The Bottom Line After 6 Weeks With Rust
I previously wrote about my plan to rewrite ReductStore in Rust. I am happy to announce that the migration is now complete. It was a very interesting 6 weeks for me, and I would like to share my experience with you.
Project Design before Migration
I wrote ReductStore in C++20, utilizing coroutines and ranges.
For the HTTP frontend, I used uWebSockets as an HTTP server and its event loop for coroutines.
The storage engine was implemented from scratch.
I used Protobuf as a JSON and binary serializer in both the HTTP frontend and the storage engine. Many of the structures were shared between the two.
I managed dependencies with Conan and used CMake as a build system.
Codebase about 20k lines with unit tests.
About Me
I have been developing in C++ and Python for about five years. Mostly, I write services for data acquisition, processing, and storage. I like OOP, design patterns, and my C++ smells Java. However, I avoid using exceptions and follow the RAII approach.
In Rust, I have only written the "Hello, World" example.
I work on the project in my spare time 10-20 hours a week.
Migration
Initially, I planned to use the cxx.rs library and rewrite the project in small steps by wrapping Rust code and integrating it into C++.
[SPOILER] I wasn't able to handle this...
Step 1. Rewrite util functions and classes
ReductStore has its own simple logger and uses environment variables for configuration. The code was trivial and it was good to start with Rust.
Here, I learned the basics of Rust memory management, borrowing, and references. All these concepts I already knew from C++, so it was easy to understand. However, rustc did not want to compile my code in the beginning.
The biggest problem was that cxx.rs throws an exception when a Rust function returns an Err. It did
not work for me because I did not use exceptions and my C++ functions returned a structure with a result or an error. I
had to write additional wrappers to work around it. This was a very buggy layer and a source of frustration for me.
Step 2. Authentication module
ReductStore offers token authentication and an API for token management. I implemented it in C++ using OOP patterns such as Repository and Strategy. This was a great opportunity for me to learn Rust's traits and dynamic polymorphism.
During the process, I also discovered prost. In comparison with the C++ implementation and especially the Python
implementation by Google, prost is amazing. However, it seems like magic to me, mostly because I don't understand how
it works in detail.
Although I was able to integrate the new authentication module in C++, the cost was too high. As a result, I decided to move forward with the next steps without C++ integration.
Step 3. Storage engine
The storage engine was designed to store records in pre-allocated blocks and provide parallel read/write streaming to the same block. The engine can also be configured to remove old blocks to keep a bucket quota. Thus, it should keep track of all readers and writers and remove a block only when it is no longer in use.
Next, it was time to learn about Rust's smart pointers. My background in C++ helped me a lot here. However, I was a bit
surprised that I had to use nested pointers for mutable access (Rc<RefCell<T>>) instead of MutRc<T>.
Although the storage engine was the biggest part of the project, it didn't take too long to rewrite because it only uses a file system and Protobuf. Thus, I didn't have to learn many new things.
Step 4. HTTP fronted
Well... here, I broke my neck. The C++ implementation of the HTTP layer was asynchronous, so I wanted the same in Rust. I first tried to use hyper, but unfortunately, I couldn't make it work with JSON and streaming bodies in the same service. I realized that I had to learn a lot about hyper, which doesn't have rich documentation, before starting to use it.
Then I tried axum and it worked. The framework is really awesome, and I felt as if I was using Python Flask or Fast API.
However, I later discovered that my Rc<RefCell> wasn't thread-safe and had to refactor the storage engine. 😕
Step 5. Remove main.cc and CMake lists
That was the easiest part. I like using C++, but I hate its build systems. Therefore, I felt a bit sorry when I deleted the last CMakeLists.txt.
After migration
Now, ReductStore is written in Rust and developed as a Rust project. I'm planning to write a client SDK and rewritte Reduct CLI in Rust. I also would like to provide the storage engine separated from the HTTP backend as a feature, so that you can use it in you Rust applications if you need keep a history of data.
The project is open under the Business Source License (BUSL-1.1) and contributions are welcome.
TL;DR
If you're a C++ developer you shouldn't have any problems to learn Rust. However, if you decided to rewrite you C++ codebase, it isn't so easy. Cxx.rs library is useful, but the glue code is ugly and buggy. For me it would make sense to rewrite my code in one step