[
https://issues.apache.org/jira/browse/PHOENIX-1108?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14073006#comment-14073006
]
Gabriel Reid commented on PHOENIX-1108:
---------------------------------------
Related mail discussion here: http://s.apache.org/FFe
> Clarify, verify, and document intended behavior from using
> HColumnDescriptor.KEEP_DELETED_CELLS
> -----------------------------------------------------------------------------------------------
>
> Key: PHOENIX-1108
> URL: https://issues.apache.org/jira/browse/PHOENIX-1108
> Project: Phoenix
> Issue Type: Improvement
> Reporter: Gabriel Reid
> Assignee: Gabriel Reid
>
> The current default for all Phoenix tables is to enable the
> KEEP_DELETED_CELLS flag on all column families. The general functionality of
> this default should be reviewed, as well as checking that it works as
> intended (particularly in terms of the ChunkedResultIterator, which uses
> multiple scans).
> The general idea of the KEEP_DELETED_CELLS flag is that it prevents deleted
> cells from being permanently removed during a (major) compaction. If the
> number of versions to keep for a cell is small (3 is the default) then this
> won’t cause a major problem, and is in might be needed in order to function
> correctly (i.e. to handle deletes and a major compaction occurring while a
> query is being run).
> On the other hand, if the number of versions to keep for a column family is
> large (e.g. Integer.MAX_VALUE), the default of KEEP_DELETED_CELLS=true will
> mean that a delete in Phoenix never actually deletes data.
> Tasks to be performed are:
> * clear up (and document) the intended behavior that of using
> KEEP_DELETED_CELLS=true as a default in Phoenix
> * add tests to verify that this intended behavior still works with the
> ChunkedResultIterator
> * document the implications and/or workaround if a large number of versions
> is configured for a column family
--
This message was sent by Atlassian JIRA
(v6.2#6252)
