hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Matteo Bertozzi (JIRA)" <j...@apache.org>
Subject [jira] [Comment Edited] (HBASE-14837) Procedure Queue Improvement
Date Tue, 05 Jan 2016 21:11:40 GMT

    [ https://issues.apache.org/jira/browse/HBASE-14837?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15083747#comment-15083747
] 

Matteo Bertozzi edited comment on HBASE-14837 at 1/5/16 9:11 PM:
-----------------------------------------------------------------

{quote}
// TODO: should this be part of the HTD?
Not yet I'd say. Will only confuse till better baked.
{quote}
This was part of the MasterProcedureQueue file, I just renamed it in MasterProcedureScheduler
to reflect what is really doing. but most of the code is the old one

{quote}A ProcedureEvent could get in the way of both table and server events? I suppose that
makes sense. A ProcedureEvent is a major transition?{quote}
A ProcedureEvent is just a wake up for now, prioritization logic will get in later when we
will have events from the Assignment/Server Manager that must trigger something big, like
a server dying should "reschedule/retrigger" operations on another server and similar.

{quote}Whats this? AvlTree?{quote}
Andrew pointed out that there was too much allocation/deallocation going on with the queue
stuff. 
all the java data structures for put/get create extra stuff that we don't need. it's just
easier to have pointers. and avoid allocation at all, as we do now.


was (Author: mbertozzi):
{quote}A ProcedureEvent could get in the way of both table and server events? I suppose that
makes sense. A ProcedureEvent is a major transition?{quote}
A ProcedureEvent is just a wake up for now, prioritization logic will get in later when we
will have events from the Assignment/Server Manager that must trigger something big, like
a server dying should "reschedule/retrigger" operations on another server and similar.

{quote}Whats this? AvlTree?{quote}
Andrew pointed out that there was too much allocation/deallocation going on with the queue
stuff. 
all the java data structures for put/get create extra stuff that we don't need. it's just
easier to have pointers. and avoid allocation at all, as we do now.

> Procedure Queue Improvement
> ---------------------------
>
>                 Key: HBASE-14837
>                 URL: https://issues.apache.org/jira/browse/HBASE-14837
>             Project: HBase
>          Issue Type: Sub-task
>          Components: proc-v2
>    Affects Versions: 1.2.0, 1.1.2
>            Reporter: Matteo Bertozzi
>            Assignee: Matteo Bertozzi
>            Priority: Minor
>             Fix For: 2.0.0, 1.3.0, 1.2.1
>
>         Attachments: HBASE-14837-v0.patch, HBASE-14837-v1.patch
>
>
> Add event support to the procedure queue to avoid spinning and remove all the immutable
object creations from the java classes Map, Tree, ... 
> the queues that can't be executed because are waiting for an event (e.g. master initialized)
or someone else have an exclusive lock are pulled out the run queue
> https://reviews.apache.org/r/40460/



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message