[
https://issues.apache.org/jira/browse/HBASE-517?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
stack resolved HBASE-517.
-------------------------
Resolution: Won't Fix
We won't do this.
On investigation, the thought is that the cost -- development/design and
probably at runtime -- of a glue layer would be too high. For example, mapfile
and hfile have different ways of expressing compression. In the former, the
algorithm is hardcoded and there is the notion of record/block or none. In
hfile, its only block compression and you choose the algorithm to use. Or,
hfile, if no matching record returns the closest record before; mapfile the
closest after. The main knob to twiddle improving mapfile performance is index
interval. In hfile, its block size. The dissonance between just these two
types would be hard to resolve in an agreeable interface, an interface that has
to be user presentable. Punting on this, at least for now.
> Define an interface that will apply to any possible store file implementation
> -----------------------------------------------------------------------------
>
> Key: HBASE-517
> URL: https://issues.apache.org/jira/browse/HBASE-517
> Project: Hadoop HBase
> Issue Type: Sub-task
> Reporter: Bryan Duxbury
> Assignee: stack
> Fix For: 0.20.0
>
> Attachments: 517-v2.patch, 517.patch
>
>
> Step 1: define what methods we need from a concrete implementation. Once we
> have figured out what the interface should look like, it will be easier to
> imagine structures that fulfill the requirements.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
