TraceWeaver is a research prototype for transparently tracing requests through a microservice without application implementation. This repository serves as the artifact for the associated academic paper "TraceWeaver: Distributed Request Tracing for Microservices Without Application Modification" which will appear in SIGCOMM'24.
The project is organized as follows:
src/ The main traceweaver algorithm as well as baselines are provided within src/.
data/ For convenience, trace data collected from each application is made available within the data/ directory. The trace data is converted to standard JSON format as supported by the Jaeger tracing framework. Please ensure at least 100 GB of space on your local node to store the traces once extracted.
exps/ The experiments folder contain one sub-directory per experiment mentioned in the paper. The provided bash scripts execute the experiments mentioned in the paper from scratch using traces from data/.
third_party/ The applications used for evaluation are provided as git submodules in the third_party/ folder. Please follow instructions within the corresponding submodules to run the apps.
utils/ Contains miscelleanous utility scripts.
Please ensure git is installed locally on your machine. For some large datasets (e.g., data/alibaba_microservices/), Git LFS is needed to manage and track changes. Please install it in the following way.
# On macOS
brew install git-lfs
# On Debian/Ubuntu
sudo apt-get install git-lfs
# On Windows (using Chocolatey)
choco install git-lfs
And initialize Git LFS locally by doing
git lfs install
To clone this repository, run:
git clone https://github.com/Sachin-A/TraceWeaver.git && cd TraceWeaver
Install python (3.11.9) by running your platform equivalent of the following command:
sudo apt-get update
sudo apt-get install python3.11
Install virtualenv to create and activate a virtual environment env to isolate the project dependencies from your system packages:
pip install virtualenv
python3.11 -m venv env
source env/bin/activate
Now install all dependencies within the virtual environment:
pip install -r requirements.txt
Finally, proceed to exps/exp* to re-run each experiment (README within exps/ has more details).
In order to use the Gurobi Solver invoked by TraceWeaver, a Gurobi Academic license is required. Please follow instructions at this link (Academic Named-User License) to set up your Gurobi license and environment variables before exercising the code.
Note that this artifact is still being updated. We are making the source code available now, but some functionality is being updated from our private development. In addition, we plan to provide more detailed instructions to run the benchmarks, some of which relies on large preprocessed datasets from the Alibaba cluster.
- Please reach out via email ([email protected]) for support.
- Send us a PR if you are interested in contributing code.