[
https://issues.apache.org/jira/browse/HBASE-4?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12565467#action_12565467
]
Bryan Duxbury commented on HBASE-4:
-----------------------------------
We should make creating an HTable a cheap operation. In fact, I'm surprised
that it's a problem at all, since the region locations should be cached in the
connection. It's possible that the constructor is just doing something stupid.
In any case, we could certainly change GenericHandler.getTable to cache tables.
The only downside of that is that you can only have one put in progress at a
time, at least until the new BatchUpdate patch goes in.
> hbase client makes too many requests against .META.,,1 table: caching of
> .META.,,1 lookups is broken?
> -----------------------------------------------------------------------------------------------------
>
> Key: HBASE-4
> URL: https://issues.apache.org/jira/browse/HBASE-4
> Project: Hadoop HBase
> Issue Type: Bug
> Reporter: Michael Bieniosek
>
> We're running a high load of random reads through a single hbase rest server.
> When I look at the 60010 master servlet, there are a very high number of
> hits to the server hosting .META.,,1. That server has 4000 requests in the
> last "3 seconds", the server hosting ROOT (and two other regions which
> shouldn't be getting reads) has 80 requests, and all the other regionservers
> have less.
> This doesn't seem right: shouldn't the single hbase client be caching table
> lookups?
> This appears to be a hurdle for scaling our random read rate.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
