<?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 Bryan Beege Berry on Medium]]></title>
        <description><![CDATA[Stories by Bryan Beege Berry on Medium]]></description>
        <link>https://medium.com/@bryanbeegeberry?source=rss-1a14cbf495cd------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*4gGLZNOm3Jlv4hMx.</url>
            <title>Stories by Bryan Beege Berry on Medium</title>
            <link>https://medium.com/@bryanbeegeberry?source=rss-1a14cbf495cd------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 08:40:55 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@bryanbeegeberry/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[Project: MLOps-in-a-Box 2025–10–08]]></title>
            <link>https://medium.com/@bryanbeegeberry/project-mlops-in-a-box-2025-10-08-e974e990c076?source=rss-1a14cbf495cd------2</link>
            <guid isPermaLink="false">https://medium.com/p/e974e990c076</guid>
            <category><![CDATA[personal-project]]></category>
            <category><![CDATA[llmops]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[mlops]]></category>
            <category><![CDATA[llm-engineering]]></category>
            <dc:creator><![CDATA[Bryan Beege Berry]]></dc:creator>
            <pubDate>Wed, 08 Oct 2025 09:16:40 GMT</pubDate>
            <atom:updated>2025-10-08T09:16:40.792Z</atom:updated>
            <content:encoded><![CDATA[<p>It’s been a minute since I’ve worked on this. I decided to relax a bit and play with Claude Code for my <a href="https://github.com/TheBeege/warhammer-rp-character-sheet">Warhammer Fantasy RPG character sheet</a>. I made a good bit of progress on it, considering my frontend experience is limited, especially recent times. But this post isn’t about that project 😄Onward!</p><h3>Since Last Time on Dragonball Z!</h3><p>I’ve done a bit of reading.</p><p>I started reading <a href="https://www.oreilly.com/library/view/ai-engineering/9781098166298/">Chip Huyen’s AI Engineering book</a>. I vaguely remember reading something of hers previously and being impressed. I find this book quite good, too. I did some reading up on more detailed ML things, as my knowledge isn’t very deep. I now have a stronger understanding of <a href="https://jalammar.github.io/illustrated-transformer/">transformers</a>, refreshed some of my NLP knowledge and <a href="https://machinelearningmastery.com/what-are-word-embeddings/">added to it</a>, learned about <a href="https://jalammar.github.io/visualizing-neural-machine-translation-mechanics-of-seq2seq-models-with-attention/">Attention</a>, and got a general idea of how LLMs work. They’re actually a bit simpler than I expected, at least conceptually. Or maybe Chip’s writing and the articles she recommends are just really good 😆</p><p>I also started reading up on vector databases and the process of vectorization. I’m still a bit iffy on vectorization itself, but I think I have a good enough understanding to start building and learn more the hard way. I <em>think </em>I’m going to ditch Qdrant and instead use PostgreSQL. With pgvector, it seems to have good vector support, and even GCP’s Cloud SQL supports it. My only issue is that ZenML appears to only run on MySQL. I’m really tempted to dive in and submit a PR or twenty, but I gotta stay focused.</p><p>I also decided to rename this project. I believe I’m going to change it sufficiently that it won’t be as similar to the LLM Engineer’s Handbook as I originally intended. I might rename it again to LLMOps-in-a-box, but… Let’s see.</p><p>I considered reinstating the AWS bits, since I can use ZenML’s stacks to keep it as a separate thing, but I think my architecture will change enough that it’s not worth keeping their stuff. Seeing their reliance on the AWS CLI made me think, “Ya’ll motherf*****s need Terraform” because my mind is polluted by memes and the general internet. So what is this architecture I have in mind? Hell, I dunno. We’ll see once I build it. (Potential employers, please note: I’m having fun with this. For a real project, I’d do design docs and feedback sessions and all that nice stuff. I’m indulging, here.)</p><h3>Today’s Goal</h3><p>Ah… I definitely wrote this section header before I had a plan. It’s fine. My ADHD super-power lets me distract people with fun flavor things like this while I make decisions on-the-fly and hope they’re right.</p><p>So today’s goal will be multi-faceted:</p><ul><li>Verify everything actually runs</li><li>Configure the pipeline to pull my data</li></ul><p>That second one is ambiguous for a reason. I kind of want to pull my Lemmy posts, but… I’m not sure if I’ll have enough time today. So let’s leave the scope definition in a state where a proper PM would cry.</p><h4>Data Source Breakdown</h4><p>So remember <a href="https://medium.com/@bryanbeegeberry/project-mlops-in-a-box-2025-09-28-c90800401a51">my last post</a>, where I realized that this book is totally disrespecting LinkedIn’s robots.txt? Let’s breakdown potential data sources and if/how to use them.</p><p>For Medium, their robots.txt is much more permissive. But I have a question: <strong>Why the hell do we scrape everything for AI?</strong> In the case of OpenAI or Anthropic or whatever where they want to scrape <em>everything</em>, I get it. You’re not going to create custom logic for each site to grab from all their different API specs and such. Well, I don’t know if you noticed, but I’m not OpenAI. I’m only pulling from a couple of sites. Additionally, my dataset is much smaller. This means I need to take more care around the data quality. I have witnessed first-hand how not accounting for gaps in your data’s distribution profile can kill performance. So what does that mean for Medium? Well, I’m not gonna scrape their web pages. Unfortunately, Medium doesn’t allow new API integrations. However, they do have <a href="https://medium.com/feed/@bryanbeegeberry">an RSS feed</a>. It’s beautiful! I can see the tags. I can see the titles. Timestamps, post content as HTML, timestamps. It’s all in straightforward XML. Now, the LLM Engineer’s Handbook already has a scraper written for Medium… But I <strong>hate</strong> scraping. So no. I’ll rewrite my own. If this was a company project, I’d re-use their crawler, but this is <strong>my</strong> project. I’m not using a scraper unless there’s no alternative.