Mailing-List: contact user-help@cassandra.apache.org; run by ezmlm
Precedence: bulk
Reply-To: user@cassandra.apache.org
Received-SPF: pass (nike.apache.org: local policy)
Content-Type: text/plain; charset=windows-1252
Mime-Version: 1.0 (Mac OS X Mail 6.2 \(1499\))
Subject: Re: Best practices for nodes to come back
From: aaron morton <aaron@thelastpickle.com>
In-Reply-To: 
 <CABNBnCiCvc75Pv6d4PvN1DJh9s3a6c1tCmWe4aYYzrVmMiFiaQ@mail.gmail.com>
Date: Wed, 13 Mar 2013 22:22:17 -0700
Content-Transfer-Encoding: quoted-printable
Message-Id: <5568BDB7-D912-440A-8363-484511B61764@thelastpickle.com>
References: 
 <CABNBnCiCvc75Pv6d4PvN1DJh9s3a6c1tCmWe4aYYzrVmMiFiaQ@mail.gmail.com>
To: user@cassandra.apache.org

If you node has been dead for less then gc_grace you can return it to =
the cluster and run nodetool repair (without the -pr). Until repair has =
completed will be getting inconsistent results, but if you have been =
using ONE / ONE for all ops that is a possibility for everything.=20

If the node has been down for longer than gc_grace you cannot bring it =
back. Remove it from another node using nodetool and bootstrap a new =
node with the same token.=20

Cheers

-----------------
Aaron Morton
Freelance Cassandra Consultant
New Zealand

@aaronmorton
http://www.thelastpickle.com

On 12/03/2013, at 3:02 AM, julien Campan <julien.campan@gmail.com> =
wrote:

> Hi,
>=20
> I have a question about the nodes coming back to the cluster after a =
long interruption.
>=20
> First of all, I explain my use case. I have a cluster that processes =
many requests with consistency ONE and I wouldn=92t change it because of =
an incident.  Secondly, I don't want to return the old information from =
this node. So, I want to avoid reading from this node until it gets =
repaired.
>=20
> After looking at documentation I see two possible solutions:
>=20
> 1) Start the incoming node in =93survey mode=94, wait until the end of =
the hints (or end of repair) and restart the node in =93normal mode=94.
>=20
> 2) Use the =93nodetool remove node=94, remove data repository and =
change the IP address of the removed node (I=92m working in virtual =
environment), and add the node again to the cluster. In this case I will =
be in a =93autobootstrap mode".
>=20
> I wonder which solution is the best, and if there is a more simple and =
better solution to for my use case.
>=20
> Thanks.
>=20