cassandra-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Kant Kodali <k...@peernova.com>
Subject Re: non incremental repairs with cassandra 2.2+
Date Wed, 19 Oct 2016 16:57:15 GMT
Thanks! How do I do an incremental repair when I add a new node?

Sent from my iPhone

> On Oct 19, 2016, at 9:54 AM, Alexander Dejanovski <alex@thelastpickle.com> wrote:
> 
> Hi Kant,
> 
> subrange is a form of full repair, so it will just split the repair process in smaller
yet sequential pieces of work (repair is started giving a start and end token). Overall, you
should not expect improvements other than having less overstreaming and better chances of
success if your cluster is dense.
> 
> You can try to use incremental repair if you know what the caveats are and use a proper
tool to orchestrate it, that would save you from repairing all 10TB each time.
> CASSANDRA-12580 might help too as Romain showed us : https://www.mail-archive.com/user@cassandra.apache.org/msg49344.html
> 
> Cheers,
> 
> 
> 
> On Wed, Oct 19, 2016 at 6:42 PM Kant Kodali <kant@peernova.com> wrote:
> Another question on a same note would be what would be the fastest way to do repairs
of size 10TB cluster ? Full repairs are taking days. So among repair parallel or repair sub
range which is faster in the case of say adding a new node to the cluster?
> 
> Sent from my iPhone
> 
>> On Oct 19, 2016, at 9:30 AM, Sean Bridges <sean.bridges@globalrelay.net> wrote:
>> 
>> Hey,
>> 
>> We are upgrading from cassandra 2.1 to cassandra 2.2.  
>> 
>> With cassandra 2.1 we would periodically repair all nodes, using the -pr flag.  
>> 
>> With cassandra 2.2, the same repair takes a very long time, as cassandra does an
anti compaction after the repair.  This anti compaction causes most (all?) the sstables to
be rewritten.  Is there a way to do full repairs without continually anti compacting?  If
we do a full repair on each node with the -pr flag, will subsequent full repairs also force
anti         compacting most (all?) sstables?  
>> 
>> Thanks,
>> 
>> Sean
> 
> -- 
> -----------------
> Alexander Dejanovski
> France
> @alexanderdeja
> 
> Consultant
> Apache Cassandra Consulting
> http://www.thelastpickle.com

Mime
View raw message