</p><p>For Lemmy, API docs are here: <a href="https://join-lemmy.org/api/main">https://join-lemmy.org/api/main</a>. GGEZ.</p><p>Beyond those… I don’t really post content on the internet…? No, we are not going to get data from Discord. So yeah, that’s it.</p><h3>Random Thoughts While Working</h3><p><a href="https://github.com/TheBeege/MLOps-in-a-Box/blob/12fec34fddfec69690c49c7459ef764cac48c460/llm_engineering/application/crawlers/dispatcher.py">Their CrawlerDispatcher</a> is using the <a href="https://refactoring.guru/design-patterns/builder">Builder pattern</a>. Well, it’s not <em>exactly</em> the Builder pattern, but it’s the same idea. I’m a fan. Builder is definitely one of my favorite patterns, when it makes sense. They’re also using something like the <a href="https://refactoring.guru/design-patterns/strategy">Strategy pattern</a>, which is another one of my favorites.</p><p>I created accounts of HuggingFace and CometML. I hate signing up for more services these days, but I guess privacy is already a joke.</p><p>Okay, I got everything up and running until fine-tuning on SageMaker. Sounds like success. In order to get it to run, I had to modify it to pull my data properly. For some reason, Selenium was exploding when trying to parse the authors’ pages’ data. This is why crawling is a pain, boys and girls. However, it’s still using BeautifulSoup for Medium data. I also had to disable post and repository data source types, since I don’t have examples of those, yet.</p><p>Claude Code is immensely helpful. For the Medium crawler, it swapped out Selenium for Requests for me. I could have done it myself, but I would have needed to look up docs to remember, check that everything is still hooked up well, etc. Claude did it for me. It even caught that it still needed to inherit from the BaseCrawler class in order to keep everything working, which I would have missed in my first attempt.</p><p>Now, I have an issue. This system is designed to parse each article individually. If I do switch to the Medium RSS feed, I won’t be able to specify each article individually in the configurations. It only provides a listing. Additionally, it seems like the feed doesn’t support pagination. It only provides the last 10 articles. Well, shit. If Medium doesn’t do API integrations anymore and only provides the last 10 articles in their RSS feed, maybe I’m stuck with scraping. If this really will be MLOps-in-a-Box, ten records for a user in insufficient. On the other hand, I can’t expect to manually define every data source. On the other other hand, I don’t think the focus of this is on the data collection itself. Maybe that’s a problem I can dive into later. Ugh, I’ll leave it with the scraping for now. The focus of this is for me to learn about the training, evaluation, and data collection loop, not to get in the weeds of parsing web data.</p><p>In any case, my immediate goals for today are met. The system works up until training. I’m not using SageMaker, so it’s broken at that point. Seems like it’s time for me to really get my hands dirty and build my own training pipeline. Now things are getting exciting! 🎉</p><p>I’m going to skip the Lemmy data for now, since I should really focus on the actual model stuff. I know data engineering. I don’t know model training.</p><p>Random side note… These pages in this book are at least half code. I’m starting to think this book was just a money grab.</p><p>Nope, that’s just because it was the implementation part of that chapter. Reading up on RAGs, I realized I misunderstood them in some critical ways. I believed that a RAG retrieved necessary information from original data sources at query time. That is true, but they don’t simply query the original data source to find the relevant information. Instead, they vectorize (embed) chunked bits of data from the original source. At query time, they consult the vector database to identify similar data, <em>then</em> they go fetch from the original source to augment the prompt. My previous understanding was critically flawed because it lacked a definition for how to identify relevant data from the external sources.</p><h3>End</h3><p>I feel like things were a little all over the place today, but progress was made. Tomorrow, I’ll get training up and running… somehow. Should be a fun struggle.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e974e990c076" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Project: MLOps-in-a-Box 2025–09–28]]></title>
            <link>https://medium.com/@bryanbeegeberry/project-mlops-in-a-box-2025-09-28-c90800401a51?source=rss-1a14cbf495cd------2</link>
            <guid isPermaLink="false">https://medium.com/p/c90800401a51</guid>
            <category><![CDATA[personal-project]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[llm-engineering]]></category>
            <category><![CDATA[mlops]]></category>
            <category><![CDATA[llmops]]></category>
            <dc:creator><![CDATA[Bryan Beege Berry]]></dc:creator>
            <pubDate>Wed, 08 Oct 2025 05:16:36 GMT</pubDate>
            <atom:updated>2025-10-08T05:16:36.800Z</atom:updated>
