GCP Bigquery 上 DBT 中的 Left Join 性能

Question

I'm trying to build a reusable model which will be used in multiple data pipelines constructed on DBT running on GCP Bigquery. I basically want to get a column value from two disjointed tables.

I created a DBT model UNION-ing those two tables

data for tbl1 - col1#####col2#######col3 ABC1 ###### A1 ######## T1 ABC2 ###### A2 ######## T2 ABC3 ###### A3 ######## T3

data for tbl2 - col1 ###### col2 A1 ######## T11 A2 ######## T21 A3 ######## T31

data for fact table - col1 ####### col2 ABC1 ####### XX ABC2 ####### YY XYZ ####### A1 DEF ####### A3

For the data in fact table, row 1 & 2 must match with the tbl1 data and row 3 & 4 must match with the tbl2 data.

with tbl1 as 
(select col1,col2,col3
from project1.dataset1.table1 ) 

,tbl2 as 
(select col2, col3
from project1.dataset1.table2 )

,tbl1_tbl2 as 
(
select col3, 
case when col1 is not null then col1
when col2 is not null then col2 end as single_col
from (
select col3,
case when flg = 'tbl1' then col1 end as col1,
case when flg = 'tbl2' then col2 end as col2
from
(select 'tbl2' as flg, null as col1 ,col3, col2 from tbl2
union all 
select 'tbl1' as flg, col1,col3,col2 from tbl1
)
) 
)

This is how the model will be used in data pipeline - fact joined with the model. But the challenge here is that when either of the JOIN columns' value is matching with the UNION-ed tables then I need the data. But the OR condition in the LEFT JOIN.....ON fails in Bigquery. When I say fails is that BQ goes on for a ride. I'm not sure if I'm approaching it correctly or if there's a better way to solve this issue. Please help!

select col3
from project1.dataset1.fact_table fact
left join 
tbl1_tbl2 tbl
on 
tbl.single_col = fact.col1
or tbl.single_col = fact.col2 --This is where BQ fails

Answer 1

Using an or in a join predicate causes terrible performance in many different warehouses. You can replace this with two left joins, and coalesce the values from the two joined tables:

select coalesce(a.col3, b.col3) as col3
from project1.dataset1.fact_table fact
left join tbl1_tbl2 a
    on a.single_col = fact.col1
left join tbl1_tbl2 b
    on b.single_col = fact.col2