Aaron Gottlieb created HIVE-20173:
-------------------------------------
Summary: MetaStoreDirectSql#executeWithArray should not catch
RuntimeExceptions from JDO
Key: HIVE-20173
URL: https://issues.apache.org/jira/browse/HIVE-20173
Project: Hive
Issue Type: Bug
Components: Metastore
Affects Versions: 1.1.0
Reporter: Aaron Gottlieb
When attempting to test the existence of a Hive database, the Metastore will
query the backing database.
The method MetaStoreDirectSql#executeWithArray catches all exceptions, turning
them into MetaExceptions.
Further up the stack, the ObjectStore#getDatabase explicitly catches
MetaExceptions and turns them into NoSuchObjectExceptions.
Finally, RetryingHMSHandler explicitly looks for NoSuchObjectExceptions and
does _not_ retry them, thinking they are legitimate answers.
If the exception in MetaStoreDirectSql#executeWithArray was a runtime
JDOException due to, say, some sort of network error between the Metastore and
the backing database, this inability to query the backing database looks just
like an answer of "no database exists" higher up the stack. Any program
depending on this information will continue with an incorrect answer rather
than retrying the original getDatabase query.
I am unsure the extent of the effects of this, but I imagine that explicitly
_not_ catching RuntimeExceptions in MetaStoreDirectSql#executeWithArray will
allow the exception to raise all the way up to the RetryingHMSHandler which
will, correctly, retry the operation.
Would allowing RuntimeExceptions to be thrown from
MetaStoreDirectSql#executeWithArray be too deleterious? Or did I miss some
code path such that my observations are incorrect?
Thanks,
Aaron
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
