hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "He Liangliang (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HBASE-13984) Add option to allow caller to know the heartbeat and scanner position when scanner timeout
Date Fri, 03 Jul 2015 09:18:05 GMT

    [ https://issues.apache.org/jira/browse/HBASE-13984?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14613045#comment-14613045

He Liangliang commented on HBASE-13984:

>> why is this not an option you'd specify on Scan?
It's flexible to put into scan, includes the timeout limit. Explicit means the heartbeat is
visible to the caller instead of transparent.

>> It seems entangled with client not taking partial results?
It makes sense, I'll update this.

>> These 'general' additions made just for the exotic case of an heartbeat carrying
the next Cell from which to carry on the Scan seems like too much exposure on a pivotal class
like ScannerCallable
We need to read this 'next' in ClientScanner, similar to heartbeatMessage filed in ScannerCallable.
Or in another word, it's piggybacked data in a heartbeat message, so we can wrap heartbeatMessage
flag and next together into an object and put it in ScannerCallable?

> Add option to allow caller to know the heartbeat and scanner position when scanner timeout
> ------------------------------------------------------------------------------------------
>                 Key: HBASE-13984
>                 URL: https://issues.apache.org/jira/browse/HBASE-13984
>             Project: HBase
>          Issue Type: Improvement
>          Components: Scanners
>            Reporter: He Liangliang
>            Assignee: He Liangliang
>         Attachments: HBASE-13984-V1.diff
> HBASE-13090 introduced scanner heartbeat. However, there are still some limitations (see
HBASE-13215). In some application, for example, an operation access hbase to scan table data,
and there is strict limit that this call must return in a fixed interval. At the same time,
this call is stateless, so the call must return the next position to continue the scan. This
is typical use case for online applications.
> Based on this requirement, some improvements are proposed:
> 1. Allow client set a flag whether pass the heartbeat (a fake row) to the caller (via
ResultScanner next)
> 2. Allow the client pass a timeout to the server, which can override the server side
default value
> 3. When requested by the client, the server peek the next cell and return to the client
in the heartbeat message

This message was sent by Atlassian JIRA

View raw message