You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@cassandra.apache.org by Aiman Parvaiz <ai...@flipagram.com> on 2015/08/15 02:10:14 UTC

Need advice for multi DC C* setup

Hi all
We are planning to move C* from EC2 (region A) to VPC in region B. I will
enumerate our goals so that you guys can advice me keeping in mind the
bigger picture.

Goals:
- Move to VPC is another region.
- Enable Vnodes.
- Bump up RF to 3.
- Ability to have a spark cluster.

I know this is a LOT of work and I know this all might not be possible in
one go.

Existing cluster in EC2 is using RF=2, simple snitch and simple replication.

I am not sure what would be the best way to approach this task. So please
anyone who has done this and would like to share anything I would really
appreciate the effort.

Thanks

Re: Need advice for multi DC C* setup

Posted by Aiman Parvaiz <ai...@flipagram.com>.
Over the weekend and after some more looking around and following this old
mailing list post

https://mail-archives.apache.org/mod_mbox/incubator-cassandra-user/201406.mbox/%3CCA+VSrLopop7Th8nX20aOZ3As75g2jrJm3ryX119dekLYNHqFwA@mail.gmail.com%3E

I was able to get my 2 node test env to move to over 3 node RF 3 cluster in
private subnet VPC in region B. I updated the RF when I started replication
between C* in public subnet to private subnet in region B.

The catch here is that since the test env was 2 nodes and both nodes were
in one AZ changing snitch to EC2Multiregion didnt afftect the replica
placement and hence I was able to get away with a rolling restart  but in
production I have 10 nodes spread over 2 AZ running simple snitch. I wonder
what would be the best way to change snitch live in this scenario.

One way I think would be to get all nodes in one AZ and then switch the
snitch that way ec2multiregion would report all nodes in 1 Rack but I am
open to suggestions also if this is a valid concern.

Thanks

On Sun, Aug 16, 2015 at 1:46 AM, Prem Yadav <ip...@gmail.com> wrote:

> I meant the existing nodes must be in the default VPC if you did not
> create one,
> In any case, you can use the VPC peering.
>
> On Sun, Aug 16, 2015 at 5:34 AM, John Wong <go...@gmail.com> wrote:
>
>> > The EC2 nodes must be in the default VPC.
>> Did you really mean the default VPC created by AWS or just a VPC? Because
>> I would be very surprise default VPC must be used.
>>
>> On Sat, Aug 15, 2015 at 2:50 AM, Prem Yadav <ip...@gmail.com> wrote:
>>
>>>
>>> The EC2 nodes must be in the default VPC.
>>>
>>> create a ring in the VPC in region B. Use VPC peering to connect the
>>> default and the region B VPC.
>>> The two rings should join the existing one. Alter the replication
>>> strategy to network replication so that the data is replicated to the new
>>> rings. Repair the keyspaces.
>>> Once it is done, you can decommission the existing ring.
>>>
>>> For spark,if you are using datastax version, it comes with spark. You
>>> just need to change a config and spark starts along with cassandra. A
>>> separate ring is advised for analytics stuff.
>>>
>>>
>>> On Sat, Aug 15, 2015 at 1:10 AM, Aiman Parvaiz <ai...@flipagram.com>
>>> wrote:
>>>
>>>> Hi all
>>>> We are planning to move C* from EC2 (region A) to VPC in region B. I
>>>> will enumerate our goals so that you guys can advice me keeping in mind the
>>>> bigger picture.
>>>>
>>>> Goals:
>>>> - Move to VPC is another region.
>>>> - Enable Vnodes.
>>>> - Bump up RF to 3.
>>>> - Ability to have a spark cluster.
>>>>
>>>> I know this is a LOT of work and I know this all might not be possible
>>>> in one go.
>>>>
>>>> Existing cluster in EC2 is using RF=2, simple snitch and simple
>>>> replication.
>>>>
>>>> I am not sure what would be the best way to approach this task. So
>>>> please anyone who has done this and would like to share anything I would
>>>> really appreciate the effort.
>>>>
>>>> Thanks
>>>>
>>>
>>>
>>
>


-- 
*Aiman Parvaiz*
Lead Systems Architect
aiman@flipagram.com
cell: 213-300-6377
http://flipagram.com/apz

Re: Need advice for multi DC C* setup

Posted by Prem Yadav <ip...@gmail.com>.
I meant the existing nodes must be in the default VPC if you did not create
one,
In any case, you can use the VPC peering.

