Depends. If a machine fails hard and HDFS 2.x is setup correctly I'd expect 
this to be the ZK timeout (180s by default in 0.94, but can be lowered) + a few 
minutes to rejigger things.

-- Lars



________________________________
 From: Demai Ni <[email protected]>
To: "[email protected]" <[email protected]>; lars hofhansl 
<[email protected]> 
Sent: Wednesday, March 19, 2014 9:16 AM
Subject: Re: How much time HBase will take if one Region Server down.
 


Lars,

I didn't get the 15~20 min from the article. I played a bit with my clusters on 
94 early last year, and got a ball-park number. Well, my cluster was using all 
the default hbase configuration at that time, so it definitely wasn't well 
tuned.  So for HBase 94, what's a reasonable MTTR # from your experience? Thanks

Demai




On Tue, Mar 18, 2014 at 10:51 PM, lars hofhansl <[email protected]> wrote:

Many of the improvement were made in HDFS, so benefit 0.94 equally.
>Not sure where you read 15-20 minutes for 0.94 in that article.
>
>-- Lars
>
>
>
>________________________________
> From: Demai Ni <[email protected]>
>To: "[email protected]" <[email protected]>
>Sent: Tuesday, March 18, 2014 10:59 AM
>Subject: Re: How much time HBase will take if one Region Server down.
>
>
>
>a good article to start with:
>http://hortonworks.com/blog/introduction-to-hbase-mean-time-to-recover-mttr/
>
>I would say 15~20 min for 94 or earlier; it can be reduced to a couple min
>with HBase 96+ and Hadoop .
>
>Demai
>
>
>
>On Tue, Mar 18, 2014 at 10:55 AM, Upendra Yadav <[email protected]>wrote:
>
>> How much time HBase will take if one Region Server down and try to read
>> Region that belong to this down Region Server.
>>
>> Means Is there service down for those all region that belongs to dead RS?
>>
>> How much time HMaster will take to assign those region to live RS? What is
>> the dependencies that may increase region assignment in such situation.
>>
>> Thanks....
>>

Reply via email to