zookeeper-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Murali Krishna. P" <muralikpb...@yahoo.com>
Subject Re: Changing hosts
Date Thu, 12 May 2011 06:07:23 GMT
Thanks for the responses,
     I have the luxury of stopping the clients during the operations. So, I would go with
the second approach of cloning.

Murali Krishna

From: Ted Dunning <ted.dunning@gmail.com>
To: Alexander Shraer <shralex@yahoo-inc.com>
Cc: "user@zookeeper.apache.org" <user@zookeeper.apache.org>; Murali Krishna. P <muralikpbhat@yahoo.com>;
"zookeeper-user@hadoop.apache.org" <zookeeper-user@hadoop.apache.org>
Sent: Thursday, 12 May 2011 5:27 AM
Subject: Re: Changing hosts


I think that this process does a slightly different thing.  Your process is good for cloning
a cluster, but it doesn't address the problem of transitioning a working cluster.  My process
never has two clusters with the same data so all transactions will always be applied to a
single notional version of the data.

The reason that this is important is a part of the process that neither of us mentioned.  That
is how to transition the clients.  My assumption is that before the transition, all clients
would re-open their ZK connection with all 6 nodes in the list of servers.  Once this is
done, my process will lead the clients through the transition in a way that all updates will
be visible to all clients.  At the end, the clients should (eventually) trim their list of
servers to the shorter list of new servers.

With a cluster clone operation, there will be moments when some clients connect to one cluster
and some connect to the other.  That makes it hard to understand how this will work well.

The OP can probably clarify which task they really wanted to accomplish.


On Wed, May 11, 2011 at 4:49 PM, Alexander Shraer <shralex@yahoo-inc.com> wrote:

Hi Ted,
>There's a simpler way that works. Suppose that the original servers are A, B and C, and
the new ones
>are D, E, F. Configure D, E and F to be in the configuration A, B, C, D, E, F and start
them. Don't do any changes to
>A, B and C. After D, E and F synch with the leader (or at least 2 of them do), turn them
off, change their config files to D, E, F and bring them up again.
>> -----Original Message-----
>> From: Ted Dunning [mailto:ted.dunning@gmail.com]
>> Sent: Tuesday, May 10, 2011 8:39 AM
>> To: user@zookeeper.apache.org; Murali Krishna. P
>> Cc: zookeeper-user@hadoop.apache.org
>> Subject: Re: Changing hosts
>> Step 1:  configure two of the new hosts to be part of a 5 node cluster
>> containing all of the
>> original nodes.
>> Step 2:  reconfigure each of the original 3 nodes to be part of the the
>> 5
>> node cluster.
>> Step 3: bounce each of the originals and start the two new servers.
>> Step 4: configure the 6th server (previously untouched) to be part of a
>> three node cluster
>> containing only the 3 new nodes.
>> Step 5: reconfigure servers 4 and 5.
>> Step 6: bounce servers 4 and 5 and start server 6.
>> Done.
>> (wait for somebody else to critique this procedure before proceeding
>> with
>> it)
>> On Tue, May 10, 2011 at 7:54 AM, Murali Krishna. P
>> <muralikpbhat@yahoo.com>wrote:
>> > Hi,
>> >      I have a zookeeper cluster (3.2.2) with 3 hosts. I need to
>> replace all
>> > the 3 hosts with different machines. What is the best way to achieve
>> this
>> > without any data loss? I can shutdown my clients during this
>> operation.
>> >
>> >
>> >
>> > Thanks,
>> > Murali Krishna
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message