一句话：提出统一运行时状态表示模型 RuntimeSlicer，通过对齐指标、追踪和日志的对比学习预训练，实现可泛化的故障管理。
现代软件系统规模与复杂度空前，故障管理日益困难。指标、追踪和日志提供互补的运行时视图，但现有方法依赖任务特定流水线，将模态预处理、表示学习与下游模型紧耦合，限制泛化。为此提出 RuntimeSlicer：一个统一运行时状态表示模型。它预训练任务无关模型，将指标、追踪和日志直接编码为单一对齐的系统状态嵌入；引入统一运行时对比学习，整合异构数据并优化跨模态对齐与时间一致性；进一步提出状态感知任务导向调优，无监督划分运行时状态并实现状态条件适配。在 AIOps 2022 数据集上的初步实验证明了其在系统状态建模与故障管理上的可行性和有效性。
Modern software systems operate at unprecedented scale and complexity, making effective failure management both critical and increasingly challenging. Metrics, traces, and logs provide complementary views of runtime behavior, yet existing approaches typically rely on task-specific pipelines that tightly couple modality-specific preprocessing, representation learning, and downstream models, limiting generalization across tasks and systems. To address this gap, we propose RuntimeSlicer, a unified runtime state representation model for generalizable failure management. RuntimeSlicer pre-trains a task-agnostic model that directly encodes metrics, traces, and logs into a single aligned system-state embedding capturing the overall runtime condition. To train RuntimeSlicer, we introduce Unified Runtime Contrastive Learning, which integrates heterogeneous training data and optimizes complementary objectives for cross-modality alignment and temporal consistency. Building on the learned embeddings, we further propose State-Aware Task-Oriented Tuning, which partitions runtime states in an unsupervised manner and enables state-conditioned adaptation for downstream tasks. Preliminary experiments on the AIOps 2022 dataset demonstrate the feasibility and effectiveness of RuntimeSlicer for system state modeling and failure management.
DOI 原文 · arXiv · PDF（开放获取） · DBLP
@inproceedings{abs-2603-21495,
title = {RuntimeSlicer: Towards Generalizable Unified Runtime State Representation for Failure Management},
author = {Lingzhe Zhang and Tong Jia and Weijie Hong and Mingyu Wang and Chiming Duan and Minghua He and Rongqian Wang and Xi Peng and Meiling Wang and Nicholas Zhang and Renhai Chen and Ying Li},
booktitle = {FSE},
year = {2026},
doi = {10.1145/3803437.3805559}
}