Mailing-List: contact user-help@zookeeper.apache.org; run by ezmlm
Precedence: bulk
Reply-To: user@zookeeper.apache.org
Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0 (Mac OS X Mail 8.2 \(2104\))
Subject: Re: rolling upgrade with wiped disk
From: Flavio Junqueira <fpj@apache.org>
In-Reply-To: <30894FD248BF284589A13D667130AD5A76F8ADD2@MX106CL01.corp.emc.com>
Date: Fri, 21 Oct 2016 14:59:38 +0100
Content-Transfer-Encoding: quoted-printable
Message-Id: <1C9568CC-491D-40B1-9A4C-5449C7231D99@apache.org>
References: <30894FD248BF284589A13D667130AD5A6E682A7C@MX106CL01.corp.emc.com> <2430F7F3-559C-4FE2-93B0-1878D4908AA5@apache.org> <30894FD248BF284589A13D667130AD5A76F8ADD2@MX106CL01.corp.emc.com>
To: user@zookeeper.apache.org
archived-at: Fri, 21 Oct 2016 13:59:48 -0000


> On 21 Oct 2016, at 14:35, Vazhenin, Maksim <Maksim.Vazhenin@dell.com> =
wrote:
>=20
> Thanks for reply, Flavio,
>=20
> "You have to be careful when you do this. You may end up losing quorum =
on txns if you wipe out the disk of a server prematurely."
> What steps could be performed to not loose quorum? If quorum is lost =
what steps are to be performed to fix it?
> Is there a chance to loose data stored in zk in this scenario?


If you wipe out the next server before a quorum is formed, then you =
could end up losing transactions. Making sure that there is a leader and =
enough followers before the next iteration is very critical.

> On node failure / node replacement it is inevitable to have disk wiped =
for one server (this should be handled by zk as this will happen =
eventually on all systems running zk).
> Does upgrading zk version on node with wiped disk adds a situation not =
supported by zk?
>=20
> "On your step 4, you can tell that a server is ready once clients are =
able to connect to it successfully."
> Do you mean client should have only this server in the connection =
string, and once this client get respond for any 'get' operation it =
means that it is safe to go ahead?

That's a way of testing, yes. A way of checking this manually is to use =
the zkCli under bin with the address of the server that you're bringing =
up. If you want to automate it, then you probably want some java code =
that waits until a session is sync connected, and the only address in =
the connect string is the one of the server you're interested in.

Thanks,
-Flavio=20

>=20
> Thanks,
> Maksim
>=20
> -----Original Message-----
> From: Flavio Junqueira [mailto:fpj@apache.org]=20
> Sent: Friday, October 21, 2016 3:59 PM
> To: user@zookeeper.apache.org
> Subject: Re: rolling upgrade with wiped disk
>=20
> Hi Maksim,
>=20
> You have to be careful when you do this. You may end up losing quorum =
on txns if you wipe out the disk of a server prematurely.
>=20
> On your step 4, you can tell that a server is ready once clients are =
able to connect to it successfully.
>=20
> Thanks,
> -Flavio
>=20
>> On 21 Oct 2016, at 09:11, Vazhenin, Maksim <Maksim.Vazhenin@dell.com> =
wrote:
>>=20
>> Hi,
>>=20
>> Is this a supported scenario for doing rolling upgrade of zookeeper =
(v3.4.5) to later version (say v3.4.9):
>>=20
>> 1. Shutdown server A (v3.4.5)
>> 2. Wipe disk with zookeeper data on server A 3. Start server A with=20=

>> new zk version (v3.4.9) 4. Wait till reconstruction complete on =
server=20
>> A (what is the indicator for completion?) 5. Go to server B and =
repeat=20
>> 1-4.
>>=20
>> Thanks,
>> Maksim
>=20