<?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 Jānis Erlihs on Medium]]></title>
        <description><![CDATA[Stories by Jānis Erlihs on Medium]]></description>
        <link>https://medium.com/@erlihs?source=rss-8b0cc5c24d98------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*nC0zDqe0ZNMJVn_xWhOBTQ.jpeg</url>
            <title>Stories by Jānis Erlihs on Medium</title>
            <link>https://medium.com/@erlihs?source=rss-8b0cc5c24d98------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 23:58:07 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@erlihs/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[Slop, slop, and done. Or how software re-invented IKEA]]></title>
            <link>https://medium.com/@erlihs/slop-slop-and-done-or-how-software-re-invented-ikea-c12749763286?source=rss-8b0cc5c24d98------2</link>
            <guid isPermaLink="false">https://medium.com/p/c12749763286</guid>
            <category><![CDATA[keepass]]></category>
            <category><![CDATA[vibe-coding]]></category>
            <category><![CDATA[vs-code-extensions]]></category>
            <category><![CDATA[keepassxc]]></category>
            <dc:creator><![CDATA[Jānis Erlihs]]></dc:creator>
            <pubDate>Sun, 01 Nov 2026 06:32:14 GMT</pubDate>
            <atom:updated>2026-11-01T06:32:14.224Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*kOyT-Qu6yPDcF9krxrmXJw.jpeg" /></figure><p><a href="https://x.com/realAlita/status/2115044679276265812">This tweet</a> got my attention. There is a vibe of nostalgia among us, IT people, even among the most progressive, 100% generated code, ones. Our once unique, highly appreciated and admirable job has shifted into..</p><blockquote>Slop, slop, and done.</blockquote><p>From a consumer perspective? It is an “look ma, they just re-invented IKEA” momentum.</p><p>Need a new table or chair? Ikea has one. Cheap as fókk. And with quality just good enough to do the job.</p><p>Or. You could find local craftsmanship to have a custom made, unique. Expensive. Long waiting time. Quality? Could be even controversial as oak or other types of hard wood might need special care.</p><p>Software has entered its Ikea phase. With AI — slop, slop and done — cheap, and just good enough to do the job. Or..</p><blockquote>Every gangsta now is his own carpenter.</blockquote><p>My latest real time story. Needed a way to share a bunch of passwords and other secrets. Evend grandma knows the KeePass is the right tool to do the thing. But you need to install an app — either KeePass or KeePassXC. I am a lazy person. So professionally lazy, that I am using VsCode for every job out there — extensions for container management, all possible DB platforms, language tools — you name it. My first reaction — is there a KeePass extension for VsCode. Surprisingly — there is not.</p><p>So, here we go. Slop, slop, some $40 for tokens (yes I know — but tests here matters as it is a bit of cryptography involved), a couple of hours over two weekends — and the world got <a href="https://marketplace.visualstudio.com/items?itemName=erlihs.vscode-kdbx">KeePass for VsCode extension</a>. An IKEAish one.</p><p>If I had to make it by hand — a half year over weekends project. Would it be better in terms of quality? Probably not. Would I have had greater satisfaction getting the job done? Definitely, yes. So here is the nostalgia part. That part.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c12749763286" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[ORM or not?]]></title>
            <link>https://medium.com/@erlihs/orm-or-not-5aa8a691f618?source=rss-8b0cc5c24d98------2</link>
            <guid isPermaLink="false">https://medium.com/p/5aa8a691f618</guid>
            <category><![CDATA[kysely]]></category>
            <category><![CDATA[oracle-database]]></category>
            <category><![CDATA[orm]]></category>
            <category><![CDATA[oracle-rest-data-services]]></category>
            <dc:creator><![CDATA[Jānis Erlihs]]></dc:creator>
            <pubDate>Sun, 26 Jul 2026 16:28:41 GMT</pubDate>
            <atom:updated>2026-07-26T16:28:41.714Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xTEGUtgYb35oBCGKB69VtQ.jpeg" /></figure><p>Maybe the real question isn’t whether we should use an ORM.</p><p>Maybe the real question is: <strong>why are we still using relational databases and SQL in 2026?</strong></p><p>This technology is so old it feels like it was around when dinosaurs were still wandering the Earth. Isn’t there anything better by now?</p><p>Seems not. Despite modern architectures, cloud-native everything, microservices, event buses, AI assistants, and enough abstraction layers to make your CPU cry… eventually it all comes down to one thing:</p><p>An old, good SQL query.</p><p>Actually… a <strong>Structured Query Language query</strong>. Yeah. Fine.</p><h4>So here we are</h4><p>We have an application.</p><p>We have a relational database.