Contract Signing Platform System Design and Implementation for Cloud Music

This article introduces the design and implementation of a contract signing platform system for NetEase Cloud Music, providing convenient file signing capabilities for business operations.

Background: In the music copyright field, signing contracts is a frequent necessity. Cloud Music needs to obtain song authorization from artists through contracts before上架 (listing) songs. Previously, the entire signing process took 15-30 days. The existing e签宝-based system had several issues: inflexible template publishing requiring code changes for every modification, rigid signing workflows unable to dynamically add/remove processing nodes, and accuracy issues in generated contract files.

Solution - Template Configuration: Contracts consist of static legal content and dynamic data (signing dates, song info, artist info). The system uses HTML + dynamic parameter filling to generate PDF documents, implemented with itextpdf . A custom Base64ImgReplacedElementFactory handles image scaling to prevent overflow:

public class Base64ImgReplacedElementFactory implements ReplacedElementFactory { public ReplacedElement createReplacedElement(LayoutContext c, BlockBox box, UserAgentCallback uac, int cssWidth, int cssHeight) { Element e = box.getElement(); if (e == null) { return null; } String nodeName = e.getNodeName(); if (nodeName.equals("img")) { String attribute = e.getAttribute("src"); FSImage fsImage; try { fsImage = buildImage(attribute, uac); } catch (BadElementException | IOException e1) { log.warn("buildImage failed",e); fsImage = null; } if (fsImage != null) { double scale = 1.0; if (fsImage.getWidth() > 8000) { scale = fsImage.getWidth() / 8000.0; } fsImage.scale((int) (fsImage.getWidth() / scale), (int) (fsImage.getHeight() / scale)); return new ITextImageElement(fsImage); } } return null; } }

Multi-version control is also supported for rollback scenarios.

Solution - Process Configuration: The system uses an event-driven architecture based on signing states and events. Each state represents a point in the workflow, and transitions between states are triggered by specific events. The platform sends MQ messages after each state change for business follow-up.

Solution - File Inspection: Using PaddleOCR library for PDF image recognition, the system parses generated PDFs and validates content accuracy. Currently covers 18 major signing scenarios with 97.9% accuracy, having identified 30,000+ historical song file anomalies.

Solution - Process Data Monitoring: Monitors specific nodes for blocking issues. For example, human review in artist signing typically takes 1-2 days; nodes blocked beyond this threshold are marked as anomalies.

Results: Template configuration has been applied to most signing scenarios, significantly reducing development costs. The process configuration solves workflow modification issues, and file inspection prevents business losses.