rcabench-platform 0.2.1

An experiment framework for Root Cause Analysis

rcabench-platform

An experiment framework for Root Cause Analysis (RCA), supporting fast development of RCA algorithms and their evaluation on various datasets.

Development Guide

Requirements

Operating System

This project is primarily developed and tested on Ubuntu 24.04 LTS or later versions. Other Linux distributions and macOS environments should be compatible with minimal configuration adjustments.

Windows is not officially supported. While some functionality may work in Windows environments (especially through WSL), we cannot guarantee full compatibility or provide dedicated support.

Toolchain

Toolchain	Version
uv	^0.7.5
just	^1.21.0
Docker Engine	*
Docker Compose	*

IDE

Recommended setup

• Visual Studio Code with the following extensions:
  • Python Extension Pack
  • Pylance
  • Ruff
  • Git Graph
  • Data Wrangler

Git

Commit Message

We follow the Conventional Commits specification.

Branching Strategy

When you are developing a new feature, create a new branch from main and name it according to the following convention:

<your github id>/feat/<feature-name>

This branch prefixed with your github id, is your own working branch. You can force-push to it freely. Do anything you want in this branch.

When you are done with the feature, create a pull request to main and invite other developers to review your code. If the code is approved, it will be merged into main. Then you can start a new branch from main and continue your work.

The main branch is the default branch for this repository. main is protected and should not be used for development. Before merging any changes into main, ensure that the following conditions are met:

• The branch is up to date with main.
• The branch is free of merge conflicts.
• The basic checks passed successfully.
• The changes will not break other developers' workflow.

We requires a linear commit history. Please use git rebase to keep your branch up to date with main. When merging your branch into main, use git merge --ff-only to ensure a fast-forward merge. This will keep the commit history clean and linear.

Workflow

Download source code

git clone [e-mail]:LGU-SE-Internal/rcabench-platform.git
cd rcabench-platform

Run basic checks

just dev

If the basic checks pass, then your python environment is ready for development.

Local development services

docker compose up -d

docker compose down

We have the following localhost services running in the background:

• neo4j: for graph visualization

Link datasets

Mount JuiceFS to your machine:

sudo juicefs mount redis://10.10.10.38:6379/1 /mnt/jfs -d --cache-size=1024

See infra/README.md for more details.

Link the datasets to the project directory:

mkdir -p data
cd data
ln -s /mnt/jfs/rcabench-platform-v2 ./

Docker image

rcabench-platform:

./scripts/docker.sh build
./scripts/docker.sh run
./scripts/docker.sh push

clickhouse_dataset:

cd docker/clickhouse_dataset
./cli.sh build
./cli.sh push

Commands

Self test

Test if the environment is set up correctly:

./main.py self test

Prepare inputs

./cli/prepare_inputs.py --help

Make rcaeval

./cli/make_rcaeval.py --help

Make rcabench

./cli/make_rcabench.py --help

Example calls:

mkdir -p /dev/shm/make
TMP=/dev/shm/make LOGURU_COLORIZE=0 POLARS_MAX_THREADS=16 ./cli/make_rcabench.py run --parallel=8 >temp/a.log 2>&1

The example call runs 8 parallel processes with 16 polars threads each, using memory storage as the temporary directory. It is tested on a VM with 128 cores and 192 GiB of RAM.

./cli/make_rcabench.py make-filtered
./cli/make_rcabench.py make-with-issues

./cli/make_rcabench.py merge-conclusion

./cli/make_rcabench.py query-fault-types rcabench
./cli/make_rcabench.py query-fault-types rcabench_filtered
./cli/make_rcabench.py query-fault-types rcabench_with_issues

Evaluation

./main.py eval --help

./main.py eval show-algorithms
./main.py eval show-datasets

Example calls:

./main.py eval single traceback-A7 rcabench_filtered ts3-ts-route-plan-service-request-delay-59s2q4

LOGURU_LEVEL=INFO ./main.py eval batch -d rcaeval_re2_tt -a random -a baro -a nsigma -a traceback-A7 --use-cpus=112 --clear >temp/a.log 2>&1

LOGURU_LEVEL=INFO ./main.py eval batch -d rcabench_filtered -a random -a baro -a nsigma -a traceback-A7 --use-cpus=112 --clear >temp/a.log 2>&1

./main.py eval perf-report rcaeval_re2_tt
./main.py eval perf-report rcabench_filtered

Notebooks

Edit the SDG Visualization notebook:

./notebooks/sdg.py