This paper explores the use of large language models (LLMs) to enhance cloud incident management, focusing on two critical tasks: root cause analysis for dependency failures and monitor categorization. The authors leverage data from different stages of the software development lifecycle (SDLC) to improve the performance of these tasks. By incorporating additional contextual information, such as service architecture, dependencies, and functionalities, the proposed methods outperform state-of-the-art (SoTA) methods in both scenarios. **Root Cause Analysis for Dependency Failures:** - **Problem Formulation:** The root cause of incidents often involves upstream service dependencies, which require detailed understanding and reasoning. - **Data Preparation:** The dataset includes 353 historical incidents, with a focus on dependency failures. Incident metadata, upstream service dependencies, and service properties are curated. - **Methods and Baselines:** Five strategies are developed, including the proposed method (InC DEP) that combines incident metadata, in-context examples, and upstream service details. - **Evaluation:** Lexical and semantic metrics are used to evaluate the effectiveness of different prompting strategies. The proposed method (InC DEP) significantly improves F1-score for dependency failure classification. **Monitor Categorization:** - **Problem Formulation:** The goal is to classify monitors into resource and SLO classes using monitor metadata and additional contextual information. - **Data Preparation:** 260 real-world monitors are labeled for resource and SLO classes, incorporating service descriptions and component functionalities. - **Methods and Baselines:** Experiments are conducted with different combinations of data sources, including only monitor metadata, monitor metadata with service descriptions, and monitor metadata with service and component descriptions. - **Evaluation:** Precision, recall, F1-score, and accuracy metrics are used to evaluate the performance. The proposed method (InC DEP) shows significant improvements in SLO classification, particularly for classes like "Availability" and "Freshness." **Experimental Results:** - **Root Cause Analysis:** The proposed method (InC DEP) achieves a higher F1-score for dependency failure classification compared to other methods. - **Monitor Categorization:** The inclusion of service descriptions improves the accuracy of SLO and resource class predictions, especially for certain classes. **Lessons Learned and Threats:** - Additional contextual data significantly enhances performance, but irrelevant information can sometimes reduce it. Contextual data should align with the task to achieve better results. Overall, the paper demonstrates the effectiveness of leveraging X-lifecycle data to improve the accuracy and efficiency of cloud incident management using LLMs.This paper explores the use of large language models (LLMs) to enhance cloud incident management, focusing on two critical tasks: root cause analysis for dependency failures and monitor categorization. The authors leverage data from different stages of the software development lifecycle (SDLC) to improve the performance of these tasks. By incorporating additional contextual information, such as service architecture, dependencies, and functionalities, the proposed methods outperform state-of-the-art (SoTA) methods in both scenarios. **Root Cause Analysis for Dependency Failures:** - **Problem Formulation:** The root cause of incidents often involves upstream service dependencies, which require detailed understanding and reasoning. - **Data Preparation:** The dataset includes 353 historical incidents, with a focus on dependency failures. Incident metadata, upstream service dependencies, and service properties are curated. - **Methods and Baselines:** Five strategies are developed, including the proposed method (InC DEP) that combines incident metadata, in-context examples, and upstream service details. - **Evaluation:** Lexical and semantic metrics are used to evaluate the effectiveness of different prompting strategies. The proposed method (InC DEP) significantly improves F1-score for dependency failure classification. **Monitor Categorization:** - **Problem Formulation:** The goal is to classify monitors into resource and SLO classes using monitor metadata and additional contextual information. - **Data Preparation:** 260 real-world monitors are labeled for resource and SLO classes, incorporating service descriptions and component functionalities. - **Methods and Baselines:** Experiments are conducted with different combinations of data sources, including only monitor metadata, monitor metadata with service descriptions, and monitor metadata with service and component descriptions. - **Evaluation:** Precision, recall, F1-score, and accuracy metrics are used to evaluate the performance. The proposed method (InC DEP) shows significant improvements in SLO classification, particularly for classes like "Availability" and "Freshness." **Experimental Results:** - **Root Cause Analysis:** The proposed method (InC DEP) achieves a higher F1-score for dependency failure classification compared to other methods. - **Monitor Categorization:** The inclusion of service descriptions improves the accuracy of SLO and resource class predictions, especially for certain classes. **Lessons Learned and Threats:** - Additional contextual data significantly enhances performance, but irrelevant information can sometimes reduce it. Contextual data should align with the task to achieve better results. Overall, the paper demonstrates the effectiveness of leveraging X-lifecycle data to improve the accuracy and efficiency of cloud incident management using LLMs.