Mailing-List: contact issues-help@hbase.apache.org; run by ezmlm
Precedence: bulk
Date: Fri, 6 Jun 2014 17:14:02 +0000 (UTC)
From: "Jimmy Xiang (JIRA)" <jira@apache.org>
To: issues@hbase.apache.org
Message-ID: <JIRA.12710152.1398287682303.82689.1402074842992@arcas>
In-Reply-To: <JIRA.12710152.1398287682303@arcas>
References: <JIRA.12710152.1398287682303@arcas>
Subject: [jira] [Commented] (HBASE-11059) ZK-less region assignment
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit


    [ https://issues.apache.org/jira/browse/HBASE-11059?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14020063#comment-14020063 ] 

Jimmy Xiang commented on HBASE-11059:
-------------------------------------

Thanks a lot for the review, Stack!

This one is fixed:
{quote}
Leave of this sentence I'd say: " The other
+ * way around is not recommended also sophisticated users could do it
+ * somehow."
{quote}

For migration instruction, I will copy it to the release note.

This one is fixed:
{quote}
This message will come out wrong?
+ LOG.info("Finished region assignment in (failover=" + failover
+ + ")" + (System.currentTimeMillis() - startTime) + "ms");
{quote}

That's right. It's better to be done in one single transaction explicitly. Currently, I think we have transactions as long as only one RS is doing the merge/split for the same set of regions.
{quote}
We have transactions right as long as all in the same region? Which is the case when only single meta.
{quote}


By the way, I did some test. With a patch v3 (to-be-uploaded),  it took about 316 seconds (~5 minutes) to assign 1 million regions. All these regions are empty so take no time to open. Including loading the regions from meta (twice, one to restore the state, the other one to init the region assignment snapshot for favored nodes plan. I think the second one can be removed. I can fix it in a separate issue), it took about 395 seconds (~6.5 minutes).

Without the patch, I don't have a number yet. It's still running. I have increased the ZK jute buffer several times.

> ZK-less region assignment
> -------------------------
>
>                 Key: HBASE-11059
>                 URL: https://issues.apache.org/jira/browse/HBASE-11059
>             Project: HBase
>          Issue Type: Improvement
>          Components: master, Region Assignment
>            Reporter: Jimmy Xiang
>            Assignee: Jimmy Xiang
>         Attachments: hbase-11059.patch, hbase-11059_v2.patch, zk-less_am.pdf
>
>
> It seems that most people don't like region assignment with ZK (HBASE-5487), which causes many uncertainties. This jira is to support ZK-less region assignment. We need to make sure this patch doesn't break backward compatibility/rolling upgrade.


--
This message was sent by Atlassian JIRA
(v6.2#6252)