[
https://issues.apache.org/jira/browse/CASSANDRA-17623?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17539505#comment-17539505
]
Doug Rohrer commented on CASSANDRA-17623:
-----------------------------------------
I was the primary author - Yifan figured out that it affected more than just
the CQLSSTableWriter path and added the test - Francisco and Yifan both paired
with me for a few hours working on making sure the fix was right, improving
things like making all of the Map code require a SortedMap implementation vs.
just fixing the one line that fixed the bug, etc.
So yeah, I think so, but not sure how we'd normally credit that work?
> Frozen maps may be serialized unsorted, causing inability to query later
> ------------------------------------------------------------------------
>
> Key: CASSANDRA-17623
> URL: https://issues.apache.org/jira/browse/CASSANDRA-17623
> Project: Cassandra
> Issue Type: Bug
> Components: CQL/Semantics
> Reporter: Doug Rohrer
> Assignee: Doug Rohrer
> Priority: Normal
> Fix For: 4.0.x, 4.1-beta, 4.x
>
> Time Spent: 2h 40m
> Remaining Estimate: 0h
>
> CASSANDRA-7859 introduced the ability to use frozen collections as parts of
> primary keys. This +requires+ all frozen maps to be persisted with their
> entries sorted by the map keys. If the map is +not+ sorted correctly, it
> becomes impossible to project all of the map values out of the map using the
> map projection/selection syntax. For example, the select below would fail if
> the map was not sorted correctly and the higher-valued key was persisted
> first:
> {code:sql}
> CREATE TABLE test.test (k text, c frozen<map<text, text>>, PRIMARY KEY (k,
> c));
> INSERT INTO test.test (k, c) VALUES ('key', {'z':'second_value',
> 'a':'first_value'});
> SELECT k, c['a'] from test.test where k='key' -- c['a'] would return NULL here
> {code}
> Additionally, if you attempted to select just that row by using the complete
> map value in a WHERE clause, which is also supported, it would return no rows
> unless the map provided by the query processor just happened to be sorted the
> same way as the persisted value.
> However, there is a bug in Maps.java where we don't actually use a SortedMap
> in {{Maps.Value#fromSerialized}}, which manifests if a client sends an
> unsorted map as a bound parameter to a query on insert or select. In either
> case, the map may not be sorted correctly, leading to either invalid data
> being persisted to disk (in the INSERT case) or the query not being able to
> be executed/returning 0 rows even though a row _should_ exist (SELECT).
> This bug affects any usage of parameterized queries (tested with the DataStax
> driver, and was originally discovered when using the CQLSSTableWriter code to
> write data locally).
--
This message was sent by Atlassian Jira
(v8.20.7#820007)
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org
For additional commands, e-mail: commits-h...@cassandra.apache.org
