Amvera, a PaaS cloud provider, has optimized its Python application build process to reduce deployment times. The company implemented a local mirror for Python packages to bypass PyPI download speed limitations, which previously dropped to 20–50 KB/s during peak loads. The mirror provides stable speeds of 300–400 MB/s.
The team integrated UV and UV PIP, which are written in Rust, to replace the standard pip. These tools enable parallel dependency installation and utilize a specialized cache. According to internal tests, this change resulted in a 7-fold increase in dependency installation speed.
To further reduce network latency, Amvera introduced a hot cache directly on build nodes for frequently used dependencies. The company also scaled its infrastructure by increasing the number of nodes and allocating more CPU and RAM resources to handle concurrent build volumes, which have reached peaks of 70 simultaneous processes.
These optimizations have significantly reduced total build times. For a test project using Streamlit, the build duration decreased from 18 minutes to 50 seconds, representing a 20-fold improvement. The company is now focusing on optimizing archive management and pod scheduling to further minimize total deployment time.
