Practical .NET and Software Architecture Tips | Microsoft MVP

Hi there! 👋

I'm a senior software engineer, primarily working in the .NET ecosystem.

Things I talk about on LinkedIn:
- C# and .NET ❤
- Software engineering and architecture 
- Distributed systems and how to build them
- Databases and optimization techniques
- Career and personal growth

One of my goals (i̶s̶) was to become a Microsoft MVP in Developer Technologies.
I became an MVP in March of 2023.

If this sounds interesting, let's connect!

Milan Jovanović

You don’t need a Dockerfile anymore.

Yes, really.

Modern .NET can publish your app directly as a container image using the SDK. No scaffolding. No boilerplate. No hand-written Dockerfiles.

Here's what you'll learn:
- How dotnet publish builds a Linux container
- How to run it locally with Docker
- How to push it to GitHub Container Registry
- Why this makes self-hosting .NET apps on a VPS much simpler

If you’ve ever copied a Dockerfile you didn’t fully understand, this is for you.

Check it out: https://lnkd.in/dUFzz3du

This is also the first step in a new series I’m working on around self-hosting .NET apps without overengineering.

No alternative text description for this image

AI in .NET is way easier than people think.

And you can start for free.

If you want a practical first project, build vector search:
- Turn text into embeddings (vectors)
- Store them in your vector store (pg_vector)
- Turn the user’s query into a vector
- Compare with cosine similarity to find the closest matches

That’s why it feels like “search by meaning” instead of keywords.

To do this cleanly in .NET, start with MEAI (Microsoft Extensions AI), the official library for LLM + embeddings.

Start here: https://lnkd.in/dkpadmVc

---

Do you want to simplify your development process? Grab my Clean Architecture template here and save 7 days of development time: https://lnkd.in/dfGYcyFC

undefined undefined

I think this makes me an Azure core developer now

/s

Should you stop using random UUIDs in .NET?

Traditional UUIDs (Guid.NewGuid()) are randomized. Great for uniqueness, but terrible for database performance.

They cause:

❌ Expensive index rebalancing
❌ Poor insert performance
❌ High fragmentation

That’s why many developers switched to ULIDs. It's 128-bit, sortable, and still unique.

They preserve temporal order, which makes them more index-friendly.

But now, there’s something new:

✅ Guid.CreateVersion7() in .NET 9+

UUID V7 is time-ordered and compatible with ULIDs, without needing a 3rd-party package.

You get the benefits of ULID, but natively supported in .NET.

Would you switch from ULID to V7 UUIDs?

---

Do you want to simplify your development process? Grab my Clean Architecture template here and save 7 days of development time: https://lnkd.in/dfGYcyFC

Not all Load Balancing strategies are created equal.

If you are using YARP (Yet Another Reverse Proxy) in .NET, you have 5 built-in ways to distribute traffic. But which one should you choose?

Here is the breakdown:

1. RoundRobin: The classic. It cycles through servers evenly. Simple, but it assumes all servers are equally fast.

2. LeastRequests: The smart choice. It sends traffic to the server currently doing the least work.

3. Random: It picks a server at random. Surprisingly effective for high-volume, uniform workloads.

4. PowerOfTwoChoices: The balanced approach. It picks two random servers and chooses the one with fewer requests. It avoids the overhead of checking every server.

5. FirstAlphabetical: The "sticky" one. It always picks the first available server. Good for failover setups, bad for load distribution.

Most people stick with RoundRobin, but switching to LeastRequests can significantly improve tail latency in systems where request times vary.

I dive deeper into how to scale horizontally using YARP here: https://lnkd.in/daDNMpmB

Which strategy is your default choice?

---
Sign up for the .NET Weekly with 75K+ other engineers, and get a free Clean Architecture template: https://lnkd.in/dcvJmi7E

Stop calling DateTime.UtcNow in business logic.

This line looks harmless... and it’s the reason your tests are a mess.

Here’s the trap:

- DateTime.UtcNow > user.SubscriptionEnd

Seems harmless, right?

