Create UNION ALL statements via a loop - sql

Question

Manually, I can select partitions in an inner query with the first code block below. Is there a way to do this in a more elegant way via a loop? I'm showing 3 partitions here, but I have about 200 and the partitions are based on a date column and therefore the partition names will need to change when I run this query again at a future date.
SELECT *
FROM (
SELECT * FROM RSS_ACQ.TRX_ARQ PARTITION("SYS_P211048") UNION ALL
SELECT * FROM RSS_ACQ.TRX_ARQ PARTITION("SYS_P210329") UNION ALL
SELECT * FROM RSS_ACQ.TRX_ARQ PARTITION("SYS_P176323")
) TRX_ARQ
;
With this statement, I've created a loop that outputs the UNION ALL statements.
BEGIN
FOR ALL_TAB_PARTITIONS IN
(
SELECT PARTITION_NAME
FROM ALL_TAB_PARTITIONS
where TABLE_OWNER = 'TABLEOWNER'
AND TABLE_NAME = 'TABLENAME'
AND PARTITION_POSITION > 123
ORDER BY partition_position DESC
)
LOOP
DBMS_OUTPUT.PUT_LINE( 'SELECT * FROM RSS_ACQ.TRX_ARQ PARTITION(\"'
|| ALL_TAB_PARTITIONS.PARTITION_NAME || '\") UNION ALL');
END LOOP;
END;
And in this block, I've attempted to use the loop inside the inner query. It's not yet formatted correctly and I'll need to avoid having UNION ALL for the very last partition.
SELECT *
FROM (
BEGIN
FOR ALL_TAB_PARTITIONS IN
(
SELECT PARTITION_NAME
FROM ALL_TAB_PARTITIONS
where TABLE_OWNER = 'TABLEOWNER'
AND TABLE_NAME = 'TABLENAME'
AND PARTITION_POSITION > 123
ORDER BY partition_position DESC
)
LOOP
DBMS_OUTPUT.PUT_LINE( 'SELECT * FROM RSS_ACQ.TRX_ARQ PARTITION(\"'
|| ALL_TAB_PARTITIONS.PARTITION_NAME || '\") UNION ALL');
END LOOP;
END;
) TRX_ARQ
;
Here are some of the errors, but there were also many more. They are syntax errors pointing to other parts of the query so I would expect that I have an issue with escaping the quotes.
Error starting at line : 99 in command -
END LOOP
Error report -
Unknown Command
Error starting at line : 100 in command -
END
Error report -
Unknown Command
Error starting at line : 101 in command -
)
Error report -
Unknown Command
Error starting at line : 102 in command -
) TABLENAME
Error report -
Unknown Command

Answer 1

This is a bit of a guess, but it's too long for a comment.
I am assuming your table is interval partitioned. In that case, getting all the data from partition positions > 123 is the same as getting all the rows with a higher date than the highest date in partition 123.
You can obtain that date from ALL_TAB_PARTITIONS and then use it to query the table. Like this:
WITH FUNCTION get_high_value RETURN DATE IS
l_high_val_expr ALL_TAB_PARTITIONS.HIGH_VALUE%TYPE;
l_high_value DATE;
BEGIN
SELECT high_value
INTO l_high_val_expr
FROM all_tab_partitions
WHERE table_owner = 'RSS_ACQ'
AND table_Name = 'TRX_ARQ'
and partition_position = 123;
EXECUTE IMMEDIATE 'SELECT ' || l_high_val_expr || ' FROM DUAL' INTO l_high_value;
RETURN l_high_value;
END;
SELECT * FROM rss_acq.trx_arq
-- Replace "partitioned_date_column" with the name of the column on which the
-- table is interval partitioned.
WHERE partitioned_date_column > get_high_value;

Answer 2

We can't execute an anonymous PL/SQL block in a SELECT statement.
What you need to do is spool the output of the ALL_TAB_PARTITIONS loop to a file (or a SQL worksheet if you're using an IDE like SQL Developer). This will give you a script you can run separately after editing it (you need to trim UNION ALL from the final generated SELECT.
Probably there are more elegant ways of achieving the same thing, but the task seems sufficiently wrong that it doesn't strike me as being worth the effort. You want to query 200 partitions in a single statement. That is a brute force operation and there isn't mush to be gained from querying named blocks. In fact, producing a union of 200 separate queries may be more expensive than a single query. So why not try something like this?
select * from RSS_ACQ.TRX_ARQ
where partition_key_col >= date '2018-08-01' -- or whatever
"I think you are overlooking the 12c feature of using PL/SQL in the WITH clause"
That 12c feature is for functions not procedures, so it won't help the OP run their code. It would be possible to use a WITH clause function but that would require:
creating a type with the same projection as the target table
and a nested table type based on that type
a WITH clause function which assembles and executes a dynamic SQL statement
we can't use REF CURSORs in SQL so ...
the function has to execute the dynamic select INTO a local collection variable ...
then loop over the collection and PIPE ROW to output those rows ...
so the main query can call the function with a table() call
Can a WITH clause function be pipelined? I can't find anything in the documentation to say we can't (don't have access to 12c right now to test).