            <content:encoded><![CDATA[<p>This is mostly just a journal for the project I’m working on. I’m posting this after the fact. Previously, I’d been keeping notes within the repository, but I decided they don’t belong there. So here we go.</p><p>Repository: <a href="https://github.com/TheBeege/MLOps-in-a-Box">https://github.com/TheBeege/MLOps-in-a-Box</a></p><h3><strong>TODO</strong></h3><p>Run the data engineering stuff on my data</p><ul><li>Get everything running via Docker Compose for local dev</li><li>Set up OpenStack on my server to treat it like a cloud</li><li>Kubernetes-ify everything that’s not on Kubernetes</li><li>Install <a href="https://www.comet.com/docs/opik/self-host/kubernetes/">CometML/Opik on Kubernetes</a></li><li>Set up a self-hosted model registry instead of Hugging Face</li></ul><p><strong>Done</strong></p><ul><li>✅ Get things up and running locally with uv</li></ul><h3><strong>Log</strong></h3><h4><strong>Chapter 2</strong></h4><p>Learned about <a href="https://poethepoet.natn.io/">Poe the Poet</a>. I’d been hoping for a task runner in the Python ecosystem like NPM can do. Seems tightly coupled to Poetry and uv, but I’m okay with it for now.</p><p>ZenML seems really cool, but the self-hosted versus cloud offerings have the fatal catch that all these systems have: RBAC requires their cloud offering. To me, this is a deal-breaker if you want a system like this for sensitive things like medicine. I’ve run into similar frustrations with Prefect, though we just threw Keycloak in front of it. I was working on completely blocking access to Prefect and having another system only be able to trigger pipelines, but I ran out of time for that.</p><p>The frustrating thing about all these tools is that they’re missing some integrations and paywall some features. I understand that people need to make money, but I’d love to have an open-source, completely pluggable system for all components. I want a ZenML with RBAC that can hook up to some identity platform or that can defer RBAC to another system like Keycloak. I understand that’s a hard problem to solve, but call me greedy, I guess. I suppose putting interfaces with RBAC and whatnot in front of these kinds of systems is the way to go, but it just feels like it should be unnecessary. If I actually had enough free time, I’d just build my own.</p><p>They keep talking about the stacks part of ZenML and how it was a deciding factor in their tool choice. I’m not yet sure I believe them, but maybe I should try it be re-instating the AWS bits and adding my OpenStack things as a stack in ZenML.</p><p>CometML/Opik look cool. I’m looking forward to learning about and playing with those.</p><p>I’m not using SageMaker. AWS has really nice systems and some of the best customer service I’ve ever experienced, but I hate Amazon as a company. I think building my own Kubernetes-based training system will be a good experience, anyway.</p><h4><strong>Chapter 3</strong></h4><p>Decent data architecture so far. I probably would have done it the same.</p><p>But… Maybe it’s because I’ve been out of the data game for a minute or because everyone does things on commodity hardware these days, but I don’t get why everyone says systems like MongoDB or PostgreSQL can’t handle millions of records. I literally just led a team to do that over the past few months on PostgreSQL, and it performs fine. Hell, we’re not even using the columnar plugins that we should be using because GCP doesn’t support them. Maybe there’s something I don’t know about, yet. We haven’t used these systems to supply training data quite yet; they’re more focused on analytics. I’m sure there’s a good reason that I’ll learn soon enough. I just don’t want to pay for Snowflake or BigQuery, though I have heard that they can be cheap if you’re smart about it.</p><p>I’m nearly out of time for my current session. I’m hoping I’ll have some time after dinner to actually write some damn code.</p><p>I took a quick look at <a href="https://www.linkedin.com/robots.txt">LinkedIn’s </a><a href="https://www.linkedin.com/robots.txt">robots.txt</a>. This is funny:</p><pre>User-agent: anthropic-ai<br>Disallow: /<br>User-agent: ChatGPT-User<br>Disallow: /<br>User-agent: ClaudeBot<br>Disallow: /<br>User-agent: Claude-Web<br>Disallow: /<br>User-agent: Claude-User<br>Disallow: /<br>User-agent: Claude-SearchBot<br>Disallow: /<br>User-agent: cohere-ai<br>Disallow: /<br>User-agent: GPTBot<br>Disallow: /</pre><p>That list goes on. And there’s one critical one:</p><pre>User-agent: *<br>Disallow: /</pre><p>Also, I forgot that Microsoft owns LinkedIn. I just got sad again. Ends up that reading posts is via their Community Management API, which “are only available to registered legal organization for commercial use cases only.” Lovely. I wonder if I should send a copy of this book to them, since it explicitly violates their robots.txt. In any case, I guess I can’t used LinkedIn content, or at the very least, I don’t want to go through the hassle it would take to ethically get access. I do find it funny that they wrote a LinkedIn crawler but don’t have configurations to actually crawl it. So they encourage their readers to violate terms, but they don’t do it themselves. Assholes. I’m deleting the crawler. I’m out of time to test if I broke some import, though.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c90800401a51" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Automating Infrastructure with Docker]]></title>
            <link>https://medium.com/codeseoul/automating-infrastructure-with-docker-32f85cab37c4?source=rss-1a14cbf495cd------2</link>
            <guid isPermaLink="false">https://medium.com/p/32f85cab37c4</guid>
            <category><![CDATA[devops]]></category>
            <category><![CDATA[infrastructure]]></category>
            <category><![CDATA[automation]]></category>
            <category><![CDATA[tutorial]]></category>
            <category><![CDATA[docker]]></category>
            <dc:creator><![CDATA[Bryan Beege Berry]]></dc:creator>
            <pubDate>Sat, 14 Oct 2017 02:09:54 GMT</pubDate>
