Medicine

Giving Every Modality a Voice in Microservice Failure Diagnosis via Multimodal Adaptive Optimization

The framework consists of three main stages: feature encoding, modality fusion, and optimization balancing. In the feature encoding stage, we design specific encoders for metrics, logs, and traces based on their characteristics. During the modality fusion stage, we use a fusion module with channel attention to combine the original statistical features and the failure classification features extracted by the encoders. In the optimization balancing stage, Medicine employs a modality evaluation component to identify high-yield and low-yield modalities, then uses a gradient suppression component to optimize high-yield modalities and a feature enhancement component to strengthen low-yield modalities.

Description

Folder Structure

Already Existing Folders

• ./dataset: it is used to preprocess and load data from specific path
• ./models: it stores models used in experiment
• ./cache: it stores the bert-base-uncased model used in experiment

Auto Generated Folders

• ./data: it stores the intermediary files of each preprocessed dataset
• ./result: it stores the result of each experiment

File Description

• config.py：the config file of D1(AIops22) and D2(GAIA)
• main.py：the entry file of experiment
• trainer.py：the script file of traun and test model
• utils.py：the utils file

Environment

• Linux Server 20.04.1 LTS
• Intel(R) Xeon(R) CPU E5-2650 v4@ 2.20GHz
• Python version 3.9

Getting Started

Recommend conda environment or venv

Run the following commands in turn at bash or shell

1. pip install -r requirements.txt
2. python main.py --dataset gaia or python main.py --dataset aiops22

All result will be saved in result/

Reproduce

1. download gaia and aiops22 dataset we provide for you
2. change dataset config in ./config.py at the position described below:

   gaia = {
       "dataset_dir": "path to downloaded gaia dataset entry"
   }
   aiops22 = {
       "dataset_dir": "path to downloaded aiops22 dataset entry"
   }

3. remove ./data folder
4. run python main.py --dataset gaia or python main.py --dataset aiops22

All result will be saved in result/

Some Issues

1. Have some trouble with downloading bert-base-uncased by using transformers

download pytorch version of bert-base-uncased from huggingface

then copy them into ./cache/

Citing Medicine

Medicine paper is published in IEEE/ACM International Conference on Automated Software Engineering 2024. If you use Medicine, we would appreciate citations to the following paper:

Giving Every Modality a Voice in Microservice Failure Diagnosis via Multimodal Adaptive Optimization.

By Lei Tao, Shenglin Zhang, Zedong Jia, Jinrui Sun, Dan Pei, et.al.

BibTex:

@inproceedings{tao2024medicine,
  title={Giving Every Modality a Voice in Microservice Failure Diagnosis via Multimodal Adaptive Optimization},
  author={Tao, Lei and Zhang, Shenglin and Jia, Zedong and Sun, Jinrui and Ma, Minghua and Li, Zhengdan and Sun, Yongqian and Yang, Canqun and Zhang, Yuzhi and Pei, Dan},
  booktitle={Proceedings of the 39th IEEE/ACM International Conference on Automated Software Engineering},
  pages={1107--1119},
  year={2024}
}