[
https://issues.apache.org/jira/browse/CASSANDRA-10001?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14661162#comment-14661162
]
Stefania commented on CASSANDRA-10001:
--------------------------------------
Reproduced with equivalent unit test which can be added to
o.a.c.cql3.validation.entities.CollectionsTest:
{code}
@Test
public void testMapWithLargePartition() throws Throwable
{
Random r = new Random();
long seed = System.nanoTime();
System.out.println("Seed " + seed);
r.setSeed(seed);
int len = (1024 * 1024)/100;
createTable("CREATE TABLE %s (userid text PRIMARY KEY, properties
map<int, text>) with compression = {}");
final int numKeys = 200;
for (int i = 0; i < numKeys; i++)
{
byte[] b = new byte[len];
r.nextBytes(b);
execute("UPDATE %s SET properties[?] = ? WHERE userid = 'user'", i,
new String(b));
}
Object[][] rows = getRows(execute("SELECT properties from %s where
userid = 'user'"));
assertEquals(1, rows.length);
assertEquals(numKeys, ((Map)rows[0][0]).size());
}
{code}
{code}
java.lang.AssertionError:
Expected :200
Actual :126
<Click to see difference>
at org.junit.Assert.failNotEquals(Assert.java:689)
at
org.apache.cassandra.cql3.validation.entities.CollectionsTest.testMapWithLargePartition(CollectionsTest.java:56)
{code}
> Bug in encoding of sstables
> ---------------------------
>
> Key: CASSANDRA-10001
> URL: https://issues.apache.org/jira/browse/CASSANDRA-10001
> Project: Cassandra
> Issue Type: Bug
> Reporter: T Jake Luciani
> Assignee: Stefania
> Priority: Blocker
> Fix For: 3.0 beta 1
>
>
> Fixing the compaction dtest I noticed we aren't encoding map data correctly
> in sstables.
> The following code fails from newly committed {{
> compaction_test.py:TestCompaction_with_SizeTieredCompactionStrategy.large_compaction_warning_test}}
> {code}
> session.execute("CREATE TABLE large(userid text PRIMARY KEY, properties
> map<int, text>) with compression = {}")
> for i in range(200): # ensures partition size larger than
> compaction_large_partition_warning_threshold_mb
>
>
> session.execute("UPDATE ks.large SET properties[%i] = '%s' WHERE
> userid = 'user'" % (i, get_random_word(strlen)))
> ret = session.execute("SELECT properties from ks.large where userid =
> 'user'")
> assert len(ret) == 1
> self.assertEqual(200, len(ret[0][0].keys()))
> {code}
> The last assert is failing with only 91 keys. The large values are causing
> flushes vs staying in the memtable so the issue is somewhere in the
> serialization of collections in sstables.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