</p><p>Now we need to write queries.</p><p>How?</p><h4>Plain SQL</h4><p>The old-school approach is still there: write SQL as text.</p><p>As long as you’re using parameter binding (remember that SQL injection thingy you learned in kinder garden), you’re generally fine.</p><p>The downside?</p><p>You have to know SQL.</p><p>It’s usually not type-safe.</p><p>IDE support is… let’s call it “optimistic.”</p><p>AI isn’t much better. Most of the hallucinations I’ve seen lately involve confidently suggesting columns that simply don’t exist. Which is actually understandable — someone else probably named that column, and AI has no clue what your schema really looks like. MCP that gigs into your DB helps but you have to create the skill. Meh.</p><h4>ORM</h4><p>Someone invented the ORM. Actually, lots of people did, so fortunately there’s no single person who has to apologize for it.</p><p>With the rise of OOP, Java, and Hibernate, an entire generation of enterprise software learned a valuable debugging shortcut:</p><blockquote>- knock, knock, who’s there?</blockquote><blockquote>[ 10 seconds to minute, each time random, pause ]</blockquote><blockquote>- Java</blockquote><blockquote>- Why so slow, Java?</blockquote><blockquote>[ 10 seconds to minute, each time random, pause ]</blockquote><blockquote>- I do hibernate.</blockquote><p>Yes, I’m exaggerating. Slightly.</p><p>ORMs try to map object-oriented models into SQL. They build joins. They discover relationships. They generate queries.</p><p>Mechanically.</p><p>Without business context.</p><p>Without understanding <em>why</em> you’re asking the database something.</p><p>It’s like taking the craft of writing SQL and replacing it with something that knows all the rules but none of the intent.</p><p>Sometimes that’s exactly what you want.</p><p>Sometimes it definitely isn’t.</p><p>But sooner or later it becomes <strong>pain</strong>.</p><h4>The middle ground</h4><p>The community has spent years swinging between handwritten SQL and ORMs, trying to solve a problem that probably doesn’t have a perfect solution.</p><p>The closest thing to a sensible middle ground I’ve seen lately is <a href="https://kysely.dev/"><strong>Kysely</strong></a>.</p><p>It’s basically plain SQL…</p><p>…except you write it in TypeScript.</p><p>You still think in terms of queries. You’re not pretending your database is an object graph. But you get type safety, autocompletion, and compile-time validation.</p><p>Even more interesting is its migration approach. Versioning database changes suddenly feels natural. Multiple developers can work against the same codebase without inventing elaborate rituals around SQL scripts.</p><p>Definitely worth checking out.</p><h4>The idea</h4><p>I have a dream. A very niche dream. An Oracle Database kind of dream.</p><p>What if we took the ideas behind Kysely and applied them to Oracle Database — but pushed them a bit further?</p><p>Not just type-safe table definitions.</p><p>Not just type-safe query builders.</p><p>Not just typescript migrations replacing sqlcl complexities.</p><p>What about <strong>type-safe PL/SQL based builders</strong>?</p><p>What about <strong>generating DBMS package APIs that are checked at compile time</strong>?</p><p>And together with Oracle REST Data Services (ORDS), <strong>exposing package procedures as REST endpoints automatically?</strong></p><p>The whole paradigm starts to shift.</p><p>Instead of writing a backend in Java, .NET, or Node.js that talks to Oracle…</p><p>…you write your backend in TypeScript.</p><p>And the runtime is Oracle Database.</p><h4>Next</h4><p>Lets burn some tokens to get this s*it done:)</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5aa8a691f618" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Gems: Three Things That Helped Me the Most in H1 2026]]></title>
            <link>https://medium.com/@erlihs/gems-three-things-that-helped-me-the-most-in-h1-2026-c2b72110c263?source=rss-8b0cc5c24d98------2</link>
            <guid isPermaLink="false">https://medium.com/p/c2b72110c263</guid>
            <category><![CDATA[typescript]]></category>
            <category><![CDATA[developer-experience]]></category>
            <category><![CDATA[developer-tools]]></category>
            <dc:creator><![CDATA[Jānis Erlihs]]></dc:creator>
            <pubDate>Sun, 07 Jun 2026 05:53:28 GMT</pubDate>
