You are viewing a plain text version of this content. The canonical link for it is here.

Posted to mapreduce-user@hadoop.apache.org by Bing Jiang <ji...@gmail.com> on 2015/07/01 11:30:35 UTC

Move blocks between Nodes

hi, guys.

I want to move all the blocks from 50 datanodes to another 50 new
datanodes. There is a very easy idea that we can add the new 50 nodes to
hadoop cluster firstly, then decommission the other 50 nodes one by one.
But, I believe it is not an efficient way to reach the goal.
So I plan to get the help of  the idea of Hdfs balancer, which limits the
movements  happen to the 100 nodes. But I need to disable write_op for
those nodes to be decommissioned.

Is there a way to make DataNode in safe mode (read only) mode?

Of course, regarding to blocks movement between nodes,  Any thoughts will
be appreciated.



-- 
Bing Jiang

Re: Move blocks between Nodes

Posted by Gurmukh Singh <gu...@yahoo.com>.

The best way to do is by decommissioning nodes as said by Rich.

Another way could be to bump up the replication factor by setrep command 
(without the -w option) and wait for a few hours and then reset the 
replication factor back to original.

Then start the decommissioning process, it will be much quicker.

On 01/07/15 8:39 pm, Rich Haase wrote:
> Hi Bing,
>
> I would recommend that you add your 50 new nodes to the cluster and 
> the decommission the 50 nodes you want to get rid of.  You can do the 
> decommission in one operation (albeit a lengthy operation) by adding 
> the nodes you want to decommission to your HDFS exclude file and 
> running `hdfs dfsadmin -refreshNodes`.  The decommission process will 
> ensure that the data from your old nodes are distributed across your 
> cluster.
>
> Cheers,
>
> Rich
>
>> On Jul 1, 2015, at 3:30 AM, Bing Jiang <jiangbinglover@gmail.com 
>> <ma...@gmail.com>> wrote:
>>
>> hi, guys.
>>
>> I want to move all the blocks from 50 datanodes to another 50 new 
>> datanodes. There is a very easy idea that we can add the new 50 nodes 
>> to hadoop cluster firstly, then decommission the other 50 nodes one 
>> by one. But, I believe it is not an efficient way to reach the goal.
>> So I plan to get the help of  the idea of Hdfs balancer, which limits 
>> the movements  happen to the 100 nodes. But I need to disable 
>> write_op for those nodes to be decommissioned.
>>
>> Is there a way to make DataNode in safe mode (read only) mode?
>>
>> Of course, regarding to blocks movement between nodes,  Any thoughts 
>> will be appreciated.
>>
>>
>>
>> -- 
>> Bing Jiang
>>
>
> *Rich Haase*| Sr. Software Engineer | Pandora
> m (303) 887-1146 | rhaase@pandora.com <ma...@pandora.com>
>
>
>
>

-- 
--
Thanks and Regards

Gurmukh Singh

Re: Move blocks between Nodes

Posted by Gurmukh Singh <gu...@yahoo.com>.

The best way to do is by decommissioning nodes as said by Rich.

Another way could be to bump up the replication factor by setrep command 
(without the -w option) and wait for a few hours and then reset the 
replication factor back to original.

Then start the decommissioning process, it will be much quicker.

On 01/07/15 8:39 pm, Rich Haase wrote:
> Hi Bing,
>
> I would recommend that you add your 50 new nodes to the cluster and 
> the decommission the 50 nodes you want to get rid of.  You can do the 
> decommission in one operation (albeit a lengthy operation) by adding 
> the nodes you want to decommission to your HDFS exclude file and 
> running `hdfs dfsadmin -refreshNodes`.  The decommission process will 
> ensure that the data from your old nodes are distributed across your 
> cluster.
>
> Cheers,
>
> Rich
>
>> On Jul 1, 2015, at 3:30 AM, Bing Jiang <jiangbinglover@gmail.com 
>> <ma...@gmail.com>> wrote:
>>
>> hi, guys.
>>
>> I want to move all the blocks from 50 datanodes to another 50 new 
>> datanodes. There is a very easy idea that we can add the new 50 nodes 
>> to hadoop cluster firstly, then decommission the other 50 nodes one 
>> by one. But, I believe it is not an efficient way to reach the goal.
>> So I plan to get the help of  the idea of Hdfs balancer, which limits 
>> the movements  happen to the 100 nodes. But I need to disable 
>> write_op for those nodes to be decommissioned.
>>
>> Is there a way to make DataNode in safe mode (read only) mode?
>>
>> Of course, regarding to blocks movement between nodes,  Any thoughts 
>> will be appreciated.
>>
>>
>>
>> -- 
>> Bing Jiang
>>
>
> *Rich Haase*| Sr. Software Engineer | Pandora
> m (303) 887-1146 | rhaase@pandora.com <ma...@pandora.com>
>
>
>
>

-- 
--
Thanks and Regards

Gurmukh Singh

Re: Move blocks between Nodes

Posted by Gurmukh Singh <gu...@yahoo.com>.

The best way to do is by decommissioning nodes as said by Rich.

Another way could be to bump up the replication factor by setrep command 
(without the -w option) and wait for a few hours and then reset the 
replication factor back to original.

Then start the decommissioning process, it will be much quicker.

On 01/07/15 8:39 pm, Rich Haase wrote:
> Hi Bing,
>
> I would recommend that you add your 50 new nodes to the cluster and 
> the decommission the 50 nodes you want to get rid of.  You can do the 
> decommission in one operation (albeit a lengthy operation) by adding 
> the nodes you want to decommission to your HDFS exclude file and 
> running `hdfs dfsadmin -refreshNodes`.  The decommission process will 
> ensure that the data from your old nodes are distributed across your 
> cluster.
>
> Cheers,
>
> Rich
>
>> On Jul 1, 2015, at 3:30 AM, Bing Jiang <jiangbinglover@gmail.com 
>> <ma...@gmail.com>> wrote:
>>
>> hi, guys.
>>
>> I want to move all the blocks from 50 datanodes to another 50 new 
>> datanodes. There is a very easy idea that we can add the new 50 nodes 
>> to hadoop cluster firstly, then decommission the other 50 nodes one 
>> by one. But, I believe it is not an efficient way to reach the goal.
>> So I plan to get the help of  the idea of Hdfs balancer, which limits 
>> the movements  happen to the 100 nodes. But I need to disable 
>> write_op for those nodes to be decommissioned.
>>
>> Is there a way to make DataNode in safe mode (read only) mode?
>>
>> Of course, regarding to blocks movement between nodes,  Any thoughts 
>> will be appreciated.
>>
>>
>>
>> -- 
>> Bing Jiang
>>
>
> *Rich Haase*| Sr. Software Engineer | Pandora
> m (303) 887-1146 | rhaase@pandora.com <ma...@pandora.com>
>
>
>
>

-- 
--
Thanks and Regards

Gurmukh Singh