[
https://issues.apache.org/jira/browse/HBASE-18715?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16555016#comment-16555016
]
xijiawen commented on HBASE-18715:
----------------------------------
xuge V5 87!
> isolate large query
> -------------------
>
> Key: HBASE-18715
> URL: https://issues.apache.org/jira/browse/HBASE-18715
> Project: HBase
> Issue Type: New Feature
> Reporter: chenxu
> Assignee: chenxu
> Priority: Major
> Attachments: HBASE-18715-master-v1.patch,
> HBASE-18715-master-v2.patch, HBASE-18715-master-v3.patch,
> HBASE-18715-master-v4.patch, isolate large query.pdf, isolate result.png
>
>
> large queries are very expensive than the general operation, and often delay
> insensitive.
> if not isolate, Handlers may be all used by the large query,
> lead to _Put_ or _Get_ slowly, so there need be an mechanism to isolate the
> large query.
> for example we can use a separate _RpcExecutor_ to handle this.
> for the following type of queries we can think it is a large query:
> (1) _Scan_ has no startkey or endkey
> (2) Client call _ResultScanner.next()_ more than a certain threshold
> (3) Client call custom coprocessor which involves large queries(such as
> _AggregateService_)
> for large queries, RegionServer should meet the following requirements
> (1) dynamic configure if isolate large query is turn on
> (2) setup certain Handlers to deal with large queries only
> (3) in offpeak hours, large queries are not isolate
> (4) dynamic configure if cache Blocks is turn on if the query is a large query
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
