[
https://issues.apache.org/jira/browse/CASSANDRA-10631?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15024572#comment-15024572
]
Sylvain Lebresne commented on CASSANDRA-10631:
----------------------------------------------
So, there were more bugs that the change to CQLTester (so prepared statements
are reused) was triggering. The first one, I've created CASSANDRA-10758 for:
fixing it is not trivial enough that I feel confident shoving it in that
(unrelated) ticket. To work-around that ticket, I've disabled the prepared
statement reuse for the 2 class of tests that were failing otherwise (and I'll
add a comment in CASSANDRA-10768 to remove that work-around).
The 2nd problem was that the implementation of the {{DROP INDEX}} statement was
stateful: we were looking up the metadata for table for which the index is
dropped and saving it in an instance field, reusing it afterwards. Except that
it's during this lookup that we're throwing an exception if the index doesn't
exists, so this saving the result of the lookup was breaking if a {{DROP
INDEX}} was prepared and reused. One could argue that preparing any schema
related statement is borderline useless since you can't use bind marker
anywhere (I guess you save parsing ...), but we actually don't disallow it and
hence should probably fix that. Now, the main reason for saving the result of
that metadata lookup was due to the fact that we need that metadata when
creating the {{Event.SchemaChange}}, but by the time {{changeEvent()}} is
called, the index has been dropped and the lookup method would fail. Anyway,
what that mean is that {{changeEvent}} needed access to values that where only
available in {{announceMigration}} (there is a few other function related
statement that saves stuffs in fields for the very same reason) and so I make
the change of removing the {{changeEvent()}} and making the event be the result
of the {{announceMigration()}} call. I happen to think it's cleaner that way
anyway and the change is trivial, so I've included a commit for this one.
Only the 3.0 branch has new commits (there were no tests failure in 2.2 so I've
let it alone):
||3.0||
|[branch|https://github.com/pcmanus/cassandra/commits/10631-3.0]|
|[utests|http://cassci.datastax.com/job/pcmanus-10631-3.0-testall/4/]|
|[dtests|http://cassci.datastax.com/job/pcmanus-10631-3.0-dtest/1/]|
The unit test had 2 failures but those are junit timeouts so I suspect this is
just a case of being unluky. I've restarted the job one more time to check
however.
> JSON Update not working with PreparedStatement
> ----------------------------------------------
>
> Key: CASSANDRA-10631
> URL: https://issues.apache.org/jira/browse/CASSANDRA-10631
> Project: Cassandra
> Issue Type: Bug
> Components: CQL
> Environment: Windows 7, Datastax Java Driver 2.2.0-rc2
> Reporter: Henrik Karlsson
> Assignee: Sylvain Lebresne
> Fix For: 2.2.x
>
> Attachments: test-json.zip
>
>
> When using PreparedStatement to insert and update a row with JSON the first
> "INSERT INTO {tablename} JSON ?" statement works OK. But when calling it a
> second time with a changed value on a field the field is not updated. If I
> use a SimpleStatement instead ("INSERT INTO {tablename} JSON '"+json+"'") the
> modified field is updated as expected.
> Attaching a test project that shows the problem.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
