Towards the Self-Healing of Infrastructure as Code Projects Using Constrained LLM Technologies

The generalization of the use of cloud computing and edge computing solutions in industry requires innovative techniques to keep up with the complexity of these scenarios. In particular, the large heterogeneity of the infrastructural devices and the myriad of services offered by the various private...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:2024 IEEE/ACM International Workshop on Automated Program Repair (APR) S. 1 - 4
Hauptverfasser: Diaz-de-Arcaya, Josu, Lopez-de-Armentia, Juan, Zarate, Gorka, Torre-Bastida, Ana I.
Format: Tagungsbericht
Sprache:Englisch
Veröffentlicht: ACM 20.04.2024
Schlagworte:
Applied computing → IT architectures
automated patching
Cloud computing
Codes
Computational modeling
Conferences
Decision analysis
IaC
Industries
Information systems → Computing platforms
Infrastructure as Code
Large language models
LLMs
Maintenance engineering
self-healing
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:The generalization of the use of cloud computing and edge computing solutions in industry requires innovative techniques to keep up with the complexity of these scenarios. In particular, the large heterogeneity of the infrastructural devices and the myriad of services offered by the various private and cloud providers represent a challenge. Infrastructure as Code (IaC) technologies have been adopted to reduce the complexity of these scenarios, but even IaC technologies have their drawbacks, as the errors resulting from their use often combine the complexities of the underlying layers and require a high level of expertise. In this regard, the recent upsurge of Large Language Models represents an opportunity as they are able to tackle different problems. In this article, we aspire to shed light on the automated patching of IaC projects with the help of LLMs. We evaluate the suitability of this hypothesis by using a well-known LLM that is able to solve all the scenarios we envisioned and assess the possibility of doing the same with smaller, offline LLMs, which could lead to the use of these technologies in resource-constrained environments, such as edge computing.
DOI:10.1145/3643788.3648014