cassandra-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Jean Carlo <jean.jeancar...@gmail.com>
Subject Re: Restore cassandra snapshots
Date Wed, 18 Oct 2017 06:22:36 GMT
Hi Pradeep,

Because you use sstableloader, you don't need to restore de system keyspace.

Your procedure is correct to me.

Best regards

On Oct 18, 2017 4:22 AM, "Pradeep Chhetri" <pradeep@stashaway.com> wrote:

Hi Anthony

I did the following steps to restore. Please let me know if I missed
something.

- Took snapshots on the 3 nodes of the existing cluster simultaneously
- copied that snapshots respectively on the 3 nodes of the freshly created
cluster
- ran sstableloader on each of the application table. ( I didn't restore
the system related tables ) on all of three node.

I was assuming that since I ran from all the three snapshots, all the
tokens should be there so thought that this will not cause any data loss.

Do you see that I might have data loss. I     am not sure how to verify the
data loss although I did ran count of few table to verify the row count.

Thank you the help.


On Wed, 18 Oct 2017 at 5:39 AM, Anthony Grasso <anthony.grasso@gmail.com>
wrote:

> Hi Pradeep,
>
> If you are going to copy N snapshots to N nodes you will need to make sure
> you have the System keyspace as part of that snapshot. The System keyspace
> that is local to each node, contains the token allocations for that
> particular node. This allows the node to work out what data it is
> responsible for. Further to that, if you are restoring the System keyspace
> from snapshots, make sure that the cluster name of the new cluster is
> exactly the same as the cluster which generated the System keyspace
> snapshots.
>
> Regards,
> Anthony
>
> On 16 October 2017 at 23:28, Jean Carlo <jean.jeancarl48@gmail.com> wrote:
>
>> HI,
>>
>> Yes of course, you can use sstableloader from every sstable to your new
>> cluster. Actually this is the common procedure. Just check the log of
>> cassandra, you shouldn't see any errors of streaming.
>>
>>
>> However, because the fact you are migrating from on cluster of N nodes to
>> another of N nodes, I believe you can just copy and paste your data node
>> per node and make a nodetool refresh. Checking obviously the correct names
>> of your sstables.
>> You can check the tokens of your node using nodetool info -T
>>
>> But I think sstableloader is the easy way :)
>>
>>
>>
>>
>> Saludos
>>
>> Jean Carlo
>>
>> "The best way to predict the future is to invent it" Alan Kay
>>
>> On Mon, Oct 16, 2017 at 1:55 PM, Pradeep Chhetri <pradeep@stashaway.com>
>> wrote:
>>
>>> Hi Jean,
>>>
>>> Thank you for the quick response. I am not sure how to achieve that. Can
>>> i set the tokens for a node via cqlsh ?
>>>
>>> I know that i can check the nodetool rings to get the tokens allocated
>>> to a node.
>>>
>>> I was thinking to basically run sstableloader for each of the snapshots
>>> and was assuming it will load the complete data properly. Isn't that the
>>> case.
>>>
>>> Thank you.
>>>
>>> On Mon, Oct 16, 2017 at 5:21 PM, Jean Carlo <jean.jeancarl48@gmail.com>
>>> wrote:
>>>
>>>> Hi,
>>>>
>>>> Be sure that you have the same tokens distribution than your original
>>>> cluster. So if you are going to restore from old node 1 to new node 1, make
>>>> sure that the new node and the old node have the same tokens.
>>>>
>>>>
>>>> Saludos
>>>>
>>>> Jean Carlo
>>>>
>>>> "The best way to predict the future is to invent it" Alan Kay
>>>>
>>>> On Mon, Oct 16, 2017 at 1:40 PM, Pradeep Chhetri <pradeep@stashaway.com
>>>> > wrote:
>>>>
>>>>> Hi,
>>>>>
>>>>> I am trying to restore an empty 3-node cluster with the three
>>>>> snapshots taken on another 3-node cluster.
>>>>>
>>>>> What is the best approach to achieve it without loosing any data
>>>>> present in the snapshot.
>>>>>
>>>>> Thank you.
>>>>> Pradeep
>>>>>
>>>>
>>>>
>>>
>>
>

Mime
View raw message