            <atom:updated>2026-06-07T05:53:28.252Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ZTRavJ-q45XP7xceMEp9lw.jpeg" /></figure><h4>Kysely</h4><p><a href="https://kysely.dev/">https://kysely.dev/</a></p><p><strong>Kysely </strong>is an absolute gem. It eliminates the need for Liquibase and made database migrations simple again.</p><p>Database migrations are now just db-up and db-downone-liners in the CI/CD pipeline. No XML files, no heavyweight tooling to learn.</p><p>And — Kysely is <strong>not an ORM in the traditional sense</strong>.</p><p>Instead of defining entities and letting a framework generate SQL behind the scenes, you write queries directly in TypeScript:</p><pre>const users = await db<br>  .selectFrom(&#39;users&#39;)<br>  .select([&#39;id&#39;, &#39;email&#39;])<br>  .where(&#39;active&#39;, &#39;=&#39;, true)<br>  .execute()</pre><p>What you get is:</p><ul><li>Full SQL expressiveness</li><li>Excellent type safety</li><li>Great autocomplete</li><li>No ORM magic</li><li>Easy-to-review migrations</li></ul><p>It feels like writing SQL, but with TypeScript helping you avoid mistakes before they ever reach production. For teams that like SQL but want modern developer ergonomics, Kysely hits it.</p><p>There surely is “kysely for Oracle Database” in “things to do in my life” list…</p><h4><strong>Turborepo</strong></h4><p><a href="https://turborepo.dev/">https://turborepo.dev/</a></p><p>A monorepo with ten TypeScript applications and shared packages sounds great in theory.</p><p>Until build times start hurting.</p><p>Turborepo solves that problem.</p><p>Its incremental build system understands dependency graphs and only rebuilds projects affected by your changes. Touch a shared package and only the applications depending on it get rebuilt. Change a frontend app and your backend services stay untouched.</p><p>The result is:</p><ul><li>Faster local development</li><li>Faster CI pipelines</li><li>Simpler dependency management</li><li>Shared tooling across the entire organization</li></ul><p>It simply works and no quirks observed so far.</p><h4><strong>Oxlint + Oxc Formatter</strong></h4><p><a href="https://oxc.rs/">https://oxc.rs/</a></p><p>The JavaScript ecosystem by the efforts of team behind VoidZero is getting faster, much faster.</p><p>With Oxlint and the Oxc formatter are becoming viable replacements for ESLint and Prettier. Had tired out in real world projects, it works.</p><p><strong>And does it so fast that formatting and linting becomes almost unnoticeable</strong>, even on pretty large codebases.</p><p>Combined with Vite 8, the ecosystem seems to be converging toward a future where JavaScript tooling is no longer the bottleneck. One last missing piece left:</p><p><strong>tsgo</strong>.</p><p>We’re already at the point where it’s correct to say: <em>“I’ve deployed my TypeScript app to production twice while your Java service is still warming up.”</em></p><p>Once TypeScript compilation gets the same treatment, developer experience will be close to perfect.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c2b72110c263" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[It is happening]]></title>
            <link>https://medium.com/@erlihs/it-is-happening-ed69a24b351f?source=rss-8b0cc5c24d98------2</link>
            <guid isPermaLink="false">https://medium.com/p/ed69a24b351f</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[developer-experience]]></category>
            <category><![CDATA[prompt-engineering]]></category>
            <dc:creator><![CDATA[Jānis Erlihs]]></dc:creator>
            <pubDate>Sun, 05 Apr 2026 07:10:42 GMT</pubDate>
            <atom:updated>2026-04-05T07:10:42.333Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8LUnvCpS2rWxJ29RGUBP_w.jpeg" /></figure><p>Over the past six months, I’ve had the chance to build meaningful things using practices and a tech stack that don’t make me question why I’m still in this industry.</p><p>And the most interesting part — it actually works. Instead of the classic 7–9 person agile team, a far better productivity with 2–3 micro-teams. With AI in the mix. A lot of prompting — tens to hundreds of prompts per day.</p><p>And it has consequences, but others that one might think of.</p><p>Recently, I wrote a short business-level document — an action plan. A colleague read it and said it felt “AI-ish”. The catch? I wrote it entirely myself. No prompting this time — a rare exception.</p><p>We often talk about how AI is adapting to humans. But there’s another side to it: humans are adapting to AI.</p><p>I tracked my communication style — and there it was. Catching myself saying, “you are absolutely right” to people. Structuring explanations in the way Anthropic would. Organizing notes and tasks as if everything were a markdown spec.</p><p>So, instead of AI adapting to humans..</p><p>You don’t agree?</p><p>You are absolutely right!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ed69a24b351f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[It Should Have Been a SQL Join]]></title>
            <link>https://medium.com/@erlihs/it-should-have-been-a-sql-join-ec7e8cf8bb4d?source=rss-8b0cc5c24d98------2</link>
            <guid isPermaLink="false">https://medium.com/p/ec7e8cf8bb4d</guid>
            <category><![CDATA[application-development]]></category>
            <category><![CDATA[architecture]]></category>
            <dc:creator><![CDATA[Jānis Erlihs]]></dc:creator>
            <pubDate>Sat, 31 Jan 2026 17:02:28 GMT</pubDate>
