Major Description
For tpcds Q15:
explain select ca_zip, sum(cs_sales_price) from catalog_sales, customer, customer_address, date_dim where catalog_sales.cs_bill_customer_sk = customer.c_customer_sk and customer.c_current_addr_sk = customer_address.ca_address_sk and (substr(ca_zip,1,5) in ('85669', '86197','88274','83405','86475', '85392', '85460', '80348', '81792') or ca_state in ('CA','WA','GA') or cs_sales_price > 500) and catalog_sales.cs_sold_date_sk = date_dim.d_date_sk and d_qoy = 2 and d_year = 2001 group by ca_zip order by ca_zip limit 100;
The Traits setup for the Operators are:
FIL[23]: bucketCols=[[]],numBuckets=-1 RS[11]: bucketCols=[[VALUE._col0]],numBuckets=-1 JOIN[12]: bucketCols=[[_col71], [_col71]],numBuckets=-1 FIL[13]: bucketCols=[[_col71], [_col71]],numBuckets=-1 SEL[14]: bucketCols=[[_col71], [_col71]],numBuckets=-1 GBY[15]: bucketCols=[[_col0]],numBuckets=-1 RS[16]: bucketCols=[[KEY._col0]],numBuckets=-1 GBY[17]: bucketCols=[[_col0]],numBuckets=-1 SEL[18]: bucketCols=[[_col0]],numBuckets=-1 LIM[21]: bucketCols=[[_col0]],numBuckets=-1 FS[22]: bucketCols=[[_col0]],numBuckets=-1 TS[3]: bucketCols=[[]],numBuckets=-1 RS[5]: bucketCols=[[VALUE._col0]],numBuckets=-1 JOIN[6]: bucketCols=[[_col3], [_col36]],numBuckets=-1 RS[7]: bucketCols=[[VALUE._col40]],numBuckets=-1 JOIN[9]: bucketCols=[[_col40], [_col0]],numBuckets=-1 RS[10]: bucketCols=[[VALUE._col0]],numBuckets=-1 TS[1]: bucketCols=[[]],numBuckets=-1 RS[8]: bucketCols=[[VALUE._col0]],numBuckets=-1 TS[0]: bucketCols=[[]],numBuckets=-1 RS[4]: bucketCols=[[VALUE._col3]],numBuckets=-1
This is incorrect:
Join[9] joins ca join (cs join cust). In this case both sides of join have a '_col0' column. The reverse mapping of trait propagation relies on ExprNodeColumnDesc.isSame; since this doesn't account for the tabAlias we end up with Join[9] being bucketed on cs_sold_date_sk; Join[12] has the same issue, only compounds the error.