Both fix for HIVE-13090 and HIVE-15090 are in CDH5.7.6 , so we can upgrade to this version to avoid this issue.
Alternatively, we have below workarounds if upgrade cannot be scheduled at this time.
As far as you don't use load balancer before the HS2 instances, we could change the token store to MemoryTokenStore just for HiveServer2, but leave DBTokenStore for HMS.
To achieve this, please do the following:
1) Navigate to CM > Hive > Configuration > Hive MetaStore Server > Hive Metastore Delegation Token Store
2) Select DBTokenStore (should already be done)
3) Navigate to CM > Hive > Configuration > HiveServer2 > HiveServer2 Advanced Configuration Snippet (Safety Valve) for hive-site.xml
6) Restart Hive.
In case of Hive Available load balanced HS2 setup, we could change the token store to org.apache.hadoop.hive.thrift.ZooKeeperTokenStore instead.
Note: Cloudera is testing the use of org.apache.hadoop.hive.thrift.ZooKeeperTokenStore in the use of Hive Server 2 load balanced environments. If you experience problems with this method, Cloudera suggests that you not target the load balancer in the jdbc-url for the Hive2 action, but target a specific HiveServer2 instance behind the load balancer instead