            <atom:updated>2026-01-31T17:02:28.204Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*T60Cxn148Nzg9iq4wNzTIg.jpeg" /></figure><p>If I got $10k every time I saw something like this, I’d never have to work again.</p><h4>Phase one</h4><p>An app needs to retrieve a list of items for a user.</p><p>Everything is stored in a relational database.</p><p>Items live in an Items table. Users live in a Users table.</p><p>The app makes a database request to fetch items and gets a list of records: item id, name, user id.</p><p>Then — for each of those 10 records — it makes another database request to fetch the user’s name.</p><p>That’s 11 queries for 10 rows.</p><p>Sure, this example is trivial. But the principle is the same. Add more related entities. Add more attributes. Suddenly you’re making hundreds or thousands of queries just to render a single page.</p><p>It should have been a SQL join.</p><h4>Phase two</h4><p>Things are slow, so the next solution takes the stage: caching.</p><p>Now the first request is still slow, the next ones are fast, memory usage explodes, cache invalidation becomes a nightmare, and the app’s reputation shifts from slow to slow and unreliable.</p><p>It should have been a SQL join.</p><h4>Phase three</h4><p>Things are still bad, so someone suggests Elasticsearch.</p><p>Now you have indexing jobs, re-indexing jobs, orchestration, observability, fallbacks, dashboards, alerts — and an app that is maybe less slow, but definitely more fragile. Around this time, adding a single new field costs half a kingdom and a royal marriage.</p><p>It should have been a SQL join.</p><h4>The grand finale</h4><p>But none of this matters, because now we have AI.</p><p>AI will fix it. Or maybe the next model will. Or maybe agents will team up, remove humans from the loop entirely, and update their rule book with a single line:</p><p>It should have been a SQL join.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ec7e8cf8bb4d" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AWR is easy]]></title>
            <link>https://medium.com/@erlihs/awr-is-easy-fc3d7c389a80?source=rss-8b0cc5c24d98------2</link>
            <guid isPermaLink="false">https://medium.com/p/fc3d7c389a80</guid>
            <category><![CDATA[monitoring]]></category>
            <category><![CDATA[oracle-database]]></category>
            <dc:creator><![CDATA[Jānis Erlihs]]></dc:creator>
            <pubDate>Tue, 16 Dec 2025 12:32:47 GMT</pubDate>
            <atom:updated>2025-12-16T12:32:47.616Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ubl-IB9I9uitG37N2HPwew.jpeg" /></figure><p>Investigating database performance issues is hard, they said.</p><p>Not if you Oracle Database.</p><h4>Automatic Workload Repository</h4><p>In Oracle databases, the <strong>Automatic Workload Repository</strong> is a built-in performance monitoring feature that:</p><ul><li>Automatically <strong>collects performance statistics</strong> at regular intervals</li><li>Stores historical data about <strong>database load and activity</strong></li><li>Allows to <strong>analyze performance issues over time</strong> using reports</li></ul><p>And getting the report is as easy as it can be with <strong>SQLcl</strong></p><pre>SQL&gt; awr list snap<br>   SNAP_ID         DBID BEGIN_INTERVAL_TIME                END_INTERVAL_TIME                  FLUSH_LEVEL    <br>__________ ____________ __________________________________ __________________________________ ______________<br>         1    952378024 12-DEC-25 06.33.02.000000000 PM    16-DEC-25 11.55.40.852000000 AM    TYPICAL<br>         2    952378024 16-DEC-25 11.55.40.852000000 AM    16-DEC-25 11.59.19.920000000 AM    TYPICAL</pre><p>will give a list of snapshots. If snapshots are not scheduled automatically e.g. hourly, you can create your own at any time by</p><pre>SQL&gt; awr create snap<br>Snapshot taken, ID: 1</pre><p>and create report between any snapshots</p><pre>SQL&gt; awr create html 1 2<br>Report written to: awr_G9D4EA764C074F7_ODBVUEADB_2025-12-12-18.33.02_2025-12-16-11.59.19.html</pre><h4>How to read?</h4><p>First, slow queries and routines in different dimensions — elapsed time, CPU time, IO Wait, Gets and so on.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*eZzEno4uKLa16cxX1XM3nA.png" /></figure><p>Ideal for systematic pro-active monitoring and improving, instead of allowing things to get wrong.</p><p>And for making part of QA practices — accompanying load and stress tests.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=fc3d7c389a80" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Blue on Blue! Not.]]></title>
            <link>https://medium.com/@erlihs/blue-on-blue-not-9a0d4831c9b4?source=rss-8b0cc5c24d98------2</link>
            <guid isPermaLink="false">https://medium.com/p/9a0d4831c9b4</guid>
            <category><![CDATA[oracle-database]]></category>
            <category><![CDATA[devops]]></category>
            <category><![CDATA[vuejs]]></category>
            <dc:creator><![CDATA[Jānis Erlihs]]></dc:creator>
