SQL 基于多个条件删除行

Question

I am trying to delete rows from a data set based on multiple criteria, but I am receiving a syntax error. Here is the current code:

With cte As (
        Select *, 
                Row_Number() Over(Partition By ID, Numb1 Order by ID) as RowNumb
        from DataSet
)
Delete from cte Where RowNumb > 1;

Where DataSet looks like this:

I want to delete all records in which the ID and the Numb1 are the same. So I would expect the code to delete all rows except:

Answer 1

WITH Clauses in Vertica only support SELECT or INSERT, not DELETE/UPDATE.

Vertica Documentation

Answer 2

The cte is a temporary table. You cannot delete from it. It is effectively read-only.

If you are trying to delete duplicates out of the original DataSet table, you have to delete from the DataSet, not from the cte table.

Try this:

 with cte as ( select ID, Row_Number() Over(Partition By ID, Numb1 Order by ID) as RowNumb from DataSet ) delete from DataSet where ID in (select ID from cte where RowNumb > 1)

Answer 3

Can't delete from CTEs. Just manually use delete syntax but rollback transactions or if you have permissions you can always replicate it and test.

Answer 4

I am not very experienced with Vertica but it seems like it is not very flexible about delete statements.

One way to do it would be to use a temporary table to store the rows that you want to keep, then truncate the original the table, and insert back into it from the temp table:

create temporary table MyTempTable as
select id, numb1, state_coding
from (select t.*, count(*) over(partition by id, numb1) cnt from DataSet) as t
where cnt = 1;

truncate table DataSet;

insert into DataSet
select id, numb1, state_coding from MyTempTable;

Note that I used a window count instead of row_number . This will remove records for which at least another record exists with the same id and numb1 , which is what I understand that you want from your sample data and expected results.

Important: make sure to backup your entire table before you do this!

Answer 5

You'd have saved me ~5 min had you pasted the data as text and not as picture - as I could not copy-paste and had to retype ...

Having said that:

Rebuild the table here:

DROP TABLE IF EXISTS input;
CREATE TABLE input(id,numb1,state_coding) AS (
          SELECT 202003,4718868,'D'
UNION ALL SELECT 202003,  35756,'AA'
UNION ALL SELECT 204281, 146199,'D'
UNION ALL SELECT 204281, 146199,'D'
UNION ALL SELECT 204346, 108094,'D'
UNION ALL SELECT 204346, 108094,'D'
UNION ALL SELECT 204389,  14642,'DD'
UNION ALL SELECT 204389,  96504,'F'
UNION ALL SELECT 204392,  22010,'D'
UNION ALL SELECT 204392,   8051,'G'
UNION ALL SELECT 204400,  74118,'D'
UNION ALL SELECT 204400, 103900,'D'
UNION ALL SELECT 204406,1387304,'D'
UNION ALL SELECT 204406,      0,'HJ'
UNION ALL SELECT 204516,    894,'D'
UNION ALL SELECT 204516,   3927,'D'
UNION ALL SELECT 204586, 234235,'D'
UNION ALL SELECT 204586, 234235,'D'
)
;

And then: Based on what was said in other responses, and keeping in mind that a mass delete of an important part of the table, not only in Vertica, is best implemented as an INSERT ... SELECT with inverted WHERE condition - here goes:

CREATE TABLE input_help AS
SELECT * FROM input
GROUP BY id,numb1,state_coding
HAVING COUNT(*) = 1;

DROP TABLE input;
ALTER TABLE input_help RENAME TO input;

At least, it works with that simplicity if the whole row is the same - I notice you don't put state_coding into the condition yourself. Otherwise, it gets slightly more complicated.

Or did you want to re-insert one row of the duplicates each afterwards?

Then, just build input_help as SELECT DISTINCT * FROM input; , then drop, then rename.