But the moment you try to test this logic, you realize the problem:

- You can’t control DateTime.UtcNow
- Your tests become flaky or complicated
- Business rules are tied to infrastructure

There are better ways:

1. Inject DateTime as a parameter — simplest, great for small methods

2. Introduce an IClock interface — more scalable, and test-friendly

By inverting the dependency on time, you make your domain logic:

✅ Easier to test
✅ More deterministic
✅ Separation of concerns

Next time you write a time-based condition, stop and ask:

“Will I need to test this?”

If the answer is yes, extract the clock.

---
Do you want to simplify your development process? Grab my Clean Architecture template here and save 7 days of development time: https://lnkd.in/dfGYcyFC

The best HttpClient config is the one you don’t repeat in every service.

In .NET, you set it up once with IHttpClientFactory (and Aspire if you’re using it):
- Discovery for service-to-service calls
- Resilience defaults so transient failures don’t hurt you

Then you just inject the client and write business code.

Want the full walkthrough?

Start here: https://lnkd.in/dQNxjHix

---

Do you want to simplify your development process? Grab my Clean Architecture template here and save 7 days of development time: https://lnkd.in/dfGYcyFC

Here are 9 excellent libraries I use in my projects:

1. YARP
2. Polly
3. EF Core
4. Refit
5. Testcontainers
6. OpenTelemetry
7. FluentValidation
8. xUnit
9. Dapper

Which library would you add to the list?

---
Sign up for the .NET Weekly with 76K+ other engineers, and get a free Clean Architecture template: https://lnkd.in/dbU3m97Q

Stop trying to force Domain-Driven Design into every feature.

Sometimes, the best architecture is the simplest one: The Transaction Script.

It’s a classic pattern from Martin Fowler, and it fits 80% of CRUD scenarios perfectly.

Instead of building complex aggregates and value objects, you just write a simple, procedural script that handles a single request:

1. 𝗩𝗮𝗹𝗶𝗱𝗮𝘁𝗲 the input data.
2. 𝗖𝗮𝗹𝗰𝘂𝗹𝗮𝘁𝗲 the result (business logic).
3. 𝗣𝗲𝗿𝘀𝗶𝘀𝘁 the changes to the database.
4. 𝗥𝗲𝘁𝘂𝗿𝗻 the response.

It is fast, readable, and easy to test.

I wrote an article showing a side-by-side comparison of this vs. a Rich Domain Model.

See the code here: https://lnkd.in/d9dwgpdq

👉 When to switch: Start with a Transaction Script. Only refactor to a Domain Model when the business logic becomes "chatty", meaning you have too many if/else blocks or complicated state transitions that are hard to track in a single method.

---
Sign up for the .NET Weekly with 76K+ other engineers, and get a free Clean Architecture template: https://lnkd.in/dbU3m97Q

In .NET, you usually have to choose between Speed and Consistency.

In-Memory Caching (IMemoryCache):
✅ Blazing fast (nanoseconds).
❌ Data is stuck on one server. If you have 10 nodes, you have 10 caches.

Distributed Caching (IDistributedCache / Redis):
✅ Shared across all servers. Consistent data.
❌ Network latency. Serialization costs. It's slower.

The Solution: HybridCache.

Don't choose one. Use both.

A Hybrid Cache creates a tiered architecture:

- L1 (Local): The app checks RAM first. If it's there, return instantly.
- L2 (Distributed): If missing locally, check Redis.
- Source: If missing in Redis, fetch from DB, then populate both L1 and L2.

While Microsoft is working on a native implementation, I still swear by FusionCache as the most robust, battle-tested library for this.

I wrote a guide on how to set this up properly: https://lnkd.in/d6NKgMVp

Are you currently using just Redis, or a multi-layer cache?

---
Sign up for the .NET Weekly with 75K+ other engineers, and get a free Clean Architecture template: https://lnkd.in/dCMpeCNe

Most OAuth demos are useless in the real world.

