[ 
https://issues.apache.org/jira/browse/HBASE-11644?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Jonathan Hsieh updated HBASE-11644:
-----------------------------------

    Description: 
Form the design doc,  mob files are not involved in the normal HBase compaction 
process.  This means deleted mobs would still take up space and that we never 
really merge mob files that accrue over time.   Currently, MOBs depend on two 
external tools:

1) A TTL cleaner that removes mobs that have passed their TTL or exceeded 
minVersions.
2) A 'sweep tool' cleaner that remove mobs that have had their references 
deleted and merges small files into larger ones.  

Today the tools are triggered by admins.  The longer term goal would be to 
integrate them into hbase such that by default mobs are cleaned.  The tools 
will be preserved however so that advanced admins can disable automatic 
cleanups and manually trigger these compaction like operaitons.  #1 would 
likely be a chore in the master while #2 requires some design work to integrate 
into hbase.

  was:The MOB files are involved in the HBase compaction. It means there's no 
chance to delete and merge the MOB files. The external tools do this, one is a 
cleaner to clean the MOB files that are expired (by TTL and minVersions), the 
other one is a sweep tool to clean the deleted Cells in HBase and merge small 
files into bigger ones. These tools are triggered by users. Besides, the 
cleaner could be a chore in HMaster.


> External MOB compaction tools
> -----------------------------
>
>                 Key: HBASE-11644
>                 URL: https://issues.apache.org/jira/browse/HBASE-11644
>             Project: HBase
>          Issue Type: Sub-task
>          Components: Compaction, master
>            Reporter: Jingcheng Du
>            Assignee: Jingcheng Du
>
> Form the design doc,  mob files are not involved in the normal HBase 
> compaction process.  This means deleted mobs would still take up space and 
> that we never really merge mob files that accrue over time.   Currently, MOBs 
> depend on two external tools:
> 1) A TTL cleaner that removes mobs that have passed their TTL or exceeded 
> minVersions.
> 2) A 'sweep tool' cleaner that remove mobs that have had their references 
> deleted and merges small files into larger ones.  
> Today the tools are triggered by admins.  The longer term goal would be to 
> integrate them into hbase such that by default mobs are cleaned.  The tools 
> will be preserved however so that advanced admins can disable automatic 
> cleanups and manually trigger these compaction like operaitons.  #1 would 
> likely be a chore in the master while #2 requires some design work to 
> integrate into hbase.



--
This message was sent by Atlassian JIRA
(v6.2#6252)

Reply via email to