            <atom:updated>2023-04-22T07:06:43.559Z</atom:updated>
            <content:encoded><![CDATA[<p>This has been updated as of 2023–04–22.</p><p>Disclaimer: I am not an expert. I’m quite the opposite. I usually learn just enough to accomplish what I need, and I move on. Do not take what I write here as best practice! This is only intended to expose the potential of technologies. I always recommend reading the official documentation or tutorials, and ideally, translating them if necessary :D</p><p>Today, I’d like to tell you about Docker. We’re going to create some mock services and automate running them in containerized, source-controlled infrastructure using Docker. What does this mean?</p><p>First, a word about a preceding alternative. Most people are familiar with virtual machines: you setup an OS the way you want, create an image of it, use some hypervisor to host virtual machines, and deploy your images at will. For many years, this has been an effective method to diversify the usage of physical compute resources. However, virtualization has its downsides. Entire operating systems have a pretty large amount of overhead in terms of resource usage. Boot times can be painful. Virtual disk images are large, making it a pain to move things around, and they’re a pain to create. (This has been mitigated by Vagrant, but that’s also a recent development.) This works, and it has worked for many years. However, are there alternatives?</p><p>Enter Docker. Docker uses what are called “containers” instead of virtual machines. These utilize Linux cgroups (unfortunately, inside a virtual machine on Mac and Windows) to create an isolated but low-overhead operating environment. Containers have noticeably less overhead than virtual machines, and startup is wicked fast. Docker specifically provides a wealth of tools to make working with containers easy and even pleasant. Similar to Vagrant, you can define a text file called a Dockerfile that defines how the container is configured. There’s a collection of CLI tools for building Docker images from Dockerfiles, pushing and pulling images from Dockerhub or Docker registry (git -&gt; Github : Docker -&gt; Dockerhub), and managing containers. There are also a wealth of tools to orchestrate Docker containers. We’ll cover what orchestration is and means later in this post.</p><p>Sounds great! How do we get started? Well, it depends on your machine.</p><p>Running Windows? Heads up that you’ll need to install the Windows Subsystem for Linux first: <a href="https://learn.microsoft.com/en-us/windows/wsl/install.">https://learn.microsoft.com/en-us/windows/wsl/install.</a> Once that’s done, check out Docker for Windows here: <a href="https://docs.docker.com/desktop/install/windows-install/">https://docs.docker.com/desktop/install/windows-install/</a></p><p>Running on Mac? You’ve got it easy. Check out Docker for Mac here: <a href="https://docs.docker.com/desktop/install/mac-install/">https://docs.docker.com/desktop/install/mac-install/</a></p><p>Running on Linux? Here’s a link for Ubuntu: <a href="https://docs.docker.com/engine/install/ubuntu/">https://docs.docker.com/engine/install/ubuntu/</a>. There are links for other distributions in the navigation pane on the left.</p><p>Double-check to make sure that the docker service is started!</p><p>Now that you have things installed, let’s build something. Let’s build a self-contained environment with an HTTP server in one container, a client in another container, and configurations for running all of it.</p><p>Let’s get started. First, we’re going to build a <em>super</em> simple HTTP server. Create a new folder (I called mine server), and create these two files in it.</p><p>server.sh:</p><pre>#!/bin/sh</pre><pre>trap exit EXIT INT TERM HUP</pre><pre>while true<br>do<br>  # Why don&#39;t we use -k? hello.http is only piped in once<br>  # We also send it to the background so that `wait` can catch the trap<br>  ncat -l $(hostname -I) 8000 &lt; hello.http &amp;<br>  wait<br>done</pre><p>hello.http:</p><pre>HTTP/1.0 200 OK</pre><pre>&lt;html&gt;<br>  &lt;body&gt;<br>    &lt;h1&gt;Hello, world!&lt;/h1&gt;<br>  &lt;/body&gt;<br>&lt;/html&gt;</pre><p>Sloppy, I know, but it works. This script uses nmap’s ncat command to serve up some text written out following HTTP to simulate a web server. Since ncat stops listening after fulfilling a request, we just loop it forever. If you weren&#39;t sure, <em>don&#39;t use ncat for a legitimate HTTP server</em>. This is just a tech demo for another tool, so I think The Right Way™ gods will forgive me.</p><p>If you have a POSIX shell available, this will run an HTTP server on port 8000. While its running, you can visit localhost:8000and see a tiny page with &quot;Hello, world!&quot;. This will work for our example.</p><p>Now, let’s containerize it. Create a new file called Dockerfile also in this folder. No extension! It should look like this:</p><pre>FROM alpine</pre><pre>RUN apk add --no-cache nmap-ncat</pre><pre>ADD ./ ./</pre><pre>EXPOSE 8000</pre><pre>CMD [&quot;./server.sh&quot;]</pre><p>This is Dockerfile syntax. You can see the full docs here: <a href="https://docs.docker.com/engine/reference/builder/">https://docs.docker.com/engine/reference/builder/</a></p><p>A Dockerfile defines a Docker image. This can be compared to a virtual machine image. Again, a Dockerfile is much like a Vagrantfile if you’re familiar with Vagrant.</p><p>FROM indicates that we&#39;re going to inherit from another Dockerfile. Yes, Docker supports inheritance. It&#39;s lovely. Basically, this image will start from the mentioned image, and we&#39;ll build on top of it. Where does this other Dockerfile/image come from? Docker Hub! <a href="https://hub.docker.com/">https://hub.docker.com/</a> In this case, we&#39;re going to base our image on Alpine.