            <pubDate>Mon, 17 Nov 2025 19:57:06 GMT</pubDate>
            <atom:updated>2025-11-17T19:57:06.375Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*oAAWOI5trY8PiUx99HlYjQ.jpeg" /></figure><p>It is still too many cases out there where downtime during release deployments exist. Here is a small concept that can help for particular technology set — Single Page Application — any React, Angular, VueJs — running directly against Oracle Rest Data Services and Oracle Database.</p><h4>Web (SPA)</h4><p>Let’s assume a Vue-based SPA served by Nginx . The goal? Every deploy feels boring and predictable.</p><p>On the web side, the idea is simple: your SPA is just static files (HTML, JS, CSS) served by Nginx inside a container. If you can update those files <strong>atomically</strong>, you get zero downtime.</p><p>Assume we have <strong>releases</strong> directory with timestamped folders and a current <em>symlink</em>:</p><ul><li>releases/2025–11–17–103000/ — new build</li><li>releases/2025–11–10–091200/ — previous build</li><li>releases/current-&gt;2025–11–10–091200/ — what nginx is currently serving</li></ul><p>Nginx only ever serves from releases/current, so the deploy process becomes:</p><p>1. Build SPA assets</p><p>2. Copy them into a new timestamped folder</p><p>3. Flip the current <em>symlink</em> to point to the new folder</p><p>4. Optionally clean up older releases</p><p>Because the <em>symlink</em> switch is atomic, users never see a half-deployed state.</p><p>A minimal nginx config for serving the SPA from releases/current looks like this (simplified):</p><pre>worker_processes 1;<br><br>events { worker_connections 1024; }<br><br>http {<br>  include       mime.types;<br>  default_type  application/octet-stream;<br><br>  sendfile on;<br>  keepalive_timeout 65;<br><br>  server {<br>    listen 80;<br>    server_name apps.example.com;  # apps site<br><br>    # Prod: SPA served from blue/green current symlink<br>    root /var/www/apps/current;<br>    index index.html;<br><br>    location / {<br>      try_files $uri $uri/ /index.html;<br>    }<br>  }<br>}</pre><p>The only trick is the root pointing to releases/current. Everything else is standard SPA hosting. A <strong>classic blue/green layout</strong> on the server:</p><ul><li>/var/www/apps/blue</li><li>/var/www/apps/green</li><li>/var/www/apps/current -&gt; /var/www/apps/blue</li></ul><p>Nginx is configured to serve from the current <em>symlink</em>, not from blue or green directly. The deploy workflow then does the following:</p><p>1. Detect which slot is active by resolving /var/www/apps/current (blue or green)</p><p>2. Pick the other slot as the <strong>inactive</strong> target</p><p>3. <em>Rsync </em>the new build into the inactive directory</p><p>4. Fix file ownership/permissions for nginx</p><p>5. Atomically flip current to point to the inactive slot</p><p>6. Validate nginx config and reload</p><p>The atomic flip on the remote host looks roughly like this:</p><pre>TARGET_DIR=/var/www/apps/green<br><br>sudo ln -sfn &quot;$TARGET_DIR&quot; /var/www/apps/current.tmp<br>sudo mv -T /var/www/apps/current.tmp /var/www/apps/current<br></pre><p>This ln + mv pattern guarantees that current is always a valid directory: clients either see the blue version or the green version, never a half-updated state. Combined with Nginx’s SPA-friendly routing, that gives you a clean blue/green rollout with instant switch-over and straightforward rollback (flip back to the previous slot).</p><h4>Database (Oracle Database)</h4><p>Frontends are easy to redeploy. Databases, not so much. DDL changes can lock tables, break running code, or force downtime if not handled carefully.</p><p>Oracle’s <strong>Edition-Based Redefinition (EBR)</strong> solves this by allowing <strong>multiple versions of your application logic to exist side by side</strong>.</p><p>How EBR works (in practice)? EBR introduces the concept of <strong>Editions</strong>:</p><ul><li>An edition is a copy-on-write layer for editioned objects (packages, views, synonyms, etc.)</li><li>You can create a new edition, change code there, and still keep the previous edition running</li><li>Sessions are bound to an edition; old sessions keep using the old code, new sessions use the new code</li></ul><p>The project’s install scripts can automatically create a <strong>new edition for each release</strong>.</p><pre>-- Create edition for the new release<br>CREATE EDITION edition_v_0_0_161 AS CHILD OF edition_v_0_0_160;<br><br>-- Deploy new PL/SQL code into the new edition<br>ALTER SESSION SET EDITION = edition_v_0_0_161;<br>@install_changes_for_0_0_161.sql<br><br>-- If all is OK, switch over the new edition as default<br>ALTER DATABASE DEFAULT EDITION = edition_v_0_0_161</pre><p>Existing sessions continue to run against edition_v_0_0_160 until they finish. New sessions (e.g. new app pods/containers) are directed to the new edition.