Towards a comprehensive understanding of web service integration: a large-scale empirical study from the developers' perspective

Siddhi Baravkar, Pratiksha Gaikwad, Zhengquan Li, Eli Tilevich, Long Cheng, Zheng SongEmpir. Softw. Eng 2025 · DOI

Introduction

In the modern software ecosystem, developers face a "paradox of choice" when integrating web services. For common tasks like weather forecasting or currency conversion, dozens of functionally equivalent APIs exist. While traditional research has focused on Quality of Service (QoS) metrics like latency and reliability, there has been a lack of large-scale empirical data on how developers actually select and implement these services in real-world projects. This study bridges that gap by analyzing developer behavior across the RapidAPI marketplace and GitHub.

Methodology

The researchers developed a comprehensive pipeline to link service metadata with actual usage in open-source projects.

Overview of the multi-step data collection and analysis pipeline.

The process involved a multi-step approach:

  1. Marketplace Crawling: Collecting metadata (latency, pricing, support levels) for over 16,000 services on RapidAPI using Selenium.
  2. GitHub Mining: Identifying service usage by searching for the unique X-RapidAPI-Host header required for these API calls.
  3. Developer Classification: Categorizing developers as "Skilled," "Average," or "Novice" based on their GitHub footprint (repositories, forks, and followers).
  4. Manual Inspection: A deep-dive review of 800 repositories to analyze error-handling, security practices, and selection rationale.

Datasets

The study utilized two primary data sources across two collection rounds (July 2023 and March 2024):

  • RapidAPI: Metadata for approximately 16,400 services.
  • GitHub: Approximately 19,700 repositories invoking these services.
  • Manual Subset: 800 repositories and 382 unique services for qualitative analysis.

The authors have released their collected data to support further research in the community.

Key Insights: What Drives Service Selection?

The study's most significant finding is that human-centric factors often outweigh technical performance metrics.

RapidAPI service metadata highlighting documentation and community support metrics.
  • Support Over Speed: 83% of analyzed GitHub repositories utilize services that provide high-quality documentation or strong community support. Developers are willing to trade some performance for a service that is easy to understand and troubleshoot.
  • "Good Enough" Latency: While latency is a factor, there are diminishing returns. Services with latencies in the 100–500ms range are often as popular as those in the 1–99ms range, suggesting developers aim for "acceptable" rather than "maximum" speed.
  • The Reliability Paradox: Popular services often show slightly lower reliability (97–99%) compared to less-used "perfect" (100%) services. This is likely due to the higher load on popular services, which developers tolerate in exchange for better support and community trust.
  • Proficiency and Priorities: Skilled developers show higher sensitivity to latency and documentation quality, recognizing these as indicators of professional engineering.

Security and Integration Quality

The empirical analysis revealed significant gaps in how services are integrated into production-ready code:

  • Security Risks: Approximately 3.74% of GitHub repositories were found to have leaked their secret RapidAPI keys, exposing them to financial loss and DDoS attacks.
  • Error Handling Neglect: A staggering 31.6% of repositories lacked any exception handling for their API calls. Only 0.4% implemented robust patterns like timeouts and retries.
  • Marketplace Volatility: The ecosystem is highly dynamic; 31.6% of services available in the first round were removed within just eight months.

Conclusion and Future Work

This study highlights that for web service providers, investing in documentation and community support is as critical as optimizing backend performance. When choosing between equivalent options, the factors impacting selection (from highest to lowest) are: 1) Support, 2) Latency/Popularity, and 3) Service Level/Pricing.

Future work should focus on developing recommendation systems that prioritize "integration-friendly" services and LLM-based tools to automatically detect security leaks and improve error-handling patterns in client code.

Summary updated

BibTeX
@article{baravkar2025towards,
  author = {Siddhi Baravkar and Pratiksha Gaikwad and Zhengquan Li and Eli Tilevich and Long Cheng and Zheng Song},
  title = {{Towards a comprehensive understanding of web service integration: a large-scale empirical study from the developers' perspective}},
  journal = {Empir. Softw. Eng.},
  year = {2025},
  doi = {10.1007/s10664-025-10647-9},
}
Figure