</p><p>RUN is simply a way to run a command to help setup the image. In this case, we&#39;re installing ncat since we&#39;ll need it for the demo.</p><p>ADD is how you get files from your filesystem into the image. Here, we&#39;re just copying everything in our directory to the image&#39;s working directory. I&#39;m lazy.</p><p>EXPOSE tells the container which ports will need network access. This is important if you&#39;re expected inbound network requests.</p><p>CMD is the command that will run when the container is started. Viola! We have our first Dockerfile setup. Let&#39;s run it:</p><pre>docker build -t demo_http_server .</pre><p>This will search the current directory for a Dockerfile, create an image from it, and tag it as demo_http_server:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/592/0*gTBgPnASFIDTFWnM.png" /></figure><p>Now, we can run the container:</p><pre>docker run -d -p 8000:8000 demo_http_server</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/672/0*meWcxKqI0sAKBHI8.png" /></figure><p>The -d flag tells Docker to detach from the container, meaning it will run in the background. The -p 8000:8000 tells docker to bind the container&#39;s port 8000 to the host&#39;s port 8000, meaning our server will be accessible via the host machine at port 8000. Lastly, we tell it the image ID or tag or the image we want to create a container from. Now, you can hit your web server from the browser.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/291/0*GCUrYHl7bYm2Sk92.png" /></figure><p>Congratulations! You have your first docker container running!</p><p>However, let’s clean up after ourselves. Let’s see what containers are currently running:</p><pre>docker ps</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*38GaZKw4hC5xuWnW.png" /></figure><p>You can see the container, its ID, the image it’s based on, the currently running command, its name, and other info. Docker will generate names for containers if you don’t provide one.</p><p>Let’s make sure to stop the container, since we won’t use this specific one anymore. Since my container’s name is suspicious_jackson, I can simply do</p><pre>docker stop suspicious_jackson</pre><p>Using the container’s ID works, too. If you run docker ps again after this, you&#39;ll see no containers running.</p><p>Fancy, huh? Let’s get more fancy.</p><p>Make a separate folder called client, and drop this shell script in there. Don’t forget to add execute permissions!</p><pre>#!/bin/sh<br># This script should take in 2 arguments: the host to curl and its port</pre><pre>url=$1<br>port=$2</pre><pre>while true<br>do<br>  curl -s $url:$port<br>  sleep 2<br>done</pre><p>Now, create a Dockerfile in that folder:</p><pre>FROM alpine</pre><pre>RUN apk add --no-cache curl</pre><pre>ADD client.sh client.sh</pre><pre>ARG server<br>ARG port=8000</pre><pre>ENV server $server<br>ENV port $port</pre><pre>CMD [&quot;sh&quot;, &quot;-c&quot;, &quot;./client.sh $server $port&quot;]</pre><p>You’ll notice this file is slightly different.</p><p>ARG allows our container to take arguments, meaning we can configure some things at image build time.</p><p>ENV sets environment variables within the container.</p><p>In this case, we’re taking arguments to the container, making them available to the shell environment, then passing them as inputs to our shell script. This script just makes an HTTP request to the given server at the given port repeatedly with a 2-second delay. Let’s see them work together.</p><p>Now, there is a way to get these two containers talking to each other running manually, but that’s dumb. It’s a lot of effort. There’s an easier way.</p><p>Instead, we can create a docker-compose file, which sets up multiple containers. I have this in a docker-compose.yml file in the folder above my server and client folders.</p><pre>version: &#39;3&#39;</pre><pre>services:</pre><pre>  server:<br>    build:<br>      context: ./server<br>    ports:<br>      - &quot;8000:8000&quot;</pre><pre>  client:<br>    build:<br>      context: ./client<br>      args:<br>        server: &#39;server&#39;<br>        port: 8000<br>    links:<br>      - &quot;server&quot;</pre><p>This file will allow us to use the docker-compose command to build and run multiple containers that are networked together to talk to each other.</p><p>version refers to the version of docker-compose we&#39;re using.</p><p>services is where we define each of the services we&#39;ll run in containers.</p><p>args corresponds to the arguments we defined in our Dockerfile.</p><p>links establishes links between services, where the linked service has a hostname entry with the name of the service. So from the client, we can reference the server container as server since we defined the link and service as server.</p><p>I won’t cover the rest, but if things are unclear to you, ask in a comment! No judgement here :)</p><p>From here, we can simply run the below:</p><pre>docker-compose up</pre><p>You’ll see the server and client both startup, and you’ll see the client request from the server every second infinitely. To stop them, just ctrl + c.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/880/0*dYRI0yVJGg8LVWon.png" /></figure><p>We could do more! There are prebuilt images for all sorts of things, like databases, message queues, and web servers. You can setup an entire application stack for your local development environment. I did this at work (and impressed my coworkers).</p><p>In production, you could use Docker Swarm, Kubernetes, or Mesos to orchestrate and manage containers for your various services. Easy, observable, fault tolerant.</p><p>We did a lot today!</p><ul><li>Installing Docker</li><li>Creating a Dockerfile</li><li>Building a Docker image</li><li>Running a Docker container</li><li>Defining services using docker-compose</li><li>Running multiple services in a network</li></ul><p>Great job!