</p><p>Over time, the edition chain can get long. Since editions are mostly metadata, the performance impact is nearly none, but you’ll typically schedule a maintenance window now and then to drop old editions back towards ORA$BASE.</p><p>Also from this moment on instead of CREATE PACKAGE … is CREATE EDITIONABLE PACKAGE … .</p><h4>Rounding up</h4><p>Deployment pipeline shall take care that the new app is deployed only if DB deployment is successful.</p><p>Also it might be worth to ship with apps config Database edition and pass it down with API requests and do ALTER SESSION SET EDITION to ensure that app always faces the right edition.</p><p>Here goes <a href="https://github.com/odbvue/odbvue">a practical example</a> how all of this works together. Not trivial, but possible. Having zero downtime for years is nice.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9a0d4831c9b4" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Master Your Infrastructure as Code]]></title>
            <link>https://medium.com/@erlihs/master-your-infrastructure-as-code-ee94c2badd04?source=rss-8b0cc5c24d98------2</link>
            <guid isPermaLink="false">https://medium.com/p/ee94c2badd04</guid>
            <category><![CDATA[terraform]]></category>
            <category><![CDATA[iac]]></category>
            <category><![CDATA[oci]]></category>
            <category><![CDATA[devops]]></category>
            <dc:creator><![CDATA[Jānis Erlihs]]></dc:creator>
            <pubDate>Thu, 30 Oct 2025 09:06:23 GMT</pubDate>
            <atom:updated>2025-10-30T09:06:23.893Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*kEbUbxYqbLr5nAPPjhdb0g.jpeg" /></figure><p>In the early days of computing, software ran locally — tightly coupled to the hardware it lived on. Organizations maintained on-premise servers in data centers, manually configuring machines and installing software directly on physical systems. Each environment was unique, and scaling meant physically adding new servers — a process that was slow, costly, and prone to human error.</p><p>The rise of <strong>Cloud Computing</strong> in the mid-2000s revolutionized this model. Services like <strong>Amazon Web Services (AWS)</strong>, <strong>Microsoft Azure</strong>, and <strong>Google Cloud Platform (GCP)</strong> introduced virtualized infrastructure accessible on demand. Instead of purchasing hardware, organizations could <em>rent</em> compute power and storage. This enabled scalability, flexibility, and reduced capital costs — but much of the configuration still relied on manual setup through cloud consoles or ad-hoc scripts.</p><p>As cloud adoption matured, the next leap was <strong>Infrastructure as Code (IaC)</strong> — the practice of defining and managing infrastructure through declarative or programmatic code. This allowed developers to describe servers, networks, and services as code that could be versioned, tested, and automated. IaC bridged the gap between development and operations, making environments reproducible and infrastructure changes auditable.</p><h4>There’s More Than Just the Big Three</h4><p>For a long time, “<em>the cloud</em>” essentially meant one of three names: AWS, Azure, or GCP. But today, there’s another serious contender worth considering — <strong>Oracle Cloud Infrastructure (OCI)</strong>.</p><p><strong>OCI</strong> is Oracle’s modern cloud platform. While Oracle is best known for its databases, OCI has evolved into a full-scale cloud provider offering compute, networking, storage, database, AI/ML, and developer services on par with the major players.</p><p>What sets OCI apart is its <strong>focus on enterprise-grade performance and cost efficiency</strong>. It provides <strong>bare metal servers</strong>, <strong>dedicated virtual networks</strong>, and <strong>autonomous database services</strong>, enabling businesses to run mission-critical applications with tight control, high throughput, and low latency.</p><p>While AWS, Azure, and GCP pioneered the public cloud era with vast service portfolios and global reach, <strong>OCI differentiates itself through architecture and economics</strong>. Its <strong>flat, non-oversubscribed network design</strong> minimizes congestion and jitter — perfect for large databases, analytics, and financial workloads. OCI also offers <strong>simpler, more transparent pricing</strong> with <strong>lower egress costs</strong>, making it attractive for data-intensive operations.</p><p>Compared to <strong>AWS,</strong> OCI is often less complex and more cost-effective. Against <strong>Azure</strong>, it provides stronger isolation and hardware-level security. And versus <strong>Google Cloud</strong>, it’s more tuned for traditional enterprise workloads and Oracle database performance rather than AI-first or developer-centric use cases.</p><p>And for startups, side projects, or proof-of-concepts, OCI’s <strong>Always Free Tier</strong> offers a generous starter set — perfect for testing and learning at zero cost, with the option to scale into enterprise-grade performance later on.