On Sun, Aug 16, 2015 at 5:34 AM, John Wong <go...@gmail.com> wrote:

> > The EC2 nodes must be in the default VPC.
> Did you really mean the default VPC created by AWS or just a VPC? Because
> I would be very surprise default VPC must be used.
>
> On Sat, Aug 15, 2015 at 2:50 AM, Prem Yadav <ip...@gmail.com> wrote:
>
>>
>> The EC2 nodes must be in the default VPC.
>>
>> create a ring in the VPC in region B. Use VPC peering to connect the
>> default and the region B VPC.
>> The two rings should join the existing one. Alter the replication
>> strategy to network replication so that the data is replicated to the new
>> rings. Repair the keyspaces.
>> Once it is done, you can decommission the existing ring.
>>
>> For spark,if you are using datastax version, it comes with spark. You
>> just need to change a config and spark starts along with cassandra. A
>> separate ring is advised for analytics stuff.
>>
>>
>> On Sat, Aug 15, 2015 at 1:10 AM, Aiman Parvaiz <ai...@flipagram.com>
>> wrote:
>>
>>> Hi all
>>> We are planning to move C* from EC2 (region A) to VPC in region B. I
>>> will enumerate our goals so that you guys can advice me keeping in mind the
>>> bigger picture.
>>>
>>> Goals:
>>> - Move to VPC is another region.
>>> - Enable Vnodes.
>>> - Bump up RF to 3.
>>> - Ability to have a spark cluster.
>>>
>>> I know this is a LOT of work and I know this all might not be possible
>>> in one go.
>>>
>>> Existing cluster in EC2 is using RF=2, simple snitch and simple
>>> replication.
>>>
>>> I am not sure what would be the best way to approach this task. So
>>> please anyone who has done this and would like to share anything I would
>>> really appreciate the effort.
>>>
>>> Thanks
>>>
>>
>>
>

Re: Need advice for multi DC C* setup

Posted by John Wong <go...@gmail.com>.
> The EC2 nodes must be in the default VPC.
Did you really mean the default VPC created by AWS or just a VPC? Because I
would be very surprise default VPC must be used.

On Sat, Aug 15, 2015 at 2:50 AM, Prem Yadav <ip...@gmail.com> wrote:

>
> The EC2 nodes must be in the default VPC.
>
> create a ring in the VPC in region B. Use VPC peering to connect the
> default and the region B VPC.
> The two rings should join the existing one. Alter the replication strategy
> to network replication so that the data is replicated to the new rings.
> Repair the keyspaces.
> Once it is done, you can decommission the existing ring.
>
> For spark,if you are using datastax version, it comes with spark. You just
> need to change a config and spark starts along with cassandra. A separate
> ring is advised for analytics stuff.
>
>
> On Sat, Aug 15, 2015 at 1:10 AM, Aiman Parvaiz <ai...@flipagram.com>
> wrote:
>
>> Hi all
>> We are planning to move C* from EC2 (region A) to VPC in region B. I will
>> enumerate our goals so that you guys can advice me keeping in mind the
>> bigger picture.
>>
>> Goals:
>> - Move to VPC is another region.
>> - Enable Vnodes.
>> - Bump up RF to 3.
>> - Ability to have a spark cluster.
>>
>> I know this is a LOT of work and I know this all might not be possible in
>> one go.
>>
>> Existing cluster in EC2 is using RF=2, simple snitch and simple
>> replication.
>>
>> I am not sure what would be the best way to approach this task. So please
>> anyone who has done this and would like to share anything I would really
>> appreciate the effort.
>>
>> Thanks
>>
>
>

Re: Need advice for multi DC C* setup

Posted by Prem Yadav <ip...@gmail.com>.
The EC2 nodes must be in the default VPC.

create a ring in the VPC in region B. Use VPC peering to connect the
default and the region B VPC.
The two rings should join the existing one. Alter the replication strategy
to network replication so that the data is replicated to the new rings.
Repair the keyspaces.
Once it is done, you can decommission the existing ring.

For spark,if you are using datastax version, it comes with spark. You just
need to change a config and spark starts along with cassandra. A separate
ring is advised for analytics stuff.


On Sat, Aug 15, 2015 at 1:10 AM, Aiman Parvaiz <ai...@flipagram.com> wrote:

> Hi all
> We are planning to move C* from EC2 (region A) to VPC in region B. I will
> enumerate our goals so that you guys can advice me keeping in mind the
> bigger picture.
>
> Goals:
> - Move to VPC is another region.
> - Enable Vnodes.
> - Bump up RF to 3.
> - Ability to have a spark cluster.
>
> I know this is a LOT of work and I know this all might not be possible in
> one go.
>
> Existing cluster in EC2 is using RF=2, simple snitch and simple
> replication.
>
> I am not sure what would be the best way to approach this task. So please
> anyone who has done this and would like to share anything I would really
> appreciate the effort.
>
> Thanks
>