whoami

I am Uddeshya Singh (X, Medium, LinkedIn), SDE-2 Backend at Gojek. I specialize in payments in my 9-5 and have been an avid distributed systems + database geek.

Abstract

In distributed systems, ensuring reliability and fault tolerance is of utmost concern. Traditional heartbeat failure detectors often fall short in accurately assessing the health of processes, especially in environments with transient failures or network partitions. This is where Phi Accrual Failure Detectors (φ-failure detectors) come into play, offering a more nuanced and adaptive approach to failure detection.

Rust is an amazing choice for distributed systems so I decided to try and build a crate for the same, which is pluggable directly into your raft / paxos / toy redis systems. The result? crates.io/phi-accrual-detector.

Top takeaways from the talk

What would the audience learn after this talk?

• Building an intuition regarding why heartbeats might not be the best usecase always for failure detection.
• What makes a failure detector amazing?
• Comparing various detection algorithms and finally elaborating on φ-failure detectors.
• Code walkthrough for implementation of the library and test run results.