[
https://issues.apache.org/jira/browse/CASSANDRA-12691?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15515666#comment-15515666
]
Sylvain Lebresne commented on CASSANDRA-12691:
----------------------------------------------
I'm not really sure I agree we should make this change and this is definitively
working as designed. What we return on a LWT failure is the data that
correspond to the condition not applying. In this case, the condition is {{IF
NOT EXISTS}}, which means no row should exist. The condition is not limited to
the columns inserted in particular. So returning the full row that does exist
and make the condition fail feels somewhat more "logical" to me.
Don't get me wrong, I'm not saying returning only the inserted columns would be
a completely wrong semantic either. But for every user that would be fine with
only getting the inserting columns back in this case, there is probably a user
for which getting the full row is actually useful, so I think the current
behavior is both slightly more consistent (as justified above) and more useful.
Lastly, and as importantly, changing this now would be a breaking change for
anyone that rely on it, which is the main reason why I'm going to close this.
Meaning that we don't break backward compatibility unless we have strong
evidence that the existence behavior is genuinely problematic and I can't see
any here (but feel free to re-open if you have some).
As an aside, I'll note that we kind of have to query all columns internally to
validate the {{IF NOT EXISTS}} condition anyway, so returning only a subset of
the columns wouldn't even buy much in term of performance of the query.
Lastly, I'm going to assume that you made a typo in your description when you
said "Actual result: Returns true" since it should obviously return false.
> LWT: Inserting Subset of columns returns all columns
> ----------------------------------------------------
>
> Key: CASSANDRA-12691
> URL: https://issues.apache.org/jira/browse/CASSANDRA-12691
> Project: Cassandra
> Issue Type: Improvement
> Components: CQL
> Environment: 3.x
> Reporter: Highstead
> Priority: Minor
> Labels: transaction, transactions
>
> See: https://github.com/gocql/gocql/issues/792#issuecomment-248983669
> When inserting a subset of the table columns with the use of light weight
> transactions the cassandra result returns a full set of unordered cassandra
> column values.
> SETUP:
> {code}
> CREATE TABLE IF NOT EXISTS test.inserttest(
> key bigint,
> session_token text,
> foo text,
> bar text,
> PRIMARY KEY(key, event_date, session_token);
> INSERT INTO test.inserttest(key, session_token, foo) VALUES (1, 'myToken',
> 'baz') IF NOT EXISTS;
> {code}
> {code}insert into test.inserttest(key, session_token, foo) VALUES (1,
> 'myToken', 'bez') IF NOT EXISTS;{code}
> Expected result: Returns False, 1, myToken, baz
> Actual result: Returns true and all column values.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
