Question

We have such query:

select X.id_x, A.id_a, B.id_b
from X
left join A on 'A|' + A.id_a = X.id_aOrB
left join B on 'B|' + B.id_b = X.id_aOrB

X is linked to A or B and id is computed in the join statement.

• X: 8 000 rows
• A: 36 000 rows
• B: 3 000 rows

This query is very slow, like 10 seconds. No index on X.id_aOrb.

Then using 2 “join” tables, updated by trigger, we don’t need to concat 'A|' + id and 'B|'+id. Results are fetch in less than 1 second. OK.

My question: why is this concat so slow? Is SQL Server no very efficient in ‘+’ when there is too much data?

Answer 1

FROM X
LEFT JOIN A on 'A|' + A.id_a = X.id_aOrB

is not sargable (SQL Server cannot use an index).

If it uses a nested loops join then it must scan A as many times as there are rows in X. It is not possible to do an index seek on A to evaluate the 'A|' + A.id_a = X.id_aOrB predicate.

If it uses a merge join it must take a copy of the data and sort it first rather than being able to use the order from the index.

The same of course applies to the join on B