hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Andrew Purtell (Commented) (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HBASE-5164) Better HTable resource consumption in CoprocessorHost
Date Tue, 10 Jan 2012 04:43:41 GMT

    [ https://issues.apache.org/jira/browse/HBASE-5164?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13183069#comment-13183069
] 

Andrew Purtell commented on HBASE-5164:
---------------------------------------

We added getTable to the CP API with that risk in mind. The notion is CPs are not allowed
to just instantiate HTables, instead they ask for a more constrained server side IPC. No such
implementation yet exists, so getTable essentially is a placeholder. 

If we consider coprocessors hosted in an external process, but still on the server, the risks
are not much lower than in process. A stuck table client in the external CP will hold up the
initiator anyway, and in fact this makes external CP handling more problematic than if table
clients were simply not allowed server side at all. However I think many useful CP use cases
must have a table access scope wider than the single region.

What we need is a "lightweight" server to server IPC? Would present HTableInterface but lack
the fat client semantics. 

                
> Better HTable resource consumption in CoprocessorHost
> -----------------------------------------------------
>
>                 Key: HBASE-5164
>                 URL: https://issues.apache.org/jira/browse/HBASE-5164
>             Project: HBase
>          Issue Type: Sub-task
>          Components: coprocessors
>            Reporter: Lars Hofhansl
>            Priority: Minor
>             Fix For: 0.94.0
>
>
> HBASE-4805 allows for more control over HTable's resource consumption.
> This is currently not used by CoprocessorHost (even though it would even be more critical
to control this inside the RegionServer).
> It's not immediate obvious how to do that.
> Maybe CoprocessorHost should maintain a lazy ExecutorService and HConnection and reuse
both for all HTables retrieved via CoprocessorEnvironment.getTable(...).
> Not sure how critical this is, but I feel without this it is dangerous to use getTable,
as it would lead to all resource consumption problems we find in the client, but inside a
crucial part of the HBase servers.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira

        

Mime
View raw message