Category: Architecture

Grant Fritchey is in rare form today:

Go and search through it for a NoSQL data management system. I’ll wait.

You found one that was named NoSQL didn’t you. Oracle NoSQL, because Oracle. Of course. However, under the Database Model what did it say? Document Store. Why?

BECAUSE NOSQL IS NOT A DATA STORAGE ENGINE!

There is not a NoSQL thing to use. You can’t compare NoSQL to anything because NoSQL is just a different screed, a rant, a tantrum. A bunch of influential developers threw themselves on the floor, screaming and kicking, spittle flying everywhere, because they didn’t want to eat their broccoli.

Wait, that was my kids when they were three.

As the voice of reason (and you know you’re in trouble when that’s the case!), non-relational engines take up 6 of the top 15 slots.  There are particular advantages to non-relational data stores—as Grant willingly notes—and as companies grow, specialized data storage can become quite useful.  But relational databases are a great starting point and for good reason:  they solve a subset of problems extremely well and solve most problems well enough.  This is probably also a good place to drop in a reference to Feasel’s Law.

Sylvia Fronczak defends the layered architecture approach to software development:

There are disadvantages to the layered architecture approach. And some aspects of it have been deemed not architecturally pure for domain-driven design.

However, there are disadvantages to most approaches. The key is to understand both the advantages and disadvantages. Pick what architectural patterns work best for your particular problem.

Here are some factors you want to keep in mind.

1. The layered architecture is very database-centric. As mentioned before, since everything flows down, it’s often the database that’s the last layer. Critics of this architecture point out that your application isn’t about storing data; it’s about solving a business problem. However, when so many applications are simple CRUD apps, maybe the database is more than just a secondary player.

2. Scaleability can be difficult with a layered architecture. This is tied to the fact that many layered applications tend to take on monolithic properties. If you need to scale your app, you have to scale the whole app! However, that doesn’t mean your layered application has to be a monolith. Once it becomes large enough, it’s time to split it out—just like you would with any other architecture.

3. A layered application is harder to evolve, as changes in requirements will often touch all layers.

4. A layered architecture is deployed as a whole. That’s even if it’s modular and separated into good components and namespaces. But that might not be a bad thing. Unless you have separate teams working on different parts of the application, deploying all at once isn’t the worst thing you can do.

I completely disagree with critique #1 (as opposed to Sylvia’s defense):  most of the time, the database will outlive that puny application by a matter of decades.  That is, when some team is developing version 8 of the application, they’ll use essentially the same database as in V1.  Creating a design which optimizes for data storage and recall is, in general, much more important than which Javascript library you’re using or how you lay out those files.

Samir Behara has some tips for designing resilient microservices:

While architecting distributed cloud applications, you should assume that failures will happen and design your applications for resiliency. A Microservice ecosystem is going to fail at some point or the other and hence you need to learn embracing failures. Don’t design systems with the assumption that its going to be sunny throughout the year. Be realistic and account for the chances of having rain, snow, thunderstorms and other adverse conditions. In short, design your microservices with failure in mind. Things don’t go as per plan always and you need to be prepared for the worst case scenario.

If Service A calls Service B which in turn calls Service C, what happens when Service B is down? What is your fallback plan in such a scenario?

• Can you return a pre-decided error message to the user?

• Can you call another service to fetch the information?

• Can you return values from cache instead?

• Can you return a default value?

Microservices have their drawbacks, but one big advantage is that they tend to be concise enough that you can reason about them more clearly than kitchen sink applications.  Planning ahead on potential failure modalities differentiates flaky services from robust services.

Deborah Melkin explains the supertype-subtype pattern in relational database architecture:

You don’t see a supertype-subtype relationship defined as such when you’re looking at the physical database. You’ll only see it explicitly in the logical data model. So what is the pattern and how do you know that you have one in your database?

This relationship exists where you have one entity that could have different attributes based on a discriminator type. One example is a person. Depending on the role of that person in relationship to the business, you will need to store different pieces of information for them. You need different information about a client than you do an employee. But you’re dealing with a person so there is shared information.

It’s a good pattern for minimizing data repetition.

Anshulee Asthana shares some thoughts after researching GraphQL implementation details:

The Existing System

• SQL Server DB with SPs, functions, views, and queries. SPs are non-modular in the sense they join various tables and return values specific to the calls being made.
• Connection with the DB is using basic ADO.NET and not with EF/LINQ, etc.
• ASP.NET WebAPI2.0 (Not .NET Core) and Java-based APIs.
• Native Android, iOS and Web Clients. Some portion of the web clients bypasses the API and talks to the DB directly.
• WebClients: ASP.NET and React.
• A React Native-based app.

System Maturity

• System has been in production for few years now and is stable.
• As the system is in a competitive space, new features are always getting added apart from usual maintenance.

Customer Ask

• Whether it makes sense to wrap our existing APIs into a GraphQL Endpoint.

• For a new feature in the react app evaluate making the new .NET based APIs in GraphQL.

It’s nice that Anshulee shared this case study, especially because there’s relatively little involving GraphQL + .NET.

Lukas Eder explains why database-first design patterns tend to work better than code-first design:

The real “truth” of your database schema, and the “sovereignty” over it, resides with your database. The database is the only place where the schema is defined, and all clients have a copy of the database schema, not vice versa. The data is in your database, not in your client, so it makes perfect sense to enforce the schema and its integrity in the database, right where the data is.

This is old wisdom, nothing new. Primary and unique keys are good. Foreign keys are good. Check constraints are good. Assertions (when they’re finally implemented) are good.

And that’s not where it ends. For instance, if you’re using Oracle, you may want to specify:

• In what tablespace your table resides
• What PCTFREE value it has
• What the cache size of your sequence (behind the identity) is

Maybe, all of this doesn’t matter in small systems, but you don’t have to go “big data” before you can profit from vendor-specific storage optimisations as the above. None of the ORMs I’ve ever seen (including jOOQ) will allow you to use the full set of DDL options that you may want to use on your database. ORMs offer some tools to help you write DDL.

But ultimately, a well-designed schema is hand written in DDL. All generated DDL is only an approximation of that.

It’s a great post.  Also check out Lukas’s responses in the comments section.

Thomas Henson has a taxonomy of non-relational databases:

Columnar Database

The first type of NoSQL database is the Columnar databases which is optimized for reading and writing columns of data as opposed to rows of data. Column-oriented storage for database tables is an help drive down the input/output requirements for database. Since the I/O profile is lowered, overall storage footprint is lowered. One main feature of Columnar Databases is their ability to compress data. Instead of data being written in traditional row orientation, Columnar databases use column orientation. Each column will be associated with column key. Checkout this example from my HBase Blog Post.

He then goes on to describe the other three types.  I agree with the taxonomy he uses.