[
https://issues.apache.org/jira/browse/HBASE-25116?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17228334#comment-17228334
]
niuyulin commented on HBASE-25116:
----------------------------------
What is about the canary access that slows user-requests?
=>the meta access may slow user-requests
Should these options be made into command-line options for the Canary tool
rather than internal configs?
=>yes,made into command-line options is more friendly to users
A task maps to a Region? Would it help if we talked of sampling rather than
task counts?
=> Yes, a task per region, but I think sampling is more complicated than
task-count. normally the read task(user region) is 1w+ and write task(canary
region) is 1k- maybe 100 +, so we set task-count-max: 5k , task-count-min:2k,
and read task will be sampling and write task will be enlarged. The task
normalize will work only when task count is not in task count range we set
> RegionMonitor support RegionTask count normalize
> ------------------------------------------------
>
> Key: HBASE-25116
> URL: https://issues.apache.org/jira/browse/HBASE-25116
> Project: HBase
> Issue Type: Improvement
> Reporter: niuyulin
> Assignee: niuyulin
> Priority: Minor
>
> large count of region task from canary may affect user normal request,
> meanwhile if region task is few, the availability monitoring may shake for
> occasional exception.
> so , if the task count is large , we will randomly trim tasks for each table,
> according to the raito of the table region count in whole tasks region count.
> If the task count is few, we will repeat tasks
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
