Sergey Shelukhin commented on HIVE-18281:

Left comments on RB, mostly minor.
I am not sure about timing/races w.r.t. leadership switching. If activities 
during leadership switch take too long should they time out? etc. I'm 
hoping/assuming the latch itself has some grace period that should be 
configurable and/or accounted for.

There also appear to be some unrelated logic changes in some places.

> HiveServer2 HA for LLAP and Workload Manager
> --------------------------------------------
>                 Key: HIVE-18281
>                 URL: https://issues.apache.org/jira/browse/HIVE-18281
>             Project: Hive
>          Issue Type: New Feature
>    Affects Versions: 3.0.0
>            Reporter: Prasanth Jayachandran
>            Assignee: Prasanth Jayachandran
>            Priority: Major
>         Attachments: HIVE-18281.1.patch, HIVE-18281.2.patch, 
> HIVE-18281.WIP.patch, HSI-HA.pdf
> When running HS2 with LLAP and Workload Manager, HS2 becomes single point of 
> failure as some of the states for workload management and scheduling are 
> maintained in-memory. 
> The proposal is to support Active/Passive mode of high availability in which, 
> all HS2 and tez AMs registers with ZooKeeper and a leader have to be chosen 
> which will maintain stateful information. Clients using service discovery 
> will always connect to the leader for submitting queries. There will also be 
> some responsibilities for the leader, failover handling, tez session 
> reconnect etc. Will upload some more detailed information in a separate doc. 

This message was sent by Atlassian JIRA

Reply via email to