iiinova — where your digital self keeps evolving

Replies

Maxim Sakhno @maxim-sakhno.bsky.social · 28/01/2026

Is #LiteStar faster than #FastAPI? Benchmarks say yes, but look at Dependency Injection. 📉 Adding just 2 flat dependencies significantly degrades the throughput. I dug into the source to find why. Thread: TaskGroups, Kahn's algorithm & performance trade-offs. 🧵 #Python #Backend #WebDev

A bar chart titled "LiteStar" showing Request Per Second (RPS) performance across different numbers of dependencies (0 to 5). The RPS drops from 41,451 with 1 dependency to 26,492 with 2 dependencies. A red arrow points to this 36% decrease with a handwritten caption asking "Why?".

362

Maxim Sakhno @maxim-sakhno.bsky.social · 28/01/2026

The scenario: simple flat dependencies. No nesting, just direct injection. Below is the LiteStar setup and the dependency graph it builds behind the scenes. 👇

On the left, LiteStar code defines two simple async dependencies and two handlers: 'handler1' using one dependency and 'handler2' using two. On the right, a hand-drawn schematic visualizes the mapping: 'handler1' connects to 'dep0', while 'handler2' connects to both 'dep0' and 'dep1'.

110

Maxim Sakhno @maxim-sakhno.bsky.social · 28/01/2026

Measuring RPS for LiteStar and FastAPI using wrk. Quite a twist! 📉 LiteStar leads at 0-1 deps, but falls behind once you add a second one. While FastAPI degrades linearly, LiteStar hits a performance cliff. The culprit is hidden in the source code. Let's find it. 🕵️‍♂️

A bar chart titled "FastAPI" showing Request Per Second (RPS) performance across different numbers of dependencies (0 to 5). The RPS starts at 35,902 (0 deps) and decreases steadily to 25,751 (5 deps).

120