hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "stack (JIRA)" <j...@apache.org>
Subject [jira] [Updated] (HBASE-11581) Add option so CombinedBlockCache L2 can be null (fscache)
Date Fri, 08 Aug 2014 14:38:12 GMT

     [ https://issues.apache.org/jira/browse/HBASE-11581?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

stack updated HBASE-11581:
--------------------------

      Resolution: Invalid
    Release Note:   (was: Add hbase.combinedblockcache.l2.null configuration.  When set to
true, BlockCache uses CombinedBlockCache where L1 is instance of LruBlockCache carrying META
blocks and L2 is null: i.e. DATA blocks are fetched each time.)
          Status: Resolved  (was: Patch Available)

Resolving as won't fix. I can get the desired effect -- caching of META blocks only and NOT
caching DATA blocks -- by setting BLOCKCACHE => 'false' (since HBASE-4683 "Always cache
index and bloom blocks").  Testing I was confused by metrics showing DATA blocks count but
the number was a persist from before the configuration change (Need to look into why metric
not reset....that is another issue).

> Add option so CombinedBlockCache L2 can be null (fscache)
> ---------------------------------------------------------
>
>                 Key: HBASE-11581
>                 URL: https://issues.apache.org/jira/browse/HBASE-11581
>             Project: HBase
>          Issue Type: New Feature
>          Components: BlockCache
>            Reporter: stack
>            Assignee: stack
>            Priority: Minor
>              Labels: beginner, beginners
>         Attachments: 11581.txt
>
>
> Add option, mostly for comparison's sake, that allows a deploy orchestrated by CombinedBlockCache
such that its L1 is LruBlockCache for META blocks but DATA blocks are fetched each time (we
don't try and cache them, no blockcache churn).
> In operation, i can see fscache coming around to cover the fetched DATA blocks such that
if the DATA blocks fit in fscache, seeks go to zero.
> This setup for sure runs slower.  Will publish numbers elsewhere.  Meantime, here is
a patch to enable this option.



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

Mime
View raw message