[
https://issues.apache.org/jira/browse/HBASE-15027?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
ramkrishna.s.vasudevan updated HBASE-15027:
-------------------------------------------
Attachment: HBASE-15027_6.patch
Other than the fixes on the comments,
I moved the CompactedHFileDischarger and CompactedHFilesDischargerHandler to
the regionserver package. Earlier it was in regionserver.compactions package.
This was done so that the getRegionserverServices() need not be exposed
outside. (currently it is default). Also found that the tests are some times
flaky hence now the CompactedHFileDischarger will also allow you to execute the
close and archival in sync way rather than the executor way. How ever that is
only for tests.
> Refactor the way the CompactedHFileDischarger threads are created
> -----------------------------------------------------------------
>
> Key: HBASE-15027
> URL: https://issues.apache.org/jira/browse/HBASE-15027
> Project: HBase
> Issue Type: Bug
> Reporter: ramkrishna.s.vasudevan
> Assignee: ramkrishna.s.vasudevan
> Fix For: 2.0.0
>
> Attachments: HBASE-15027.patch, HBASE-15027_1.patch,
> HBASE-15027_2.patch, HBASE-15027_3.patch, HBASE-15027_3.patch,
> HBASE-15027_4.patch, HBASE-15027_5.patch, HBASE-15027_6.patch
>
>
> As per suggestion given over in HBASE-14970, if we need to create a single
> thread pool service for the CompactionHFileDischarger we need to create an
> exectuor service in the RegionServer level and create discharger handler
> threads (Event handlers) and pass the Event to the new Exectuor service that
> we create for the compaction hfiles discharger. What should be the default
> number of threads here? If a HRS holds 100 of regions - will 10 threads be
> enough? This issue will try to resolve this with tests and discussions and
> suitable patch will be updated in HBASE-14970 for branch-1 once this is
> committed.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
