SQL Fuzz Testing Utilities

sqlfuzz is a command-line tool and library for generating random data files and SQL queries.

For background information on SQL fuzzing, I recommend reading the paper SparkFuzz: Searching Correctness Regressions in Modern Query Engines.

Status

• Generate random SQL queries against any provided input files
• Generate randomized data files
• Provide tooling to execute queries and capture results
  • DataFusion
  • Apache Spark
  • PostgreSQL
• Provide tooling to compare two sets of results and report differences

Installation

cargo install sqlfuzz

SQL Query Fuzzing

Use the following syntax to generate randomized SQL queries against the example data files in this repository.

sqlfuzz query \
  --table ./testdata/test0.parquet ./testdata/test1.parquet \
  --join-type inner left right full semi anti \
  --count 5 \
  --max-depth 5

Example Generated Query

SELECT __c320, __c321, __c322, __c323
FROM (
    (SELECT test1.c0 AS __c320, test1.c1 AS __c321, test1.c2 AS __c322, test1.c3 AS __c323
    FROM (test1))
    INNER JOIN
    (SELECT test1.c0 AS __c324, test1.c1 AS __c325, test1.c2 AS __c326, test1.c3 AS __c327
    FROM (test1))
    ON __c320 = __c327)
WHERE __c323 > __c320;

Data Generator

sqlfuzz can generate random data files to run the query fuzzer against. The files are generated in Parquet format.

sqlfuzz data --path ./testdata --num-files 4 --row-count 256

Executing Queries

sqlfuzz execute \
  --table ./testdata/test0.parquet ./testdata/test1.parquet \
  --sql testdata/example.sql

Comparing Results

sqlfuzz compare report1.txt report2.txt