cassandra-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From aaron morton <aa...@thelastpickle.com>
Subject Re: rolling restart after gc_grace change
Date Mon, 08 Oct 2012 20:24:35 GMT
> Is it still an issue if you don't run a repair within gc_grace_seconds ?
There is a potential issue.

You want to make sure the tombstones are distributed to all replicas *before* gc_grace_seconds
has expired. If they are not you can have a case where some replicas compact and purge their
tombstone (essentially a hard delete), while one replica keeps the original value. The result
is data returning from the dead. 

Cheers

-----------------
Aaron Morton
Freelance Developer
@aaronmorton
http://www.thelastpickle.com

On 6/10/2012, at 2:54 AM, Oleg Dulin <oleg.dulin@gmail.com> wrote:

> What if gc_grace_seconds is pretty low, say 2 mins, what happens with nodetool repair
?
> 
> That wiki page below points at a bug that has been fixed long ago. Is it still an issue
if you don't run a repair within gc_grace_seconds ?
> 
> On 2012-01-09 10:02:49 +0000, aaron morton said:
> 
> Nah, thats old style. 
> 
> gc_grace_seconds is a CF level setting now. Make the change with update column family
in the CLI or your favorite client. 
> 
> Cheers
>  
> -----------------
> Aaron Morton
> Freelance Developer
> @aaronmorton
> http://www.thelastpickle.com
> 
> On 9/01/2012, at 9:33 PM, Igor wrote:
> Hi!
> 
> On thehttp://wiki.apache.org/cassandra/Operations#Dealing_with_the_consequences_of_nodetool_repair_not_running_within_GCGraceSeconds
you can read:
> 
> "To minimize the amount of forgotten deletes, first increase GCGraceSeconds across the
cluster (rolling restart required)"
> 
> Rolling restart still required for 1.0.6?
> 
> 
> -- 
> Regards,
> Oleg Dulin
> NYC Java Big Data Engineer
> http://www.olegdulin.com/


Mime
View raw message