Mailing-List: contact user-help@cassandra.apache.org; run by ezmlm
Precedence: bulk
Reply-To: user@cassandra.apache.org
Content-Type: text/plain; charset=utf-8
Mime-Version: 1.0 (Mac OS X Mail 8.2 \(2104\))
Subject: Re: quick questions
From: Max C <mc_cassandra@core43.com>
In-Reply-To: <CADOsr-E6U7vUOP1Z8GsO=u7s1_YMmSyxTT_ozNntRmraGzj2Zw@mail.gmail.com>
Date: Sat, 17 Dec 2016 17:02:52 -0800
Content-Transfer-Encoding: quoted-printable
Message-Id: <464D32DC-414C-4B8A-9D2C-D2D71C25E662@core43.com>
References: <CADOsr-E6U7vUOP1Z8GsO=u7s1_YMmSyxTT_ozNntRmraGzj2Zw@mail.gmail.com>
To: user@cassandra.apache.org
archived-at: Sun, 18 Dec 2016 01:03:13 -0000

As Matija mentioned, quorum is RF / 2 + 1:

RF=3D1, Quorum =3D 1
RF=3D2, Quorum =3D 2
RF=3D3, Quorum =3D 2
RF=3D4, Quorum =3D 3
RF=3D5, Quorum =3D 3
RF=3D6, Quorum =3D 4
RF=3D7, Quorum =3D 4

So no, you don=E2=80=99t have to have an odd RF to achieve a quorum, as =
you see above.  Most people use RF=3D3 with a minimum of 3 nodes, =
though.  For RF=3D3, 2 of the 3 nodes need to be up in order to satisfy =
a quorum read/write.

If you can=E2=80=99t achieve a quorum and you=E2=80=99re trying to =
read/write with quorum consistency then the read/write operation will =
fail.  You could still do reads/writes with CL=3DONE, though (provided =
that at least 1 of the replicas was up).

- Max

> On Dec 17, 2016, at 1:21 pm, Kant Kodali <kant@peernova.com> wrote:
>=20
> I keep hearing that the minimum number of Cassandra nodes required to =
achieve Quorum consensus is 4 I wonder why not 3? In fact, many =
container deployments by default seem to deploy 4 nodes. Can anyone =
shine some light on this?
>=20
> What happens if I have 3 nodes and replication factor of 3 and =
consistency level: quorum? I should be able to achieve quorum level =
consensus right.
>=20
> If Total node =3D 3, RF=3D2 and consistency level =3D Quorum. Then I =
understand the quorum level consensus is not possible because the number =
of replica nodes here are 2.
> This also brings up another question does number of replica nodes =
always have to be an odd number to achieve quorum level consensus? If =
so, what happens when a replica node goes down ? it would still serve =
the requests but the quorum level consensus is not possible?
>=20
> Thanks
> kant
>=20
>=20
>=20
>=20
>=20