The mirrorlist-server uses the data created by MirrorManager2 to answer client request for the "best" mirror.
This implementation of the mirrorlist-server is written in Rust. The original version of the mirrorlist-server was part of the MirrorManager2 repository and it is implemented using Python. While moving from Python2 to Python3 one of the problems was that the data exchange format (Python Pickle) did not support running the MirrorManager2 backend with Python2 and the mirrorlist frontend with Python3. To have a Pickle independent data exchange format protobuf was introduced. The first try to use protobuf in the python mirrorlist implementation required a lot more memory than the Pickle based implementation (3.5GB instead of 1.1GB). That is one of the reasons a new mirrorlist-server implementation was needed.
Another reason to rewrite the mirrorlist-server is its architecture. The Python based version requires the Apache HTTP server or something that can run the included wsgi. The wsgi talks over a socket to the actual mirrorlist-server. In Fedora's MirrorManager2 instance this runs in a container which runs behind HAProxy. This implementation in Rust directly uses a HTTP library to reduce the number of involved components.
In addition to being simpler this implementation also requires less memory than the Python version.
Another re-implementation of a part of MirrorManager2. generate-mirrorlist-cache talks to the MirrorManager2 database to create the input files for the previously mentioned mirrorlist-server. generate-mirrorlist-cache is a drop in replacement for the Python script mm2_refresh_mirrorlist_cache.
In Fedora's setup the Python version requires up to 50 minutes and 10GB of memory. This Rust version is finished in under one minute and only requires 600MB.
The project can be built using:
$ cargo buildThe mirrorlist-server requires multiple input files which all can be created using MirrorManager2 or generate-mirrorlist-cache.