Codebase of a working open source DBMS configuration tuning framework.

DBMS

• PostgreSQL (We currently support Postgres and provide a step-by-step reference for installing Postgres 12.17 on Ubuntu 20.04.)

Workload Generator

• Benchbase (We currently use Benchbase as the workload generator and provide a step-by-step reference for running Benchbase.)

Assume using Miniconda for Python package management on Linux machines.

1. Clone this repo in your working directory:

   git clone <Cybernetics repo url>
   cd cybernetics
   

2. Create and activate the development environment:

   conda env create -f environment.yml
   conda activate cybernetics
   

3. Import Cybernetics as editable packages to the conda environment:

   conda develop <path to Cybernetics>
   

   e.g.,

   conda develop /home/tianji/cybernetics
   

Follow the steps below to run vanilla Bayesian optimization (i.e., BO-Gaussian Process) for Postgres over the TPC-C workload.

1. (a) Set your Postgres username, password and BenchBase target directory as environment variables in e.g., ~/.bashrc:

   export POSTGRES_USER=<your username>
   export POSTGRES_PASSWORD=<your password>
   export BENCHBASE_POSTGRES_TARGET_DIR=<path to directory containing BenchBase executable jar>
   

   E.g., If benchbase.jar is under ~/benchbase/target/benchbase-postgres, then set export BENCHBASE_POSTGRES_TARGET_DIR=~/benchbase/target/benchbase-postgres

   (b) Apply the changes: source ~/.bashrc

2. (a) Make a copy of cybernetics/configs/benchbase/tpcc/postgres_bo_gp.ini, name it cybernetics/configs/benchbase/tpcc/postgres_bo_gp.local.ini (.gitignore will prevent *.local.ini config files from being pushed to the repo so we won't step on each other for local changes to config files.)

   (b) Specify your local paths in cybernetics/configs/benchbase/tpcc/postgres_bo_gp.local.ini including

   dbms_info.db_cluster: where Postgres stores all data

   dbms_info.db_log_filepath: where Postgres saves logs

   results.save_path: where you want to save the experiment results

3. Start DBMS config tuning:

   python ./examples/run_dbms_config_tuning.py --config_path ./cybernetics/configs/benchbase/tpcc/postgres_bo_gp.local.ini
   

4. Once tuning is complete, check ./logs/<latest run>/cybernetics.log for logs and see BenchBase and tuning optimizer outputs in results.save_path.

Cybernetics is under active development by Tianji Cong. Please use GitHub's issue tracker for all issues and feature requests.

• BO - Gaussian Process (Vanilla BO)
• BO - Random Forest (SMAC)
• RL - DDPG

• Knob Selection - Lasso
• Knob Selection - Gini Index
• Knob Selection - fANOVA
• Knob Selection - SHAP Value
• Space Transformation - Random Linear Projection
• Space Transformation - Biased Sampling for Hybrid Knobs
• Space Transformation - Knob Value Bucketization

• Workload Mapping