A simple toy payment processor.
To profile memory usage on Linux with heaptrack, run
cargo make heaptrack [1K|10K]To profile memory usage on macOS with Instruments, install cargo-instruments and run
cargo make instruments [1K|10K]In parsing a CSV file containing different types of records, such as transactions and dispute events, using separate
structs for each type, united into an enum (see AccountActivity), offers significant
advantages over a single struct with optional fields.
Although the csv crate doesn't natively support parsing into this type of data structure, the effort to adapt the deserializer is worthwhile. It enhances safety, reduces runtime errors, and improves code clarity in the long term.
Separate structs ensure that each record type only contains its relevant fields. A transaction struct will always have fields transaction id, client id and amount, whereas a dispute event will only include transaction id and client id. This enforces clear, type-safe invariants, preventing errors caused by missing or irrelevant fields.
The Rust compiler can enforce the correctness of the data structures, eliminating the need for checking Options and
reducing runtime errors. Each record type is guaranteed to have only the fields it needs, providing strong compile-time
guarantees.
As no specific performance target has been set, the processor is primarily optimized for robustness and convenience. However, performance has still been considered where appropriate to avoid missing obvious optimizations.
As expected, the slowest part of data processing is CSV file parsing, primarily due to I/O waiting times. The commonly
used csv crate does not support asynchronous file reading, limiting optimization potential in
this area. While alternative crates with async support exist, they require further evaluation before adoption.
Although benchmarks indicate that the use of the Decimal type of the
rust_decimal crate results in approximately a 20% performance decrease, its
benefits make it a sensible choice for financial calculations.
The crate ensures there are no rounding errors, which is crucial when dealing with financial data. Additionally, common issues associated with floating point types, such as NaN, infinite, or subnormal values, are avoided since decimals inherently cannot represent these states, eliminating the need for additional checks.
As above, since no specific performance target has been set, there is no strong justification for exploring alternatives
that rely on native floating-point types. The advantages provided by rust_decimal, particularly
in terms of precision and error avoidance, outweigh the potential performance gains from using floats.