SQL select from either one or other table - sql

Question

Assume I have a table A with a lot of records (> 100'000) and a table B with has the same columns as A and about the same data amount.
Is there a possibility with one clever select statement that I can either get all records of table A or all records of table B?
I am not so happy with the approach I currently use because of the performance:
select
column1
,column2
,column3
from (
select 'A' as tablename, a.* from table_a a
union
select 'B' as tablename, b.* from table_b b
) x
where
x.tablename = 'A'

Answer 1

Offhand, your approach seems like the only approach in standard SQL.
You will improve performance considerably by changing the UNION to UNION ALL. The UNION must read in the data from both tables and then eliminate duplicates, before returning any data.
The UNION ALL does not eliminate duplicates. How much better this performs depends on the database engine and possibly on turning parameters.
Actually, there is another possibility. I don't know how well it will work, but you can try it:
select *
from ((select const.tableName, a.*
from A cross join
(select 'A' as tableName where x.TableName = 'A')
) union all
(select const.tableName, b.*
from B cross join
(select 'B' as tableName where x.TableName = 'B')
)
) t
No promises. But the idea is to cross join to a table with either 1 or 0 rows. This will not work in MySQL, because it does not allow WHERE clauses without a FROM. In other databases, you might need a tablename such as dual. This gives the query engine an opportunity to optimize away the read of the table entirely, when the subquery contains no records. Of course, just because you give a SQL engine the opportunity to optimize does not mean that it will.
Also, the "*" is a bad idea particularly in union's. But I've left it in because that is not the focus of the question.

Answer 2

you can try next solution, it's selects only from table tmp1 ('A' = 'A')
select
*
from
tmp1
where
'A' = 'A'
union all
select
*
from
tmp2
where
'B' = 'A'
SQL Fiddle demo here
check execution plan

Answer 3

Hard to tell exactly what you want without a little more context, but perhaps something like this could work?
DECLARE #TableName nvarchar(15);
DECLARE #Query nvarchar(50);
SELECT #TableName = YourField
FROM YourTable
WHERE ...
SET #Query = 'SELECT * FROM ' + #TableName
EXEC #Query
Syntax might differ a bit depending on what RDBMS you are using, and more specifically what you are trying to accomplish, but might be a push in the right direction.

Answer 4

The proper way to do this and maintain performance requires some modification to your physical table design.
If you can add a column to each table that holds your indicator column and add a check constraint on that column, you can achieve "partition" elimination on your query.
DDL:
create table table_a (
c1 ...
,c2 ...
,c3 ...
,table_ind char(1) not null generated always as 'A'
,constraint ck_table_ind check (table_ind = 'A')
);
create table table_b (
c1 ...
,c2 ...
,c3 ...
,table_ind char(1) not null generated always as 'B'
,constraint ck_table_ind check (table_ind = 'B')
);
create view v1 as (
select * from table_a
union all
select * from table_b
);
If you execute the query select c1,c2,c3 from v1 where table_ind = 'A' the DB2 optimizer will use the check constraint to recognize that no rows in table_b can match the table_ind = 'A' predicate, so it will completely eliminate the table from the access plan.
This was used (and still is in some cases) before DB2 for Linux/UNIX/Windows supported Range Partitioning. You can read more about this technique in this research paper [PDF] written by some of the IBM DB2 developers back in 2002.