hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Enis Soztutar (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HBASE-10244) store mvcc in store files per KV for longer time, discard during compactions based on scanner timeout
Date Mon, 06 Jan 2014 22:35:51 GMT

    [ https://issues.apache.org/jira/browse/HBASE-10244?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13863526#comment-13863526
] 

Enis Soztutar commented on HBASE-10244:
---------------------------------------

Yes, we are storing mvcc numbers in hfiles, but since the mvcc numbers lifecycle , and a region
scanner's lifecycle is bound by the RS lifetime we are just using the lowest mvcc read number
of all the open scanners to manage the low watermark. I agree that we can implement a timeout
based low watermark tracking. The client-side open scanners will receive the mvcc number.
They would have to connect to the next region in an absolute time-bound (like 30 min), otherwise
they will fail the scan completely. The RS then can manage a time window for mvcc numbers
to expire the mvcc read numbers since any scanner not connecting for the last 30 min will
be rejected.

> store mvcc in store files per KV for longer time, discard during compactions based on
scanner timeout
> -----------------------------------------------------------------------------------------------------
>
>                 Key: HBASE-10244
>                 URL: https://issues.apache.org/jira/browse/HBASE-10244
>             Project: HBase
>          Issue Type: Sub-task
>          Components: Compaction, regionserver
>            Reporter: Sergey Shelukhin
>            Priority: Minor
>
> -Initially, we can store via KV tag. For perf reasons, it might make sense to make it
a first-class member of KV, but that would require HFileV4-



--
This message was sent by Atlassian JIRA
(v6.1.5#6160)

Mime
View raw message