They skip the part that matters: how public clients should actually get tokens.

I just published a step-by-step video on running Keycloak in Docker + OAuth 2.0 Authorization Code flow with PKCE.

What you’ll see:
- Keycloak (latest) running locally with Docker/Compose
- Realm + user + public client setup
- Using .well-known/openid-configuration so you don’t guess endpoints
- Full PKCE flow in Postman (authorize → code → token exchange)
- Quick JWT inspection + the “why is this 401/403?” scope moment

If you’re building SPAs or mobile apps, PKCE isn’t optional. It’s the safe default.

Watch it here: https://lnkd.in/deZWiEbi

The worst kind of bug is the one that explodes 3 days after deployment.

This is the hidden danger of the standard Options Pattern in .NET.

You might bind your appsettings.json to a class, inject it, and everything looks fine. The app starts up perfectly.

But if a required API Key is missing? You won't know until a user triggers that specific feature, and the app crashes.

This is why "Fail Fast" is crucial for configuration.

If your config is invalid, your application should refuse to start. Period.

You can achieve this by extending the Options Pattern with IValidateOptions.

Instead of just binding data, you attach validation logic:

1. Define rules (e.g., "ApiKey cannot be null", "RetryCount must be > 0").
2. Register the validator.
3. If the config breaks the rules, the DI container throws an exception immediately at startup.

I took this a step further and integrated FluentValidation to make the rules cleaner and more expressive.

Here is the complete implementation code: https://lnkd.in/djz5Yfux

---
Sign up for the .NET Weekly with 75K+ other engineers, and get a free Clean Architecture template: https://lnkd.in/dcvJmi7E

Hidden danger of Options Pattern in .NET

I failed a .NET interview question that looked easy.

“Add caching to an async API.”

The tricky part isn’t caching.
It’s what happens when 50 requests hit the same cold cache at once:
- race conditions
- cache stampedes (thundering herd)
- why ConcurrentDictionary still isn’t enough
- why you can’t await inside a lock
- the SemaphoreSlim async-lock pattern (with try/finally, double-check, timeout)
- scaling it with per-key locks to reduce contention

I break it down step-by-step using a currency conversion example, and show the exact refactor path from “works locally” to “works under load.”

Watch it here: https://lnkd.in/dyN7_acX

If you need to run a task "Every Friday at 5 PM," how do you implement it?

A While(true) loop with a Thread.Sleep() is not the answer.

You need a robust scheduler that understands Cron Expressions.

A Cron expression allows you to define complex recurring schedules with a compact string.

For example:

• 15 10 * * * = 10:15 AM every day.
• 15 10 * * MON-FRI = 10:15 AM, but only on weekdays.

In the .NET ecosystem, the heavyweight champion for this is Quartz.

It parses these expressions for you and ensures your background services fire exactly when they are supposed to.

I broke down how to configure your first job and trigger in this article: https://lnkd.in/dfy-U7jp

What library do you currently use for background jobs?

---
Sign up for the .NET Weekly with 75K+ other engineers, and get a free Clean Architecture template: https://lnkd.in/dJGs8Uzv

Stop wasting the first 3 days of every project setting up boilerplate.

I built a free, production-ready Clean Architecture template to handle the plumbing for you.

It's already used by 40,000+ developers.

Here is what's inside the latest version:

- 𝗨𝗽𝗱𝗮𝘁𝗲𝗱 𝘁𝗼 .𝗡𝗘𝗧 𝟭𝟬
- 𝗡𝗲𝘄 .𝘀𝗹𝗻𝘅 𝗦𝗼𝗹𝘂𝘁𝗶𝗼𝗻 𝗙𝗼𝗿𝗺𝗮𝘁
- CQRS (No MediatR dependency)
- Structured Logging & Validation
- JWT Authentication & RBAC
- Docker Compose support

It saves you hours of setup time.

