[jira] [Created] (CASSANDRA-16784) Conditional update on super column table returns too many rows

["Marten Kenbeek (Jira)" <ji...@apache.org>]

Marten Kenbeek created CASSANDRA-16784:
------------------------------------------

             Summary: Conditional update on super column table returns too many rows
                 Key: CASSANDRA-16784
                 URL: https://issues.apache.org/jira/browse/CASSANDRA-16784
             Project: Cassandra
          Issue Type: Bug
          Components: Feature/Lightweight Transactions
            Reporter: Marten Kenbeek
         Attachments: dense_supercolumn_lwt_test.patch

We have a (dense) supercolumn table that we're trying to upgrade from 2.2 to 3.x. On this table we perform conditional updates on one of the values using a CQL query. In 2.2 this works as expected: the query returns either a single-column row with {{true}}, or a two-column row with {{false}} and the current value.

In 3.x this query is internally rewritten to work on a map, using a {{Maps.SetterByKey}} operation. The update itself is applied correctly, but when the condition fails and the update is not applied, the query returns all values in the supercolumn/map instead of just the value of the subcolumn/key from the where clause.

In particular, since it returns _just_ the values, and not the keys/mapping, there's no way to know which of the values is the value of the row we're trying to update.
-- Super column table (created through Thrift)
CREATE TABLE store.shopping_items (
    key text,
    column1 text,
    column2 text,
    "" map<text, text>,
    value text,
    PRIMARY KEY (key, column1, column2)
) WITH COMPACT STORAGE;

-- Create some data
cqlsh:store> INSERT INTO shopping_items (key, column1, column2, value) VALUES ('item1', 'default', 'id', '1');
cqlsh:store> INSERT INTO shopping_items (key, column1, column2, value) VALUES ('item1', 'default', 'color', 'blue');

-- This query returns multiple rows
cqlsh:store> UPDATE shopping_items SET value ='2' WHERE key ='item1' AND column1 = 'default' AND column2 = 'id' IF value = null;

 [applied] | value
-----------+------------
     False |     'blue'
     False |        '1'
See the attached patch for a test that succeeds on 2.2, but fails with the described behavior on 3.0 and on the latest 3.11. 



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscribe@cassandra.apache.org
For additional commands, e-mail: commits-help@cassandra.apache.org