Eugene Koifman commented on HIVE-14779:

If the client is something like CLI, then if VM crashes - all threads go away.
If the client thread just hangs forever, there is no way to know that the 
client died - as long as hearbeat is done in a separate thread (wherever it is) 
it will go on.

For HS2, the Sessions are cleaned up automatically when client goes away.  
There are several variables that can be configured to define what "goes away" 
mean.  Depending on how the client connects, it may be immediate (TCP 
connection broken) or based on inactivity period for HTTP based connections.

> make DbTxnManager.HeartbeaterThread a daemon
> --------------------------------------------
>                 Key: HIVE-14779
>                 URL: https://issues.apache.org/jira/browse/HIVE-14779
>             Project: Hive
>          Issue Type: Bug
>          Components: Transactions
>    Affects Versions: 1.3.0, 2.1.0
>            Reporter: Eugene Koifman
>            Assignee: Eugene Koifman
>            Priority: Minor
>         Attachments: HIVE-14779.patch
> setDaemon(true);
> make heartbeaterThreadPoolSize static 

This message was sent by Atlassian JIRA

Reply via email to