상세 보기
초록
We present CI-Sherpa, a lightweight, budget-aware system for automatic continuous integration (CI) failure triage and repair suggestion. CI-Sherpa addresses the high cost and complexity of debugging CI logs by combining structured log reduction, failure-type classification, and LLM-based reasoning under strict token and latency constraints. The system dynamically compresses execution logs while preserving failure-critical signals, enabling efficient diagnosis without degrading accuracy. Evaluated on 280 real CI traces across diverse failure categories, CI-Sherpa achieves near-perfect classification accuracy while reducing input tokens by approximately 94-95%, comparable to the regex baseline, and maintaining sub-millisecond latency. © 2026 Copyright held by the owner/author(s).
키워드
- 제목
- CI-Sherpa: Budget-Aware CI Failure Triage and Fix Suggestion System
- 저자
- Song, Zichen
- 발행일
- 2026-09-02
- 유형
- Proceedings Paper
- 저널명
- 19th ACM International Systems and Storage Conference, SYSTOR 2026
- 페이지
- 155 ~ 156
- 언어
- ENG
- 출판사
- Association for Computing Machinery, Inc
- 분량
- 2 페이지