Download it for free here: https://lnkd.in/dbU3m97Q

💡 Why the shift to .slnx? The new Solution format in .NET is a game-changer. It replaces the old bloated syntax with a clean XML syntax. It makes merge conflicts in your solution file a thing of the past.

AI won’t replace you in 2026.

But a developer who knows how to engineer with AI absolutely will.

I recorded a new video on how I’m using AI tooling for .NET development (mainly Cursor), and the point isn’t Cursor.

It’s the workflow.

Here’s what I do every time:
- Plan first (write the steps before touching code)
- Give AI tight constraints (what to change, what NOT to change)
- Review like it’s a PR (DI lifetimes, packages, SQL correctness, connection handling)
- Verify (build + tests) so you don’t ship “confidently wrong” code

In the video, I walk through a real refactor: moving query handlers from EF Core to raw SQL/Dapper, and show the exact spots where AI gets you 90% there... and where that last 10% can break things.

You can watch it here: https://lnkd.in/dzZiKR66

If you’re trying to “stay relevant” in the AI era, the skill isn’t prompting.

It’s owning correctness.

Data Annotations are great for simple things.

[Required]? Fine.
[MaxLength(50)]? Sure.

But what happens when you need to validate something against a database?
Or validate based on a configuration setting in appsettings.json?

Data Annotations hit a wall. You can't easily inject dependencies into an Attribute.

This is where FluentValidation shines.

Because your validators are just standard classes registered in the DI container, you can inject anything you need.

Want to check if a CouponCode is valid by calling the IPricingService? Just inject it into the OrderValidator constructor.

It turns validation from a static check into a dynamic, intelligent logic layer.

I explain how to wire this up seamlessly in ASPNET Core here: https://lnkd.in/d9nVR8fu

It is a game changer for complex business rules.

---
Sign up for the .NET Weekly with 75K+ other engineers, and get a free Clean Architecture template: https://lnkd.in/dCMpeCNe

Still not using Central Package Management in .NET?

Open your .csproj file right now.

Is it full of clutter?

<PackageReference Include="Serilog" Version="4.3.0" />
<PackageReference Include="WolverineFx" Version="5.6.0" />

Now imagine if it looked like this:

<PackageReference Include="Serilog" />
<PackageReference Include="WolverineFx" />

This is the beauty of Central Package Management.

It cleans up your project files. It stops developers from accidentally installing a "rogue version" of a library that doesn't match the rest of the team.

You control everything from a central Directory.Packages.props file at the root of your repository.

You can even force all projects to use the exact same version of a library, ensuring your build is deterministic.

I have a tutorial on how to enforce this structure here: https://lnkd.in/dGDhDu2T

Do you prefer editing XML directly or using the Visual Studio UI?

---
Sign up for the .NET Weekly with 75K+ other engineers, and get a free Clean Architecture template: https://lnkd.in/dCMpeCNe

ValueFirst

Dot net developer-Technical Lead

Idea Elan

Dot net developer

Fortune Marketing Private Limited

Back End Coding Using C#.Net, VB.Net and ASP.Net,MS Sql Server

Software Developer

HKG Rox Infotech Pvt. Ltd.

Web Developer(ASP.NET)

Having good  experience in Microsoft Technologies like .NET frame work , ASP.NET,MVC, WEB API , .Net Core Web API,ADO.NET, SQL Server,Oracle and associated technologies like HTML, JavaScript,Jquery, nodejs etc.

taplio

The Battle

Milan Jovanović

Noor Alam

Followers

Posts

Engagements

Likes

Performance ScoreMilan Jovanović's LinkedIn Performance

Ready to grow your LinkedIn presence?

taplio

The Battle

Milan Jovanović

Noor Alam

Followers

Posts

Engagements

Likes

Performance Score.css-goub6s{font-size:var(--chakra-fontSizes-md);font-weight:500;color:var(--chakra-colors-gray-600);display:block;margin-top:var(--chakra-space-1);}Milan Jovanović's LinkedIn Performance

Ready to grow your LinkedIn presence?

Performance ScoreMilan Jovanović's LinkedIn Performance