Clemens Valiente created IMPALA-6864:
----------------------------------------
Summary: COMPUTE STATS overwrites table schema in Hive Metastore
Key: IMPALA-6864
URL: https://issues.apache.org/jira/browse/IMPALA-6864
Project: IMPALA
Issue Type: Bug
Components: Catalog
Affects Versions: Impala 2.9.0
Reporter: Clemens Valiente
If Impala does not have the latest schema information from Hive, a COMPUTE
STATS overwrites the table definition stored in Hive Metastore.
short example:
{code:sql}
# Hive
create table cvaliente.testschema (id int);
{code}
{code:sql}
# Impala
INVALIDATE METADATA cvaliente.testschema;
{code}
{code:sql}
# Hive
drop table cvaliente.testschema;
create table cvaliente.testschema (id int, value string);
{code}
{code:sql}
# Impala
COMPUTE STATS cvaliente.testschema;
{code}
{code:sql}
# Hive
show columns from cvaliente.testschema
{code}
In our case, due to the size and scope of our table we did {{REFRESH TABLE
dbname.tblname PARTITION (partkey = partvalue)}} in impala
followed by {{COMPUTE INCREMENTAL STATS dbname.tblname PARTITION (partkey =
partvalue)}}
REFRESH TABLE ... PARTITION does not update the table schema unlike a simple
REFRESH PARTITION so it kept the old schema and the COMPUTE INCREMENTAL
STATS... PARTITION then updated the full table back with the old schema in Hive
metastore.
- a REFRESH TABLE ... PARTITION should still update the table metadata from
hive metastore
- a COMPUTE STATS should only write column statistics back to the metastore and
NOT update the table schema.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
