[
https://issues.apache.org/jira/browse/HBASE-11729?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14098543#comment-14098543
]
Sean Busbey commented on HBASE-11729:
-------------------------------------
bq. Also FYI.. During compaction, if all the files under compaction having max
tags length as 0, we even skip the 2 bytes tags length part with every KV and
saving that overhead. Absence of MAX_TAGS_LENGTH file info means the Cell
format is just like as in HFile V2
Okay, to make sure I understand let me try restating things.
1) If an HFile has a version number of 3.0+, I need to read the File Info Block
before I can read any Data Blocks.
2) If the File Info Block contains an entry for MAX_TAGS_LENGTH, then KeyValues
will be serialized as KeyValues + Tags (even if that length is 0). Otherwise,
they'll be serialized the same as in earlier versions.
3) a File Info Block should only contain a TAGS_COMPRESSED entry if it has a
MAX_TAGS_LENGTH entry.
Is that all correct?
> Document HFile v3
> -----------------
>
> Key: HBASE-11729
> URL: https://issues.apache.org/jira/browse/HBASE-11729
> Project: HBase
> Issue Type: Task
> Components: documentation, HFile
> Affects Versions: 0.98.0
> Reporter: Sean Busbey
> Assignee: Sean Busbey
> Priority: Trivial
> Labels: beginner
> Attachments: HBASE-11729.patch, HBASE-11729.pdf
>
>
> 0.98 added HFile v3. There are a couple of mentions of it in the book on the
> sections on cell tags, but there isn't an actual overview or design
> explanation like there is for [HFile
> v2|http://hbase.apache.org/book/hfilev2.html].
--
This message was sent by Atlassian JIRA
(v6.2#6252)
