Sanjeet Malhotra created PHOENIX-7282:
-----------------------------------------
Summary: Incorrect data in index column for corresponding BIGIT
type column in data table
Key: PHOENIX-7282
URL: https://issues.apache.org/jira/browse/PHOENIX-7282
Project: Phoenix
Issue Type: Bug
Affects Versions: 5.3.0
Reporter: Sanjeet Malhotra
If we add a new column of type BIGINT to an existing data table and use CASCADE
INDEX option then the column in index is aded as DECIMAL type. If we query such
a column and the query plan gets resolved to use index table instead of data
table then value returned will be different from the case when data table would
have been used by query plan.
IT to reproduce:
{code:java}
@Test
public void testBigIntData() throws Exception {
String dataTableName = generateUniqueName();
String indexName = generateUniqueName();
try(Connection conn = DriverManager.getConnection(getUrl())) {
conn.createStatement().execute("create table " + dataTableName
+ " (id varchar not null primary key, col1 integer)");
conn.createStatement().execute("create index " + indexName + " on " +
dataTableName + " (col1)");
conn.createStatement().execute("alter table " + dataTableName + " add
if not exists col3 bigint cascade index all");
conn.createStatement().execute("upsert into " + dataTableName + " (id,
col3) values ('a', 3)");
conn.commit();
ResultSet rs = conn.createStatement().executeQuery("select col3 from "
+ dataTableName);
while(rs.next()) {
System.out.println(rs.getObject(1));
}
}
} {code}
{{So far this issue has been observed when new column of type BIGINT is added
to data table/view. But if there is already a column of type BIGINT in the data
table.view then above error is not observed even if query uses index table as
per query plan.}}
{{Further findings so far:}}
# {{During alter table when we add new column in data table/view of type
BIGINT then we also add a column of type DECIMAL (and not BIGINT) to
corresponding index/view index.}}
# {{Above finding is not true for create table but alter table only.}}
# {{We write to data table/view in column of BIGIT type and also write same
*byte array* in index/view index also but in a column of DECIMAL type. Byte
array written in data table at HBase layer was serialized via {{PLong}} class
but as we write same byte array in index column so at the time of reading value
(from index table as per query plan) the byte array gets desrialized by
{{PDecimal}} class. As we are not using compatible serializaion and
deserialization logic for index column so, a value in data table becomes
totally another value in index table when read. Serialization logic of PLong
and PDecimal are completely different thus, so are their deserialization
logics.}}
## {{One such example we saw was that in above IT (for reproducing the error)
we insert 3 in data table column (of type BIGINT and newly added via alter) but
the corresponding desrialized value in index is `-1.010101010098E+126`}}
{{Still need to figure out why this issue only happens for new columns added
via alter DDL but not via create DDL at table/view creation time.}}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
