[
https://issues.apache.org/jira/browse/ACCUMULO-3592?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14328087#comment-14328087
]
Josh Elser commented on ACCUMULO-3592:
--------------------------------------
bq. If you're suggesting that this change will not require future RFile version
number bumps when we add stuff to this structure, then yeah, let's do it now
Right. Things that we can naturally represent inside this schema-drifting data
structure (and, with proper use by the reader) would allow old readers to read
new files. For example, we add some new value to the header, hypothetically:
some density value for key-values in the file. The reader implementation can
use the density score to perform more optimized seeks/reads. The old readers
can still function without this new attribute in the header (and would parse it
without error), while new readers can read it (when present) and leverage it to
do things more efficiently.
It obviously doesn't protect us from any/all changes, but it should reduce the
cases in which we'd otherwise need to bump the RFile version.
> Improvements to RFile
> ---------------------
>
> Key: ACCUMULO-3592
> URL: https://issues.apache.org/jira/browse/ACCUMULO-3592
> Project: Accumulo
> Issue Type: Sub-task
> Reporter: Josh Elser
> Fix For: 1.7.0
>
>
> We have a small "header" at the beginning of the RFile index at the end of
> the file. This is currently just a few writable things thrown together -- it
> would be nice to convert this into some pb/thrift message that we can
> add/remove things easily.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