</p><h4>OCI and IaC</h4><p>Now, OCI and IaC, what options do we have? Especially if being small, we do not want and can not invest heavily in managing infrastructure. Quite many options:</p><ul><li><strong>oci-cli</strong>: A command-line tool to manage Oracle Cloud resources quickly without using the console — great for scripting and automation.</li><li><strong>terraform</strong>: An Infrastructure-as-Code tool that lets you define and provision OCI (and multi-cloud) infrastructure declaratively and repeatably.</li><li><strong>pulumi</strong>: A modern IaC framework that uses real programming languages (like Python, TypeScript, or Go) to define and deploy infrastructure.</li><li><strong>ansible</strong>: An automation tool for configuration management and application deployment — ideal for post-provisioning setup and orchestration.</li><li><strong>oci-sdk</strong>: Language-specific SDKs (Python, Java, Go, etc.) that allow direct programmatic access to OCI services from your applications.</li><li><strong>raw REST API</strong>: The lowest-level interface — direct HTTP calls to OCI endpoints, offering full control but requiring more manual effort.</li></ul><h4>Why Terraform?</h4><p>After spending some time experimenting with all the available options — oci-cli, Ansible, Pulumi, the OCI SDKs, and even a few raw REST calls — I found Terraform to be the straightest and least error-prone path to getting real infrastructure up and running. The feedback loop was fast: define, apply, and watch your environment take shape exactly as described.</p><p>What helped — Sonnet Haiku 4.5 and GPT-5 (as of late 2025) guided quite well, suggesting configurations, clarifying errors, and helping reason through the setup.</p><p>So, in short, even without being a DevOps person, I was still able to get the work done in a reasonable time — and end up with a one-liner that raises a full set of enterprise-level infrastructure.</p><h4>Hands On</h4><p>So for those who are practically interested — here is <a href="https://github.com/odbvue/odbvue/tree/main/i13e/oci/basic">a code and a guide</a> on how to master OCI Free Tier with Autonomous Database, Web application server, Properly configured network, Object Storage and even Email Delivery.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ee94c2badd04" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The future winners will be]]></title>
            <link>https://medium.com/@erlihs/the-future-winners-will-be-7ae82e4dfd74?source=rss-8b0cc5c24d98------2</link>
            <guid isPermaLink="false">https://medium.com/p/7ae82e4dfd74</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[insourcing]]></category>
            <category><![CDATA[digital-transformation]]></category>
            <category><![CDATA[information-technology]]></category>
            <category><![CDATA[business]]></category>
            <dc:creator><![CDATA[Jānis Erlihs]]></dc:creator>
            <pubDate>Sun, 19 Oct 2025 13:48:52 GMT</pubDate>
            <atom:updated>2025-10-19T13:48:52.948Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hGrjtVB44inM7CWwQzEAEw.jpeg" /></figure><p>AI will change things. But not in the way many thinks.</p><p>After a year of AI hype we can be certain of few things.</p><p>AI is not a replacement of deterministic business processes. It can be used to create and adjust process automation and tools, but it is not a replacement of well written transaction processing or analytics. Due to its randomness in nature.</p><p>AI is not a replacement of IT engineers either. Yes, there are layoffs, bet lets admit, there were too many people in the industry that should have better off somewhere else in the first place.</p><p>Where is the AI good at? Right. Learning. One can super easy acquire knowledge that previously required years of experience. One can validate any idea on the fly. And one person with the right AI skills and tools now can accomplish a job of a team. What this leads to?</p><p>I believe, this leads to harsh time for classical IT vendors. The “we are better, stronger, cheaper, smarter, faster, exclusive“ type of guys. “Our unique approach is first to understand client business requirements..“ — yea right. Like everybody else is out there just to dumb code solutions.</p><p>I believe, this also is a funeral bell for majority of SaaS, as any piece of business process automation or service can easily be copied and transformed as a software as software development costs tend to zero.</p><p>Where this leads to? AI gives an opportunity for non IT business to in-source their IT functions. Up to now that was typical for banking and insurance, now, it is time for other industries to explore new frontiers.</p><h4>A Hypothetical Shore Steel Industries Ltd.</h4><p>What are pros and cons for mid size industrial entity to switch over to in-sourcing. Let’s consider a hypothetical Shore Steel Industries Ltd., a European industrial manufacturer with 2,000 employees, several warehouses, and €250M annual revenue.