If you are unfamiliar with threads in C++ I recommend you take a look at my article about threads before reading this one.

Futures

Futures are a concept of concurrent programming. A future is a way to access a result from an asynchronous operation. As a simple example:

1
2
std::future<int> fut = functionThatReturnsFuture();
int val = fut.get();

The code above should be very easy to read. We execute a function and it returns a future. Then we use this future to get a value. The interesting thing about this code is that functionThatReturnsFuture can be (and most likely is) an asynchronous operation. When we call fut.get(), our code will wait for that asynchronous operation to complete. When the operation completes, it will return an int value that will then be assigned to val.

I’m getting started with concurrency in C++ and threads seem to be a good way to get familiar with the basics.

Processes

When a user starts executing a program, the program becomes a process. A process is a set of instructions (the code) and state (memory, registers) that is managed by the operating system. The operating system tells the processor which process it should be running.

In a system with a single core, only one process can be executed at one point in time. Since there are a lot of processes running in a modern system, the operating system will take care of deciding which processes should be serviced by the CPU.

Processes can create other processes by using Fork. When a process forks, they become independent of each other. They own their own instructions and state.

A few days ago I was listening to a talk about the new features of C++ and I heard the presenter mention autools. I felt pretty dumb not knowing what he was talking about so I’m writing this post to make me feel less dumb.

Make

It all started with Stuart Feldman’s make. Make is a tool that generates files based on other files. In a makefile you can specify a list of files you want to generate and how to generate those files (based on other files). The most common use for make is to generate executables based on source code.

Make allows people to generate executables easily based on source code. People in possession of the code don’t need to know the steps to build the executable because these steps are already recorded in the makefile. Another advantage of make is that it allows for faster builds by keeping tracks of source files that haven’t changed since last time a build was run and skipping unnecessary steps. This is specially useful for large codebases that take long time to compile.

Writing a post about types in most high level languages I’m used to wouldn’t be very interesting, but I’ve recently started learning C++ and I realized that I need to understand memory a little better to be able to write and read C++ programs effectively.

There is a collection of types that should not cause many surprises. .

Integer types

Floating point types

*The sizes above are for GNU C compiler, but might vary for different compilers

*All numeric types can be modified with the unsigned keyword. This affects the minimum and maximum possible values that can be hold and the way arithmetic operations work on them, but not their size in bytes.

I’m writing C++ in the title of this article because I’m currently in a journey to learn C++. I believe the same concepts apply to C.

Writing about C++ is a little harder than writing about other languages because I keep stumbling into circular references where I need to understand A in order to understand B, but it’s very hard to understand A without understanding B.

I’m going to try to start with this article where I’ll explain why C++ has header files (files with .h extension) and how to use them.

Code separation

Before we start looking into header files, lets first look at how code is split and included in languages where there are no header files. This little example is in node.js:

In previous posts I wrote a little about distributed systems and the Raft algorithm. Today I’m going to look at one distributed key-value store that uses the Raft algorithm to achieve consistency and high availability.

From a client’s perspective, etcd will behave like any other key value store out there. It’s use of Raft underneath will make sure that there is only one leader at a given time and that the log is replicated to all nodes.

Getting ready

For this exercise I’m going to create a 5-node cluster, but before we start there are a few things we need to decide.

By default each etcd nodes uses port 2380 for communicating with clients and port 2379 for server to server communication. We will keep this default behavior.

Each node in the cluster needs to be able to communicate with the rest of the nodes in the cluster. The number of nodes in the cluster and their location needs to be configured for the cluster to be able to do some work.

In normal conditions we would have each node in a different host with a different IP Address. This would allow us to say something like: You can find node A at 10.10.10.2.

Running the cluster in a single machine makes things challenging because they would all be sharing the same IP address. To walk around this issue, we will create our own docker network and work within this network.

In a past post I wrote about distributed systems, but I intentionally omitted the subject of leaderless replication since I consider it a topic that deserves a little more focus.

In this post I’m going to explore how a leaderless system works and explain a little about how the Raft algorithm helps us achieve consensus.

Leaderless replication

As the name suggests, there are no leaders (or masters) in a leaderless setup. In this configuration all instances can receive reads and writes. If you read my previous post, you might be thinking that this sounds like master-master replication, but it is actually very different.

I mentioned two main problems in a master-master setup: Replication lag when you write to one master and then read from the other, and conflicts when you modify the same record in both masters and they try to sync. Leaderless replication doesn’t have these problems (it has others that I’ll explore soon). On top of not having those problems, a leaderless system can stay up even when instances are down (like the master-master configuration).

In computer science a distributed system is a software system in which different parts of it communicate by passing messages through a network. The different parts could be running in the same machine or distributed across the globe; as long as they communicate through an unreliable channel (a network), we can classify them as distributed and consider the challenges that come with it.

With this definition in mind, we could think about many examples of distributed systems. A single monolithic application communicating with a database could be considered a distributed system if the application communicates through a network protocol.

Although in practice, local networks can be pretty reliable, they are still vulnerable. There are two condition that can cause a plethora of problems to a system: the network being down or the network being slow. These two conditions can put the system in a wide variety of states that may give results we don’t expect.

Before we look at how these problems can affect a distributed system, lets look at a distributed system whose failure mode is mostly understood and accepted to this date: a stateless system.

I want to start using Kubernetes to manage my services. Before I go all-in I will do some playing around with their easy-to-install version that can run on a single machine.

Requirements

Before we can run Minikube locally we need to have a virtualization solution installed in our host. Since my computer supports KVM, I decided to go with it. You can check if your computer supports KVM with this command:

1
egrep -c '(vmx|svm)' /proc/cpuinfo

In a previous post I talked a little about creating installable web apps. One important thing that I didn’t cover is how to automatically prompt a user on your web app to add it to their home screen.

Serve over https

One of the requirements for chrome to show an app install banner is that the app is served over HTTPS. This can be done in a few ways depending on your server. I wrote a post on how to get free SSL certificates using Let’s encrypt that should help you get started.