ZhengtongYan / JOGGER

JOGGER

efficient Join Order selection learninG with Graph-basEd Representation (JOGGER) is an efficient optimizer for solving the Join order Selection(JOS) problem. It utilizes the curriculum learning, reinforcement learning and a tailored-tree-based attention module to generate query plan.

Important parameters

Here we have listed the most important parameters.

• -ed tree embedding dimension
• -gd graph embedding dimension
• -b batch size
• -e episode
• --model model : DQ,RTOS,JOGGER(proposed method),JOGGER w/o CLO,JOGGER w/o CTRL&CLO
• -f dataset
• -lr learning rate
• -ga gamma
• -wd weight decay

Requirements

• Python 3.7
• Pytorch 1.7
• psqlparse
• deepwalk 1.0.3

Run the JOB

1. Download JOB dataset from https://github.com/gregrahn/join-order-benchmark
2. Add JOB queries in the Directory: JOGGER/code/agents/queries/crossval_sens/IMDB_data.txt
3. Run encode_table.py to build the adjacent matrix to reflect the primary-foreign key relationships
4. Generate the table embedding matrix according to the adjacent matrix by the deepwalk package of Python
5. Run train_JOGGER_main.py to optimize the model

Run the TPC-H

1. Download TPC-H from http://www.tpc.org/tpc_documents_current_versions/current_specifications.asp
2. Generate TPC-H queries from 22 templates
3. Add TPC-H queries in the Directory: JOGGER/code/agents/queries/crossval_sens/TPCH_data.txt
4. Run encode_table.py to build the adjacent matrix to reflect the primary-foreign key relationships
5. Generate the table embedding matrix according to the adjacent matrix by the deepwalk package of Python
6. Run train_JOGGER_main.py to optimize the model