From contributor

to main maintainer

of a Rust database driver

Pierre Zemb

@PierreZ

$ whoami

• Pierre Zemb
• Staff Engineer
• Leading our serverless database offer
• Mostly working around distributed systems
• OSS maintainer
• Mostly in Rust
• Involved in local communities such as
• Squash player

​

@PierreZ

$ whoami

• Pierre Zemb
• Staff Engineer
• Leading our serverless database offer
• Mostly working around distributed systems
• OSS maintainer
• Mostly in Rust
• Involved in local communities such as
• Squash player

​

@PierreZ

5 years ago…

• I was curious towards a distributed system called FoundationDB
• I was playing it with Java, and decided to test the driver in Rust and noticed that:
• It was a community-based bindings
• several features were missing
• So, I decided to contribute to it!
• Opened 3 Pull requests

@PierreZ

No response…

@PierreZ

Then, one day…

I received a mail from the former maintainer:

• no longer employed by Clikengo since March and now only have contributor rights
• the current president seems to ignore GitHub entirely
• He can still publish the crates on crates.io
• MIT/Apache 2.0 license also poses no legal concerns
• We should move the main repository elsewhere

@PierreZ

@PierreZ

@PierreZ

Today

• 16k lines of Rust
• More than 5M downloads total
• Around ~20k daily
• New release every ~9 months
• 30 contributors in total
• 1 main contributor
• Used by:
• companies like Datadog, Clever Cloud, OVHcloud
• Maybe Cloudflare, Tangia, SpiralDB, Molnett
• projects like
• Apache OpenDAL, SurrealDB,
• mvstore, and other hobby-grade projects

@PierreZ

How do I maintain the binding?

@PierreZ

How do I maintain the binding?

How I make sure that the bindings

is working properly?

🙈🙉🙊

@PierreZ

Rust, or the curse of FFI

• The Rust code is not a full database crate, it is communicating with a C-library that is handling everything
• The rust code is only a wrapper around some C code
• The good news is that the library is handling all the complicated stuff for us!
• The bad news is that you need to connect Rust and C code!

foundationdb-rs

Written in Rust

libfdb

Written in C

FoundationDB

FoundationDB

FoundationDB

@PierreZ

FFI and Bindgen

• so you need to:
• Generate the FFI(Foreign function interface) using bindgen
• Wrap the unsafe function
• test a lot!

foundationdb-rs

Written in Rust

libfdb

Written in C

@PierreZ

Dev experience

@PierreZ

We need a lot of CI

• We need to run tests on:
• On multiple OS(Ubuntu, macOS)
• On multiple FDB's versions(From FDB 6.1 to 7.3)
• On multiple Rust compiler version(MSRV, stable, beta, nightly)
• Nightly is super useful to find code that will not compile in the future!

@PierreZ

Testing is hard

• We still need to write a lot of tests 😓
• Tests hardly scale with features cardinality
• We need to generate workloads / scenarios ⚙️

@PierreZ

The BindingTester

• official tool to ensures language bindings interact correctly with FoundationDB
1. Tests transactions, atomic ops, error handling, tuple encoding
• Verifies uniform behavior across bindings
• Runs seeded generated tests

​

@PierreZ

How does it works?

@PierreZ

How does it works?

@PierreZ

How does it works?

@PierreZ

How does it works?

@PierreZ

Benefits of the BindingTester

• Running the bindingTester has been key to maintain the bindings
• Find the unknowns bugs
• The feedback loop is great, it finds bugs during dev
• We are running it on:
• scheduled CI
• every hour thousands of tests
• around 219 days of continuous testing each month
• major PRs
• Enforce 1k successful runs
• Official bindings runs more than 100k bindingtester runs every night

@PierreZ

Binding’s safety

libfdb

(tested by Apple)

@PierreZ

Binding’s safety

fdb-rs

(tested with bindingTester)

libfdb

(tested by Apple)

@PierreZ

Binding’s safety

user’s code

​

How can we test this part???

fdb-rs

(tested with bindingTester)

libfdb

(tested by Apple)

@PierreZ

FoundationDB simulation framework

• Before developing FoundationDB, they developed a deterministic simulation framework
• The goal is to find bugs during development rather than the real world.
• Simulates and inject failures into:
• I/O: Network and Disk,
• Time,
• Cluster topology,
• In a deterministic way
• perfect repeatability

​

​

​

​

@PierreZ

@PierreZ

What is trying to achieve

@PierreZ

Introduce random network partitions

@PierreZ

Reboot up to 10 machines, keep at least 3 running

@PierreZ

@PierreZ

Will happen concurrently

@PierreZ

@PierreZ

@PierreZ

FDB and Rust

🦀

@PierreZ

FDB and Rust

🦀

@PierreZ

The workload trait

@PierreZ

Open-sourced

@PierreZ

And helping the community

@PierreZ

Binding’s safety

libfdb

(tested by Apple)

@PierreZ

Binding’s safety

fdb-rs

(tested with bindingTester)

libfdb

(tested by Apple)

@PierreZ

Binding’s safety

user’s code

(tested with the simulation)

fdb-rs

(tested with bindingTester)

libfdb

(tested by Apple)

@PierreZ

We are hiring 😉

@PierreZ

Automating things around Github

• I cannot invest time every weeks to work on the crate, so I automated a lot of things
• CI with Github Actions
• On PR
• On a schedule
• bump dependencies with dependabot
• The release process with Release-plz
• Generate CHANGELOGs
• git tags
• github releases
• publish to crates.io

@PierreZ

The good and the bad

• I’m really happy to:
• maintain the crate that is used
• have included more safety and security
• learned A LOT
• Met people online and chat with them
• I had a good codebase to work with from former maintainers
• Managed to include work-time to maintain the library

• I waited too long to automate some stuff
• Some things are still missing:
• guide, documentations, tutorials…
• More maintainers
• To have more reviews
• More feedbacks from production

@PierreZ

Thanks!

Do you have any questions?

https://pierrezemb.fr

https://github.com/PierreZ

@PierreZ