</p><p>For now, this blog entry is long enough. :) Thank you for your time! I hope you found the material helpful. Please comment with any questions, comments, or suggestions. Also let me know if you have any requests for technologies you’d like me to cover!</p><p>Huge thanks to Jiyoung with Django Girls Seoul for translating! You should definitely check them out if Python and/or Django are in your future. I also recommend my own meetup, Learn Teach Code Seoul. If you think this blog was good work, it’s a testament to how well our two groups collaborate together, so you should check out both!</p><p>I welcome feedback! Don’t worry, I won’t be offended. I see feedback as a gift for helping me improve. I’m greedy. Give me gifts!</p><p>If you want need the full source, you can find it at <a href="https://github.com/TheBeege/blog-docker">https://github.com/TheBeege/blog-docker</a></p><h3>도커로 인프라 자동화하기</h3><p>이것은 2023–04–22에 업데이트 되었습니다.</p><p>읽어주세요: 저는 전문가가 아닙니다. 사실 그 반대라고 할 수 있죠. 보통 제가 필요한 걸 할 수 있을 정도로만 배우고 넘어갑니다. 여기에 쓴 내용이 모범 사례라고 생각하지 말아주세요! 그저 기술로 할 수 있는 일들을 드러내기 위해서 썼을 뿐입니다. 저는 항상 공식문서나 튜토리얼을 읽어볼 것을 추천하는데요. 이상적으로는 필요하다면 번역을 해보는 것이 좋습니다.</p><p>오늘은 도커를 다뤄볼 건데요. 연습용 서비스를 만들어보고, 도커를 이용해 컨테이너 기반의 소스 제어 인프라에서 서비스 실행을 자동화 할겁니다. 이게 무슨 뜻일까요?</p><p>먼저, 이전에 쓰던 방식을 이야기해보겠습니다. 많은 분들이 가상 머신은 익숙하실 겁니다. 원하는대로 OS를 설정하고, 이미지를 만들고, 가상 머신을 호스트하는 하이퍼 바이저를 사용하고, 마음껏 이미지를 배포하는 것이죠. 수년동안 이 방법은 물리적 컴퓨팅 리소스를 다양하게 사용하는 효과적인 방법이었습니다. 하지만 가상화에도 단점이 있습니다. 전체 운영 체제에 리소스 사용량 측면에서 상당한 오버헤드가 있는데요. 부팅하는 시간이 고통스러울 수 있습니다. 가상 디스크 이미지는 크기가 커서 이를 옮기기에도 고통스럽고 만들기도 어렵습니다.(Vagrant를 통해서 완화되긴 했지만 이것도 최근에 개발되었습니다.) 이 방법은 수년동안 효과가 있었습니다. 그렇지만 대안은 없는 걸까요?</p><p>도커에 입문해봅시다. 도커는 가상머신 대신에 ‘컨테이너’라는 것을 사용합니다. 이들은 ‘Linux cgroups’(Mac 및 Windows의 가상 시스템 내부)를 사용해 분리되어 있지만 오버헤드가 적은 운영 환경을 생성합니다. 컨테이너는 가상머신보다 오버헤드가 현저히 적고, 부팅도 무지 빠릅니다. 도커는 특히 컨테이너 작업을 더 쉽고 즐겁게 할 수 있는 풍부한 도구를 제공합니다. Vagrant와 마찬가지로 컨테이너 구성 방법을 정의하는 Dockerfile이라는 텍스트 파일을 정의할 수 있습니다. Dockerfile들을 읽고 도커 이미지를 만들거나, Dockerhub 또는 Docker registry (git -&gt; Github : Docker -&gt; Dockerhub)에서 이미지를 내려받거나 올리고, 컨테이너를 관리하는 CLI 도구들도 있습니다. 그밖에 도커 컨테이너들을 조정(orchestrate)하는 툴들도 많습니다. ‘조정(Orchestration)’이 무엇인지는 이 포스팅의 후반부에서 다루겠습니다.</p><p>좋습니다! 그럼 어떻게 시작하면 좋을까요? 여러분이 갖고있는 머신에 따라 다릅니다.</p><p>Windows를 실행하시나요? 먼저 Linux용 Windows 하위 시스템을 설치해야 한다는 점에 유의하세요: <a href="https://learn.microsoft.com/en-us/windows/wsl/install">https://learn.microsoft.com/en-us/windows/wsl/install</a>.</p><p>설치가 완료되면 여기에서 ‘ Windows ‘용 ‘ Docker ‘를 확인하세요 : <a href="https://docs.docker.com/desktop/install/windows-install/">https://docs.docker.com/desktop/install/windows-install/</a>.</p><p>맥을 사용한다면 간단합니다. 맥을 위한 도커는 이곳을 확인하세요: <a href="https://docs.docker.com/desktop/install/mac-install/">https://docs.docker.com/docker-for-mac/install/</a></p><p>리눅스를 사용한다면 아래 링크에 Ubuntu를 위한 내용이 있습니다: <a href="https://docs.docker.com/engine/install/ubuntu/">https://docs.docker.com/engine/install/ubuntu/</a></p><p>도커 서비스가 시작되었는지 다시 한번 확인해주세요!</p><p>이제 설치가 끝났으니 뭔가를 만들어봅시다. 하나의 컨테이너에 HTTP 서버, 다른 컨테이너에 클라이언트 및 모든 컨테이너를 실행하기 위한 설정을 포함하는 독립적인(self-contained) 환경을 구축해보죠.</p><p>이제 시작해봅시다. 먼저 <em>완전</em> 간단한 HTTP 서버를 만들어보겠습니다. 새로운 폴더를 만들고 (저는 server라고 했습니다), 아래 2개의 파일을 만듭니다.</p><p>server.bash:</p><pre>#!/bin/sh</pre><pre>trap exit EXIT INT TERM HUP</pre><pre>while true<br>do<br>  # `-k`를 사용하는건 어떤가요? `hello.http`는 단 한번만 사용하면 됩니다.<br>  # 또한 `wait`가 함정을 잡을 수 있도록 후방으로 보내기도 합니다.<br>  ncat -l $(hostname -I) 8000 &lt; hello.http &amp;<br>  wait<br>done</pre><p>hello.http:</p><pre>HTTP/1.0 200 OK</pre><pre>&lt;html&gt;<br>  &lt;body&gt;<br>    &lt;h1&gt;Hello, world!&lt;/h1&gt;<br>  &lt;/body&gt;<br>&lt;/html&gt;</pre><p>조잡하긴 하지만 잘 실행됩니다. 이 스크립트는 nmap의 ncat 명령을 사용해서 웹서버를 시뮬레이션하기 위해 HTTP 다음에 작성된 텍스트를 실행합니다. ncat 은 요청을 실행하고 나서는 요청 받는 것을 멈추기 때문에 이것을 영원히 반복하게 됩니다. 확실하지 않다면, <em>실제 HTTP 서버를 위해서는 ncat을 사용하지 마세요</em>. 이것은 다른 툴을 보여주기 위한 하나의 기술 데모일 뿐입니다. 그러니, &#39;올바른 방법의 신&#39;은 나를 용서해주겠죠.</p><p>POSIX 셸을 사용할 수 있다면, 8000 포트에서 HTTP 서버를 실행할 것입니다. 실행이 되는 동안 localhost:8000에 접속해서 &quot;Hello, world!&quot;가 적혀있는 간단한 페이지를 확인할 수 있습니다. 우리 예제에서는 이정도면 됩니다.</p><p>이제 컨테이너화해봅시다. 같은 폴더에 Dockerfile 이라는 새로운 파일을 만들어주세요. 확장자명은 없습니다! 아래와 같이 보여야 합니다.</p><pre>FROM alpine</pre><pre>RUN apk add --no-cache nmap-ncat</pre><pre>ADD ./ ./</pre><pre>EXPOSE 8000</pre><pre>CMD [&quot;./server.sh&quot;]</pre><p>이것은 Dockerfile 의 신택스입니다. 전체 문서는 여기서 확인하세요: <a href="https://docs.docker.com/engine/reference/builder/">https://docs.docker.com/engine/reference/builder/</a></p><p>Dockerfile은 도커 이미지를 정의합니다. 이것은 가상 머신의 이미지와도 비슷합니다. 다시 한번 말씀드리지만, 여러분이 Vagrant에 익숙하다면 Dockerfile은 Vagrant와 매우 비슷합니다.</p><p>FROM은 다른 도커파일에서 상속받을 것이라는 것을 나타냅니다. 네, 도커는 상속을 지원합니다. 좋죠. 기본적으로 이미지는 위에서 언급한 이미지에서 시작해서, 우리는 그 위에다 빌드하게 됩니다. 이 다른 Dockerfile/이미지는 어디에서 오는 것일까요? 바로 Docker Hub입니다! <a href="https://hub.docker.com/">https://hub.docker.com/</a> 이 경우에는 Alpine에 이미지를 저장하겠습니다.</p><p>RUN은 이미지 설치를 도와주는 명령어입니다. 이 경우, 우리는 데모에 필요한 ncat을 설치합니다.</p><p>ADD는 파일 시스템에서 파일을 이미지로 받아오는 방법입니다. 여기서 그냥 디렉토리의 모든 것을 복사해서 이미지의 작업 디렉토리에 붙여넣겠습니다. 저는 게으르니까요.</p><p>EXPOSE는 컨테이너에 네트워크 접근이 필요한 포트를 알려줍니다. 인바운드 네트워크 요청이 예상되는 경우 이것이 중요합니다.</p><p>CMD는 컨테이너가 시작될 때 실행될 명령입니다. 우와! 첫 번째 Dockerfile 설정을 완료했습니다. 이제 실행해봅시다:</p><pre>docker build -t demo_http_server .</pre><p>그러면 현재 디렉토리에서 Dockerfile이 있는지 검색해서, 이미지를 만들고, demo_http_server라는 태그를 지정하게 됩니다:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/592/0*7UFyyHQT6yXlxiVv.png" /></figure><p>이제 컨테이너를 실행할 수 있습니다:</p><pre>docker run -d -p 8000:8000 demo_http_server</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/672/0*WaPe1e1TqbJn4h4H.png" /></figure><p>-d 플래그는 Docker가 컨테이너에서 분리되도록 지시하는데, 이는 백그라운드에서 실행되는 것을 의미합니다. -p 8000 : 8000은 docker에 컨테이너의 8000 포트를 호스트의 8000 포트에 바인드하도록 지시합니다. 즉, 우리 서버는 호스트 머신을 통해 8000 포트에 접근할 수 있습니다. 마지막으로 도커에게 어떤 이미지 ID 또는 태그 또는 이미지를 컨테이너로 만들고 싶은지 알려줍니다. 이제 브라우저에서 웹 서버에 접근할 수 있습니다.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/291/0*FxbWo3TxIPxupkpH.png" /></figure><p>축하합니다! 