Details
-
Bug
-
Status: Closed
-
Major
-
Resolution: Fixed
-
None
-
None
Description
Currently it caches IMetastoreClient on init() but the same connection is indirectly saved using ThreadLocal (since Hive.get() uses ThreadLocal).
So it may cause something other than DbTxnManager to use the same connection. It needs to use the same Metastore client as the caller
Attachments
Attachments
Issue Links
- is related to
-
HIVE-19587 HeartBeat thread uses cancelled delegation token while connecting to meta on KERBEROS cluster
- Patch Available
- supercedes
-
HIVE-15937 RetryingMetaStoreClient.invoke() should retry Trift exception even if not wrapped in MetaException
- Resolved