As Jonathan mentioned, if you are using vnodes , you should back up nodetool 
output and assign same token to nodes and then copy corresponding sstables.

If using, initial token, then assign same value for subsequent node.

Sstable loader should work independently, not sure why you are getting wrong 
counts for that

Sent from my iPhone

> On May 13, 2017, at 9:34 AM, Jonathan Haddad <j...@jonhaddad.com> wrote:
> 
> Did you create the nodes with the same tokens?
> 
>> On Sat, May 13, 2017 at 8:44 AM srinivasarao daruna <sree.srin...@gmail.com> 
>> wrote:
>> Hi, 
>> 
>> We have a cassandra cluster built on Apache Cassandra 3.9 with 6 nodes and 
>> RF = 3. As part of re-building the cluster, we are testing the backup and 
>> restore strategy. 
>> 
>> We took the snapshot and uploaded the files to S3 and data has been saved 
>> the data with folder names (backup_folder1 - 6 for nodes 1 - 6).
>> Created a new cluster with the same number of nodes, and copied the data 
>> from S3 and created the schema.
>> 
>> Strategy 1: (using nodetool refresh)
>> 1) Copied back the data from S3 into one machine each based on the folders 
>> created (backup_folder1  - 6 to 6 nodes)
>> 2) and performed nodetool refresh on the cluster.
>> 
>> Ran the count:
>> 
>> Count on previous cluster: 12125800
>> Count on new cluster: 10504780
>> 
>> Strategy 2: using sstableloader
>> 
>> 1) Copied back the data from S3 into one machine each based on the folders 
>> created (backup_folder1  - 6 to 6 nodes)
>> 2) and performed sstableloader on each node.
>> 
>> Ran the count:
>> 
>> Count on previous cluster: 12125800
>> Count on new cluster: 11705084
>> 
>> 
>> Looking at the results, i have bit disappointed that neither of the approach 
>> resulted 100% restore for me. 
>> If there is an error in taking the backup, it should have not given 
>> different counts.
>> 
>> Any ideas on successful back-up and restore strategies.? and what could ve 
>> gone wrong in my process.?
>> 
>> Thank You,
>> Regards, 
>> Srini
>> 

Reply via email to