chore: extract comparison tool from fuzzer

Author: comphead

dev/benchmarks/tpcbench.py

@@ -91,9 +91,12 @@ def main(benchmark: str, data_path: str, query_path: str, iterations: int, outpu
                         df.explain()
 
                         if write_path is not None:
-                            output_path = f"{write_path}/q{query}"
-                            df.coalesce(1).write.mode("overwrite").parquet(output_path)
-                            print(f"Query {query} results written to {output_path}")
+                            if len(df.columns) > 0:
+                                output_path = f"{write_path}/q{query}"
+                                df.coalesce(1).write.mode("overwrite").parquet(output_path)
+                                print(f"Query {query} results written to {output_path}")
+                            else:
+                                print(f"Skipping write: DataFrame has no schema for {output_path}")
                         else:
                             rows = df.collect()
                             print(f"Query {query} returned {len(rows)} rows")

fuzz-testing/README.md

@@ -103,3 +103,19 @@ $SPARK_HOME/bin/spark-submit \
 ```
 
 Note that the output filename is currently hard-coded as `results-${System.currentTimeMillis()}.md`
+
+### Compare existing datasets
+
+To compare a pair of existing datasets you can use a comparison tool. 
+The example below is for TPC-H queries results generated by pure Spark and Comet
+
+
+```shell
+$SPARK_HOME/bin/spark-submit \    
+    --master $SPARK_MASTER \
+    --class org.apache.comet.fuzz.ComparisonToolMain     
+    target/comet-fuzz-spark3.5_2.12-0.12.0-SNAPSHOT-jar-with-dependencies.jar \
+    compareParquet --input-spark-folder=/tmp/tpch/spark --input-comet-folder=/tmp/tpch/comet
+```
+
+The tool takes a pair of existing folders of the same layout and compares subfolders treating them as parquet based datasets