Samarth Jain commented on PHOENIX-3511:

[~singamteja], apologies for not being clearer about what that threshold really 
means. It is basically used by a scheduled thread to check whether the scanner 
is due for lease renewal or not. So if the time since last renewal is greater 
than this threshold, then the thread attempts to renew the lease of the 
scanner. With default value of client scanner timeout, lease renewal will be 
attempted if the duration since last attempt has been > 0.75 minutes. 

My question for you would be - have you seen MR jobs fail with client scanner 
timeout exceptions? If yes, I would probably change the 
phoenix.scanner.lease.threshold to a lower value like 30 seconds and 
phoenix.scanner.lease.renew.interval to 15 seconds. This would cause attempts 
to renew leases to be made more frequently. You don't want to set it too low a 
value though since every renew lease call is an RPC and making too many of them 
will prevent HBase from doing the actual work.

> Async Secondary index MR job fails for large data > 200 M records
> -----------------------------------------------------------------
>                 Key: PHOENIX-3511
>                 URL: https://issues.apache.org/jira/browse/PHOENIX-3511
>             Project: Phoenix
>          Issue Type: Bug
>    Affects Versions: 4.9.0
>            Reporter: Loknath Priyatham Teja Singamsetty 
>            Assignee: Loknath Priyatham Teja Singamsetty 
>             Fix For: 4.9.0, 4.9.1
>         Attachments: phoenix-3511-v2.patch, phoenix-3511.patch

This message was sent by Atlassian JIRA

Reply via email to