여러분의 첫 번째 도커 컨테이너가 작동 중입니다!</p><p>하지만 정리를 좀 해봅시다. 현재 실행중인 컨테이너를 봅시다.</p><pre>docker ps</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*QtTUSaNaisRW4aZM.png" /></figure><p>이제 컨테이너, 해당 ID, 기반 이미지, 현재 실행중인 명령, 이름 및 기타 정보를 볼 수 있습니다. 컨테이너의 이름을 지정하지 않으면 Docker가 이름을 생성해줍니다.</p><p>이제 이 컨테이너를 더 이상 사용하지 않을 것이기 때문에 컨테이너를 중지합시다. 제 컨테이너의 이름이 suspicious_jackson이므로, 아래처럼 간단하게 중지할 수 있습니다.</p><pre>docker stop suspicious_jackson</pre><p>컨테이너의 ID를 사용할 수도 있습니다. 나중에 docker ps를 다시 실행하면 실행중인 컨테이너가 없다는 것을 볼 수 있습니다.</p><p>멋지죠? 더 멋진 것을 해봅시다.</p><p>다른 폴더에 client라는 이름으로 파일을 만들고 아래의 셸 스크립트를 적어주세요.</p><pre>#!/bin/sh<br># This script should take in 2 arguments: the host to curl and its port</pre><pre>url=$1<br>port=$2</pre><pre>trap exit EXIT INT TERM HUP</pre><pre>while true<br>do<br>  curl -s $url:$port<br>  sleep 2<br>done</pre><p>이제 그 폴더에 Dockerfile을 만들어주세요:</p><pre>FROM alpine</pre><pre>RUN apk add --no-cache curl</pre><pre>ADD client.sh client.sh</pre><pre>ARG server<br>ARG port=8000</pre><pre>ENV server $server<br>ENV port $port</pre><pre>CMD [&quot;sh&quot;, &quot;-c&quot;, &quot;./client.sh $server $port&quot;]</pre><p>이 파일이 아까와는 약간 다르다는 것을 눈치채실 겁니다.</p><p>ARG는 컨테이너가 인자를 받을 수 있게 해줍니다. 이는 이미지를 빌드할 때 몇 가지 사항을 설정할 수 있음을 의미합니다.</p><p>ENV는 컨테이너 내의 환경 변수를 설정합니다.</p><p>이 경우 우리는 컨테이너에 인자를 받아서 쉘 환경에서 사용할 수 있게 만든 다음, 셸 스크립트에 입력값으로 전달합니다. 이 스크립트는 지정된 포트에서 지정된 서버에 2초 간격으로 반복되는 HTTP 요청을 만듭니다. 이 모두가 어떻게 작동하는지 봅시다.</p><p>이제, 이 두개의 컨테이너가 서로 수동으로 통신하는 방법이 있지만, 그것은 많은 노력이 필요합니다. 더 쉬운 방법이 있습니다.</p><p>대신 여러 개의 컨테이너를 설정하는 도커 작성 파일을 만들 수 있습니다. 저는 server와 client의 폴더 위에 있는 폴더에 docker-compose.yml에 설정해두었습니다.</p><pre>version: &#39;3&#39;</pre><pre>services:</pre><pre>  server:<br>    build:<br>      context: ./server<br>    ports:<br>      - &quot;8000:8000&quot;</pre><pre>  client:<br>    build:<br>      context: ./client<br>      args:<br>        server: &#39;server&#39;<br>        port: 8000<br>    links:<br>      - &quot;server&quot;</pre><p>이 파일은 docker-compose 명령을 사용할 수 있게 해주는데, 서로 네트워크로 연결된 여러 컨테이너를 만들고 실행할 수 있는 스크립트입니다.</p><p>version은 우리가 사용하고있는 docker-compose의 작성 버전을 나타냅니다.</p><p>services는 우리가 컨테이너에서 실행할 각각의 서비스를 정의하는 곳입니다.</p><p>args는 Dockerfile에서 정의한 인수에 해당합니다.</p><p>links는 서비스 간 링크를 생성합니다. 링크된 서비스는 서비스 이름과 같은 hostname 엔트리를 갖습니다. 우리는 링크와 서비스를 server로 정의했으므로 클라이언트에서 서버 컨테이너를 server로 참조할 수 있습니다.</p><p>나머지 부분을 다루지 않겠지만, 명확하지 않은 부분은 댓글을 달아주세요! 어떤 질문이든 상관없습니다.</p><p>여기에서 우리는 아래의 명령어를 간단하게 실행할 수 있습니다:</p><pre>docker-compose up</pre><p>서버와 클라이언트가 모두 시작되는 것을 볼 수 있습니다. 매 초마다 서버가 클라이언트에 보내는 요청을 무한히 보게 될 것입니다. 이를 중지하려면 그냥 Ctrl + C를 누르면됩니다.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/880/0*aBzMNxY_7GPU39LM.png" /></figure><p>더 많은 것을 할 수도 있습니다! 데이터베이스, 메시지 큐, 웹 서버와 같은 모든 종류의 것에 대해 미리 작성된 이미지가 있습니다. 로컬 개발 환경에 맞게 전체 응용 프로그램 스택을 설정할 수도 있습니다. 저는 직장에서 이것을 했습니다 (그리고 동료들을 감동시켰죠).</p><p>프로덕션 환경에서는 Docker Swarm, Kubernetes 또는 Mesos를 사용하여 다양한 서비스를 위한 컨테이너를 조율하고 관리할 수 ​​있습니다. 쉽고, 관찰할 수 있고, 오류를 견딥니다(fault tolerant).</p><p>우리는 오늘 많은 것을 했습니다!</p><ul><li>도커 설치</li><li>Dockerfile 만들기</li><li>도커 이미지 만들기</li><li>도커 컨테이너 실행</li><li>docker-compose를 이용한 서비스 정의</li><li>한 네트워크에서 여러 서비스 실행</li></ul><p>잘했어요!</p><p>첫 글 치고는 길었네요 :) 시간 내 주셔서 감사합니다! 자료가 도움이 되었기를 바랍니다. 질문, 의견, 제안이 있다면 댓글을 달아주세요. 그리고 제가 다루었으면 하는 기술이 있다면 알려주세요!</p><p>번역을 도와준 Django Girls Seoul의 <a href="https://jyhwng.github.io/">지영에게</a> 감사를 드립니다! Python이나 Django에 관심이 있다면 꼭 한번 그들을 만나보세요. 제가 운영하는 Learn Teach Code Seoul의 밋업도 추천합니다. 이 블로그가 좋다고 생각한다면, 우리 두 그룹이 얼마나 잘 콜라보하는지 보여주는 것이죠. 그러니 두 그룹 모두 한번 확인해보세요!</p><p>피드백은 환영합니다! 걱정하지 마세요. 저는 상처받지 않습니다. 피드백은 개선을 위한 선물이라고 생각합니다. 저는 욕심이 많아요. 선물을 주세요!</p><p>전체 소스가 필요하다면 여기서 보실 수 있습니다. <a href="https://github.com/TheBeege/blog-docker">https://github.com/TheBeege/blog-docker</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=32f85cab37c4" width="1" height="1" alt=""><hr><p><a href="https://medium.com/codeseoul/automating-infrastructure-with-docker-32f85cab37c4">Automating Infrastructure with Docker</a> was originally published in <a href="https://medium.com/codeseoul">CodeSeoul</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Why I Hate the Internet in Korea]]></title>
            <link>https://medium.com/@bryanbeegeberry/why-i-hate-the-internet-in-korea-826d4e44d4a8?source=rss-1a14cbf495cd------2</link>
            <guid isPermaLink="false">https://medium.com/p/826d4e44d4a8</guid>
            <category><![CDATA[web-development]]></category>
            <category><![CDATA[korea]]></category>
            <category><![CDATA[rant]]></category>
            <category><![CDATA[web-design]]></category>
            <category><![CDATA[programming-rant]]></category>
            <dc:creator><![CDATA[Bryan Beege Berry]]></dc:creator>
            <pubDate>Mon, 25 Sep 2017 07:05:09 GMT</pubDate>
            <atom:updated>2017-09-25T07:05:09.077Z</atom:updated>
            <content:encoded><![CDATA[<p>I just came across this monstrosity: <a href="https://www.wanted.co.kr/events/meetup201709">https://www.wanted.co.kr/events/meetup201709</a></p><p>At first, you know, whatever. Looks cool. It’s some advertisement for an event. Great.</p><p>Until you try to translate it. Or copy text. Or use a screenreader. Or try to find it on Google.</p><p>This page is only composed of images. Only images. This is just images with text in them.</p><p>My rage is equivalent to that of a thousand stars. My hatred of this carries such velocity as to escape across the event horizon of black holes. I want to crush this site into the tiny gluons and quarks that make up the memory bits on the servers hosting this abomination.</p><p>The worst part? <strong>This is all over Korea.</strong> Everywhere.</p><p>It wouldn’t even be that hard to build correctly! There are plenty of themes that support this format. The text is minimal. The design tricks are common. This could take a junior front-end dev only one or two hours to build correctly. This is pure laziness.</p><p>It’s funny, too. The government is wondering why it can’t attract foreign tech talent. This. Thiiiiiiiiiiis.</p><p>If someone could translate this and send me the translation, I would love you forever. If we could then share it on all of the social media in Korean, maybe we could get through to people. I’m just going to be sad for the rest of the day. kthxbye</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=826d4e44d4a8" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[First Time Django]]></title>