</p><p>Today, they rely on SAP (sorry, SAP, nothing personal :)) for ERP, ADP for payroll, Specialized vendors for logistics and transport, Agencies for e-commerce and client portals, Their total IT vendor bill: roughly €1.7M per year.</p><p>Now imagine they decide to in-source. They hire a 5 person AI empowered team, capable of delivering with AI copilots, open-source frameworks, and low-code tools. Such team can rebuild an internal ERP ecosystem — unified across production, HR, logistics, and finance — within 12–18 months.</p><p>Annual cost: €750k (including infrastructure). Savings: nearly €1M per year. Payback period: less than 12 months.</p><p>Beyond the costs — the new agility is truly remarkable — process changes can now be implemented overnight instead of multi-month vendor cycles.</p><h4>The Risk Shift</h4><p>Sure, this model isn’t free of risks. In-sourcing increases dependency on internal talent, security responsibility, and continuity planning. But these risks are controllable. Documentation, automation, and cross-training mitigate most of them.</p><p>What’s eliminated, however, is vendor lock-in, slow change cycles, inflated license fees, and the cultural divide between “business” and “IT.”</p><h4>The Strategic Advantage</h4><p>When digital operations are in-house, feedback loops shorten. Data stays internal. Productivity increases — not through layoffs, but through ownership.</p><p>Over time, these in-house systems evolve into living digital organisms: constantly tuned by the people who use them. That is not just cost optimization — that is competitive differentiation.</p><h4>What This Means for the Industry</h4><p>For classical IT vendors and SaaS providers, this is the real disruption.</p><p>The future winners will be those who help businesses to build themselves, not those who keep selling black-box software or inflated consulting hours.</p><p>AI will not replace business systems. It will replace the distance between business and technology. And that will be enough to rewrite the entire IT services landscape.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7ae82e4dfd74" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Time has not to be linear]]></title>
            <link>https://medium.com/@erlihs/time-has-not-to-be-linear-c727d45ebf5e?source=rss-8b0cc5c24d98------2</link>
            <guid isPermaLink="false">https://medium.com/p/c727d45ebf5e</guid>
            <category><![CDATA[time-management]]></category>
            <category><![CDATA[agile]]></category>
            <category><![CDATA[life]]></category>
            <dc:creator><![CDATA[Jānis Erlihs]]></dc:creator>
            <pubDate>Sat, 04 Oct 2025 03:57:07 GMT</pubDate>
            <atom:updated>2025-10-04T03:57:07.505Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*DiD8eJVNZpbrIVpmQejdnw.jpeg" /></figure><p>I am old enough to remember when two year project where you had to do work breakdown and Gantt chart by weeks before start. Did not work out well.</p><p>I am young enough to be “agile“ where we do all work in two week sprints. Does not work out really well either.</p><p>So what still works?</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2F9LqK8GiIMYw%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3D9LqK8GiIMYw&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2F9LqK8GiIMYw%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/dc1f4214727ff5b1de827fec3f05f573/href">https://medium.com/media/dc1f4214727ff5b1de827fec3f05f573/href</a></iframe><h3>Non-linear time</h3><p><strong>Day — Week — Month — Quarter — Year — 5 years — Lifetime</strong></p><p>This is one of ways how thinking and planning works.</p><p><strong>Day</strong> gets very specific, doable in a day tasks. Like “write an article about non-linear time scales”.</p><p><strong>Week</strong> gets issues stacked up with ideas of articles that could be written that week — typically no more that one or two.</p><p><strong>Month </strong>— month gets a target topic — need to write 4 articles.</p><p><strong>Quarter</strong> gets even broader — just “keep writing articles to get reads up to 200 per month“.</p><p><strong>Year</strong> — “Enable paywall and get first revenue“</p><p>Re-planning? On daily basis. Task cards move and emerge and archive between Day and Week quite dynamically. Focus is kept to accomplish that one or two things each day. At the same time — purpose of doing daily tasks is not lost as quarter and year reminds what was the reason for this.</p><h4>So what still works?</h4><p>Not treating time as a straight line. Non-linear scale has many benefits — main of which is not to loose focus for “do it now” and keep up with “how to do it” in medium term, and up to “why even we are doing this” in a long run.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c727d45ebf5e" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>