hbase-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Ted Yu <yuzhih...@gmail.com>
Subject Re: Load balancer - Static RegionPlans
Date Tue, 30 Aug 2011 14:29:21 GMT
>> if the new RS that came up due to some reason fails start up
The Ops should find out why the new RS doesn't come up.

Currently balancer has limited time for each balance() call. This means a
portion of 130 RegionPlans would go through the scenario below before
quantum for this round of balancing is used up.
Next round of balancing would generate new RegionPlans.

Cheers

On Tue, Aug 30, 2011 at 7:17 AM, Ramkrishna S Vasudevan <
ramakrishnas@huawei.com> wrote:

> Hi
>
>
>
> Currently whe load balancer starts balancing a cluster, the balance cluster
> algo determines the RS to which the regions has to be moved for balancing.
>
> Assuming initially there are 2 RS and 400 regions.
>
> Now when a new RS comes up a load balancing will start.
>
>
>
> Roughly 130+ regions will be shared and those many regions will be
> scheduled
> to be moved to the new RS.
>
>
>
> Now after the region plan is assigned if the new RS that came up due to
> some
> reason fails start up, the balancer tries to assign the 130 regions
>
> to this new RS but fails saying Connection Exception.
>
> Anyway the regions will be assigned to another RS once again.
>
>
>
> But the master will try 130 times for every region and then decide on
> changing the region plan.
>
> Can we optimise this?
>
>
>
> Best Regards
>
> Ram
>
>
>
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message