            <link>https://medium.com/djangogirlsseoul-codecamp/first-time-django-c55202d1b2ec?source=rss-1a14cbf495cd------2</link>
            <guid isPermaLink="false">https://medium.com/p/c55202d1b2ec</guid>
            <category><![CDATA[django]]></category>
            <category><![CDATA[python]]></category>
            <dc:creator><![CDATA[Bryan Beege Berry]]></dc:creator>
            <pubDate>Fri, 24 Feb 2017 05:13:30 GMT</pubDate>
            <atom:updated>2017-02-24T05:56:50.514Z</atom:updated>
            <content:encoded><![CDATA[<p>I recently had the privilege of joining up with the <a href="http://www.djangogirlsseoul.org/">Django Girls Seoul</a> study group for <a href="https://www.djangoproject.com/">Django</a>. Having previous experience with the <a href="http://mean.io/">MEAN</a> stack, I was prepared for something conceptually easy but an absolute mess in practice. I was worried I’d need to memorize dozens of different technologies, untangle dependency hell, and pray to the unholy gods that glued together this ship made of toothpicks while in the middle of a monsoon.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LSyjRz9CqR5IYtPcFvICZQ.jpeg" /><figcaption>You can guess how that would end up</figcaption></figure><p>To put it shortly…</p><blockquote>I was gloriously wrong</blockquote><p>Seriously, this was amazing.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Tjb8Zz__xu5rf1vUVbT3dA.jpeg" /><figcaption>Yeah… It was more like this.</figcaption></figure><p>Signing up was easy, and they even assured me they’d have English speakers to help me out. We met at <a href="http://dcamp.kr/">D.CAMP</a>, which as far as I can tell is a startup incubator. It started pretty typically. The organizers introduced themselves and spoke about the study group, then they had everyone introduce themselves.</p><p>준범 then proceeded to give a brief lecture on Django, REST, and MVC, or in Django’s case, MVT. I was able to follow most of it, considering most of the tech terms have English origins. The way he explained it was excellent. As someone who also runs an educational meetup, I often feel the desire to jump in if someone is explaining something poorly. With 준범, not even once. From the order of how he progressed through the logic to how he balanced keeping things simple while maintaining accuracy, the concepts were very well explained. He even got an applause afterwards. Hell yeah.</p><p>We then started brainstorming ideas. We ended up dividing into three groups, each with its own project: a user management system, a blog, and a Python version of Instagram, or Pystagram. While I’m not much of an Instagram user, I joined that project, as it seemed to have the most interesting problem set.</p><p>The class is currently ongoing, and the Pystagram project is based on this <a href="http://blog.hannal.com/category/start-with-django-webframework/">series from Kay on the rails</a>. Even running it through a translator, it’s really straightforward. Each of us on the project is working individually, since everyone learns at a different pace, but it’s excellent having access to people with experience. I’ve both asked for help and given help, which is the best of both worlds.</p><p>Already knowing <a href="https://en.wikipedia.org/wiki/Model%E2%80%93view%E2%80%93controller">MVC</a>, Django is #easyMode. Django does things slightly differently, using MVT instead of MVC. It’s actually a welcome change. I’ve often found my Controllers to be boringly simple. Django handles maybe half of the Controller logic built into the framework, and you use the View to decide how exactly to display your data. The Template is literally that: an HTML template. Simple. Define your data as models. Massage and format your data in the view. Drop it into an HTML via a template. Voilà. Done.</p><p>It even handles database migrations for you. Changed how your models are setup? No problem! Django comes with a manage.py file that will generate and apply schema changes for you. The generated migration files even fit nicely into source control.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/500/1*NkA9fJZoUH2iQWPzYhPU9A.png" /></figure><p>The manage.py script does more than database migrations. It will generate portions of your app for you, much like <a href="http://yeoman.io/">Yeoman</a>, but with less crying and <a href="http://www.imgrum.net/media/1224233373905271893_1574454942">gnashing of teeth</a>. This lets you focus on writing your logic instead of writing boilerplate. It’s also organized in a sane way.</p><p>Django also has built-in and extension apps, consideration of static assets for deployment, and a built-in admin interface. Overall, it’s just satisfying to work with.</p><p>In short, participating in the Django girls study group has been wonderful: great people and great technology. I strongly recommend it to anyone, new or experienced, that wants to experiment with Python-driven web servers.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c55202d1b2ec" width="1" height="1" alt=""><hr><p><a href="https://medium.com/djangogirlsseoul-codecamp/first-time-django-c55202d1b2ec">First Time Django</a> was originally published in <a href="https://medium.com/djangogirlsseoul-codecamp">Django Girls Seoul | 장고걸스서울</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>