[
https://issues.apache.org/jira/browse/HBASE-19645?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16304672#comment-16304672
]
Anoop Sam John commented on HBASE-19645:
----------------------------------------
1. Off heap MSLAB can be there which makes off heap buffers and keep in pool
2. Off heap bucket cache having DBB pools
3. Short circuit reads uses
4. RPC side we have BBPool which will have fixed sized BBs and a max count for
this
5. The RPC client/server impl internally uses Netty / NIO which again creates
on demand DBBs and even can pool certain number.
All these we have to consider in this calc and decision.
> Add global barrier for offheap usage
> ------------------------------------
>
> Key: HBASE-19645
> URL: https://issues.apache.org/jira/browse/HBASE-19645
> Project: HBase
> Issue Type: Task
> Reporter: stack
>
> See discussion at the end of HBASE-16747. In synopsis, we do not have a
> global limit for offheap usage; we do no global accounting. There are the
> offheap buffers, short-circuit read buffers, offheap memstores, etc. Ideally,
> as for onheap, when we approach threshold, we'd bring on flushing to clear
> out offheap usage or make use of onheap. TODO.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
