You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@cassandra.apache.org by Jason Brown <ja...@gmail.com> on 2017/08/30 17:05:50 UTC

Re: Hints replay incompatible between 2.x and 3.x

Hi Andrew,

This question is best for the user@ list, included here.

Thanks,

-Jason

On Wed, Aug 30, 2017 at 10:00 AM, Andrew Whang <an...@gmail.com>
wrote:

> In evaluating 3.x, we found that hints are unable to be replayed between
> 2.x and 3.x nodes. This introduces a risk during the upgrade path for some
> of our write-heavy clusters - nodes will accumulate upwards of 1TB of hints
> if a node goes/remains down for <1hr.
>
> Any suggestions to mitigate this issue?
>

Re: Hints replay incompatible between 2.x and 3.x

Posted by Aleksey Yeshchenko <al...@apple.com>.
It doesn’t work between any two majors (1.2 and 2.0, 2.0 and 2.1, 2.1 and 3.0).

The reason is that hint delivery doesn’t proceed between two nodes unless they have the same schema version,
and in every recent major release we made changes that made schema calculation differ even without any DDL
statements going on.

There might be an argument for relaxing that restriction, and/or for making schema versioning more stable, but we are where we are right now.

—
AY

On 30 August 2017 at 18:06:00, Jason Brown (jasedbrown@gmail.com) wrote:

Hi Andrew,  

This question is best for the user@ list, included here.  

Thanks,  

-Jason  

On Wed, Aug 30, 2017 at 10:00 AM, Andrew Whang <an...@gmail.com>  
wrote:  

> In evaluating 3.x, we found that hints are unable to be replayed between  
> 2.x and 3.x nodes. This introduces a risk during the upgrade path for some  
> of our write-heavy clusters - nodes will accumulate upwards of 1TB of hints  
> if a node goes/remains down for <1hr.  
>  
> Any suggestions to mitigate this issue?  
>  

Re: Hints replay incompatible between 2.x and 3.x

Posted by Erick Ramirez <fl...@gmail.com>.
1TB of hints suggests you don't have enough capacity in your cluster. The
only way around that is to add more nodes.

On Thu, Aug 31, 2017 at 3:05 AM, Jason Brown <ja...@gmail.com> wrote:

> Hi Andrew,
>
> This question is best for the user@ list, included here.
>
> Thanks,
>
> -Jason
>
> On Wed, Aug 30, 2017 at 10:00 AM, Andrew Whang <an...@gmail.com>
> wrote:
>
>> In evaluating 3.x, we found that hints are unable to be replayed between
>> 2.x and 3.x nodes. This introduces a risk during the upgrade path for some
>> of our write-heavy clusters - nodes will accumulate upwards of 1TB of
>> hints
>> if a node goes/remains down for <1hr.
>>
>> Any suggestions to mitigate this issue?
>>
>
>