[
https://issues.apache.org/jira/browse/HBASE-696?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12609357#action_12609357
]
jimk edited comment on HBASE-696 at 6/30/08 1:32 PM:
--------------------------------------------------------------
stack wrote:
> Remove bloomfilter options. Only one bloomfilter type makes sense in hbase
> context.
True.
> Also, make bloomfilter self-sizing; you know size when flushing.
However, you can't easily know the size when doing a compaction.
Question: Is the bloomfilter based on the row key; row and column; row, column
and timestamp; row and timestamp?
It seems as if basing the bloomfilter solely on the row key would be the most
useful. If you are doing a get or scan with LATEST_TIMESTAMP, that won't match
anything in the bloomfilter if the timestamp is included. Similarly
row/family:member doesn't make sense if you are fetching by column wildcard
(family: ).
Using row/family: might be another option.
>Putting in 0.2 for now because its API change (for the simpler). We can punt
>later.
With respect to the API change, would it be sufficient to change
HColumnDescriptor so that bloomFilter is a boolean ? That would require a
migration step.
BloomFilterDescriptor could then be moved to
org.apache.hadoop.hbase.regionserver and become package private.
was (Author: jimk):
stack wrote:
> Remove bloomfilter options. Only one bloomfilter type makes sense in hbase
> context.
True.
> Also, make bloomfilter self-sizing; you know size when flushing.
However, you can't easily know the size when doing a compaction.
Question: Is the bloomfilter based on the row key; row and column; row, column
and timestamp; row and timestamp?
It seems as if basing the bloomfilter solely on the row key would be the most
useful. If you are doing a get or scan with LATEST_TIMESTAMP, that won't match
anything in the bloomfilter if the timestamp is included. Similarly
row/family:member doesn't make sense if you are fetching by column wildcard
(family:).
Using row/family: might be another option.
>Putting in 0.2 for now because its API change (for the simpler). We can punt
>later.
With respect to the API change, would it be sufficient to change
HColumnDescriptor so that bloomFilter is a boolean ? That would require a
migration step.
BloomFilterDescriptor could then be moved to
org.apache.hadoop.hbase.regionserver and become package private.
> Make bloomfilter true/false and self-sizing
> -------------------------------------------
>
> Key: HBASE-696
> URL: https://issues.apache.org/jira/browse/HBASE-696
> Project: Hadoop HBase
> Issue Type: Improvement
> Reporter: stack
> Fix For: 0.2.0
>
>
> Remove bloomfilter options. Only one bloomfilter type makes sense in hbase
> context. Also, make bloomfilter self-sizing; you know size when flushing.
> Putting in 0.2 for now because its API change (for the simpler). We can punt
> later.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
