RPMeta Service Documentation

RPMeta Service Documentation

RPMeta provides a systemd service for running the prediction server as a background service. This document describes how to install, configure, and manage the service.

Installation

Disclaimer: The service is shipped only to Fedora.

The RPMeta server service is installed as part of the rpmeta-server package:

sudo dnf install rpmeta-server

Configuration

Service Configuration

The service uses a configuration file located at /etc/rpmeta/config.toml. You can create or modify this file to customise the API server settings:

# Create or edit the configuration file
sudo vi /etc/rpmeta/config.toml

Example configuration options:

[api]
# Host to bind the server to
# Use "localhost" for local access only, "0.0.0.0" to allow external connections
host = "localhost"
# Port to run the server on
port = 9876
# Enable debug mode for more verbose output
debug = false

[logging]
# Log format
format = "[%(asctime)s] [%(levelname)s] [%(name)s]: %(message)s"
# Date format for logs
datefmt = "%Y-%m-%d %H:%M:%S"
# Log file location
file = "/var/log/rpmeta.log"

The service also needs to specify the model data (path, name, categories) and the configuration file. This is configured via an environment file located at /etc/sysconfig/rpmeta.

Model Files

The service expects model files to be installed in /usr/share/rpmeta/models/ or in the environment file:

• The model files themselves (trained using the rpmeta-trainer package)
• A categories.json file containing category data types

To use your own trained model:

1. Train a model using the rpmeta trainer command
2. Copy the model and categories.json files to the appropriate location:

sudo mkdir -p /usr/share/rpmeta/models
sudo cp /path/to/your/models /usr/share/rpmeta/models/
sudo cp /path/to/your/categories.json /usr/share/rpmeta/models/

Environment file

This file allows you to customise the service parameters without directly modifying the service file.. The environment file contains these defaults:

Parameter	Description	Default Value
MODEL_DIR	Path to the model directory	/usr/share/rpmeta/models
MODEL_NAME	Name of the model to use	lightgbm
CATEGORIES	Path to categories file	/usr/share/rpmeta/models/categories.json
CONFIG	Path to configuration file	/etc/rpmeta/config.toml
EXTRA_OPTS	Additional options to pass to the service	(empty)

Managing the Service

Works as you are used to a typical systemd service.

Testing the Service

Once the service is running, you can test it by sending a prediction request:

curl -X POST http://localhost:9876/predict -H "Content-Type: application/json" -d '{
  "package_name": "example-package",
  "epoch": 0,
  "version": "1.0.0",
  "mock_chroot": "fedora-42-x86_64",
  "hw_info": {
    "cpu_model_name": "Intel Xeon Processor (Cascadelake)",
    "cpu_arch": "x86_64",
    "cpu_model": "85",
    "cpu_cores": 6,
    "ram": 15324520,
    "swap": 8388604
  }
}'

If everything is working correctly, you should receive a response with a prediction value:

{
  "prediction": 1234
}