr><p><a href="https://stories.celebrate.company/how-to-deploy-a-black-horse-2f81ed65cd47">How to deploy a black horse</a> was originally published in <a href="https://stories.celebrate.company/">celebrate company</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Don’t fear the walking dead.]]></title>
            <link>https://stories.celebrate.company/dont-fear-the-walking-dead-6baca1a96f6f?source=rss-efbce6210af8------2</link>
            <guid isPermaLink="false">https://medium.com/p/6baca1a96f6f</guid>
            <category><![CDATA[php]]></category>
            <category><![CDATA[tech]]></category>
            <category><![CDATA[ecommerce-web-development]]></category>
            <category><![CDATA[unit-testing]]></category>
            <category><![CDATA[code-coverage]]></category>
            <dc:creator><![CDATA[Sebastian Thoss]]></dc:creator>
            <pubDate>Mon, 24 Oct 2016 06:57:29 GMT</pubDate>
            <atom:updated>2017-02-24T14:35:54.658Z</atom:updated>
            <content:encoded><![CDATA[<p>Legacy software can be like a zombie: somehow it still works, but nobody would consider it live and well, and the thought of having to touch it makes you want to run away. So what can you do to get rid of it? In this post we’ll show you how we replaced the most important pages of our monolithic legacy system with a black horse called fury.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*rHxC5o-xzJ40WbNklncACA.png" /></figure><p>But let’s start at the beginning. Kartenmacherei started 6 years ago with a standard e-commerce platform, to which we added a lot of custom code over the years. With each new custom feature we used the core functionalities less and less of. After doing that for some years we slowly but surely got into serious performance issues. On top of that, we developers didn’t feel comfortable to even touch the code because of possible unknown side effects.</p><h4>“Our legacy shop turned into a zombie …</h4><p>… and most developers were scared to touch it. At some stage we decided it was time to get rid of our zombie. The question was how we were going to do this, and whether we would have to replace everything at once. We decided that the first step would be to replace the product detail pages and category pages only.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*x0-PKLCt52WHdixB1x2Xsg.png" /></figure><p>Our idea was to put the new software between the web server and our legacy software (zombie). This results in the fact that all requests go to the new system first. If this component doesn’t know the URL, it lets our zombie answer the request.</p><h4>So fury was born.</h4><p>Fury because we want our new code to be like the horse from fifties TV series. Fast, reliable and beautiful (clean code and SOLID). We started to think how we could achieve these visions for our new component. We had a look on the pages we wanted to replace and realised that all users see the same content on each request. A question popped in our minds. Do we need to render HTML on request at all?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*neB2ZTMKWblQawn_etj7eQ.png" /></figure><p>No! We decided to render HTML into key-value storage. So on request we just looked up content for a given URL and send it back. As most other online shops, we have filters too. If we would pre-render each combination into it’s own entry in storage, we would end up with several millions of entries. Well, that&#39;s not gonna work. We decided to render each product tile on a category page as a single entry in storage. Every time someone applies a filter, we do a new search with their category and selected filters. This will give us information about which products to show. With this list of SKUs we now go to storage and get the HTML for each product. We now have to just glue them together and send them back to the browser.</p><h4>Where is all this data coming from?</h4><p>For this we put a backend component which is completely decoupled from the frontend. It populates the data to key-value storage and searches. We call it Collect &amp; Export. This script will collect all needed data from the legacy database. During Export we render HTML for our key-value storage and generate data for our search engine.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*QTq3hGHwqNDFc0pD7xzdwg.png" /></figure><p>If a request comes in, web server asks fury frontend. This will go to key-value and gets the HTML. Sometimes fury has to ask search if filters were applied. On top, fury will read from legacy session storage to get information like login status and how many items a customer has in their whishlist and basket. With all of this applied, fury responds with the status “200 OK” to the web-server. In case the customer goes to checkout which is not part of fury, fury will return the status “404 Not Found”. In this case the web server will hand the request to the legacy system and there the request will be processed as usual.</p><h4>All this is done without any framework …</h4><p>… and just a short list of software dependencies. In the process, we learned a few things by doing things. We’re now fully responsible for our code — so there’s no possibility of blaming the framework anymore. We have to find and fix all the security issues our self because we miss a big contributing community. We can now perform refactoring more easily because of our 100% code coverage. We still depend to our zombie, but they are encapsulated in repository classes. From there only new and good structured objects are allowed to enter the fury environment.</p><h4>The zombie still lives …</h4><p>… but now we are in a position to first work on our biggest pain points and then focus on replacing other parts of the zombie step by step.</p><p>Doing this step by step enabled us to build, test and launch the first wave in less than 6 month. From now on we are able to ship product increments significantly faster.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*1NWpPSBG-JbZln_lltmzXA.png" /></figure><p><a href="https://tech.kartenmacherei.de/how-to-deploy-a-black-horse-2f81ed65cd47#.jrko8kpar">Next time</a> I’d like to show you how we deploy fury and how we make sure that every page works before we allow the customer to see the content.</p><ul><li><a href="http://www.slideshare.net/kartenmacherei/dont-fear-the-walking-dead-how-we-get-rid-of-our-monolith">Slides on SlideShare</a></li><li><a href="http://www.kartenmacherei.de/">Our website based on Fury</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6baca1a96f6f" width="1" height="1" alt=""><hr><p><a href="https://stories.celebrate.company/dont-fear-the-walking-dead-6baca1a96f6f">Don’t fear the walking dead.</a> was originally published in <a href="https://stories.celebrate.company/">celebrate company</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>