Rao created CASSANDRA-5671:
------------------------------

             Summary: cassandra automatic token generation issue: each 
datacenter doesnot span the complete set of tokens in NetworkTopologyStrategy
                 Key: CASSANDRA-5671
                 URL: https://issues.apache.org/jira/browse/CASSANDRA-5671
             Project: Cassandra
          Issue Type: Bug
          Components: Core
    Affects Versions: 1.2.5
            Reporter: Rao


When a route is saved, some of the routes save time is taking log longer 
(200ms+) than the other routes (<30ms). When analysed, it looks like the 
routeId (primary key which is a UUID) has a token that maps to a different 
datacenter than the current one, so the request is going accross dc and is 
taking more time. 

We have the following configuration for the keyspace: 2 nodes in each 
datacenter and with replication factor of 2. 

CREATE KEYSPACE grd WITH replication = {
  'class': 'NetworkTopologyStrategy',
  'HYWRCA02': '2',
  'CHRLNCUN': '2'
};

Cassandra Version:  Cassandra 1.2.5 
Using Virtual tokens generated (num_tokens: 256)
partitioner: org.apache.cassandra.dht.Murmur3Partitioner
On save we are using the consistency level of ONE. 
On read we are using the consistency level of local_quorum.

So in this case am expecting the the tokens to be generated in such a way that 
the each datacenter spans the complete set of tokens. So when a save happens it 
always goes to the local data center. Also on reads too, it should go to the 
local dc.

some examples of the nodetool getendpoints:
[cassdra@hltd217 conf]$ nodetool -h hltd217.hydc.sbc.com -p 20000 getendpoints 
grd route 22005151-a250-37b5-bb00-163df3bf0ad6
135.201.73.144 (dc2)
135.201.73.145 (dc2)
150.233.236.97 (dc1)
150.233.236.98 (dc1)

[cassdra@hltd217 conf]$ nodetool -h hltd217.hydc.sbc.com -p 20000 getendpoints 
grd route d1e86f4e-6d74-3bf6-8d76-27f41ae18149
150.233.236.97 (dc1)
135.201.73.144 (dc2)
150.233.236.98 (dc1)
135.201.73.145 (dc2)

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira

Reply via email to