[
https://issues.apache.org/jira/browse/PHOENIX-3454?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Samarth Jain updated PHOENIX-3454:
----------------------------------
Attachment: Screen Shot 2016-11-04 at 1.29.43 PM.png
See the attached debugger screenshot for PhoenixIndexBuilder.executeAtomicOp().
The reason is because the the mutations returned by PRowImpl.toRowMutations()
isn't returning a sorted list of mutations. The code in
PhoenixIndexBuilder.executeAtomicOp() inherently expects the mutations to be
sorted wrt column qualifier cells since it is doing the following:
{code}
for (Mutation source : mutations) {
flattenCells(source, flattenedCells);
}
tuple.setKeyValues(flattenedCells);
{code}
One option would be to sort the list of flattened cells over in
PhoenixIndexBuilder before setting in the tuple. But then
PRowImpl.toRowMutations() is called in other places too.
I ended up discovering this as part of my work for encoding column qualifiers
where we use the column qualifier "0" instead of "_0" for the empty key value
column.
> ON DUPLICATE KEY construct doesn't work correctly when using lower case
> column names
> ------------------------------------------------------------------------------------
>
> Key: PHOENIX-3454
> URL: https://issues.apache.org/jira/browse/PHOENIX-3454
> Project: Phoenix
> Issue Type: Bug
> Reporter: Samarth Jain
> Attachments: Screen Shot 2016-11-04 at 1.29.43 PM.png
>
>
> See this test case for a repro:
> {code}
> @Test
> public void testDeleteOnSingleLowerCaseVarcharColumn() throws Exception {
> Properties props = PropertiesUtil.deepCopy(TEST_PROPERTIES);
> Connection conn = DriverManager.getConnection(getUrl(), props);
> conn.setAutoCommit(false);
> String tableName = generateUniqueName();
> String ddl = " create table " + tableName + "(pk varchar primary key,
> \"counter1\" varchar, \"counter2\" smallint)";
> conn.createStatement().execute(ddl);
> String dml = "UPSERT INTO " + tableName + " VALUES('a','b') ON
> DUPLICATE KEY UPDATE \"counter1\" = null";
> conn.createStatement().execute(dml);
> conn.createStatement().execute(dml);
> conn.commit();
> ResultSet rs = conn.createStatement().executeQuery("SELECT * FROM " +
> tableName);
> assertTrue(rs.next());
> assertEquals("a",rs.getString(1));
> assertEquals(null,rs.getString(2));
> assertFalse(rs.next());
>
> dml = "UPSERT INTO " + tableName + " VALUES('a','b',0)";
> conn.createStatement().execute(dml);
> dml = "UPSERT INTO " + tableName + " VALUES('a','b', 0) ON DUPLICATE
> KEY UPDATE \"counter1\" = null, \"counter2\" = \"counter2\" + 1";
> conn.createStatement().execute(dml);
> dml = "UPSERT INTO " + tableName + " VALUES('a','b', 0) ON DUPLICATE
> KEY UPDATE \"counter1\" = 'c', \"counter2\" = \"counter2\" + 1";
> conn.createStatement().execute(dml);
> conn.commit();
> rs = conn.createStatement().executeQuery("SELECT * FROM " +
> tableName);
> assertTrue(rs.next());
> assertEquals("a",rs.getString(1));
> assertEquals("c",rs.getString(2));
> assertEquals(2,rs.getInt(3));
> assertFalse(rs.next());
> conn.close();
> }
> {code}
> After changing the column names to upper case (or removing the quotes), the
> test passes.
> FYI, [~jamestaylor]
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
