[jira] [Commented] (PHOENIX-2859) RPC Heartbeats for aggregation coprocessors

["Samarth Jain (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/PHOENIX-2859?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15259125#comment-15259125 ] 

Samarth Jain commented on PHOENIX-2859:
---------------------------------------

[~ndimiduk] - could/should this be tied with renewing scanner leases? Does renewing a scanner lease today also reset the time tracker for RPC (a cursory look at the HBase code doesn't say so) ? We already have a framework in Phoenix to renew scanner leases. May be we can utilize it for RPC heartbeats too. 

> RPC Heartbeats for aggregation coprocessors
> -------------------------------------------
>
>                 Key: PHOENIX-2859
>                 URL: https://issues.apache.org/jira/browse/PHOENIX-2859
>             Project: Phoenix
>          Issue Type: Improvement
>            Reporter: Nick Dimiduk
>
> Phoenix supports a range of analytical functions, including {{count}}, {{sum}}, {{min}}, {{max}}, &c. Many of these are heavily supported by aggregation coprocessors. When the Phoenix table contains any significant amount of data, these aggregation functions can take a while. Mostly these queries will be killed by phoenix query timeout or hbase rpc timeout. Both of these settings must be increased in order to support these aggregation queries at all. This is pretty annoying because there's two configs managing similar things.
> We could better support these kinds of queries if our coprocessors supported a heart-beat functionality, similar to what was added in for HBase scanners in HBASE-13090. This lets us leave HBase's RPC timeout as the default and let queries be managed by Phoenix's query timeout. That keeps the cluster snappy and responsive for usual workloads, but still supports basic aggregations.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)