Adding Brisk Node to existing 0.8.0 cluster.

(9 posts) (2 voices)

• Started 1 year ago by agriffith
• Latest reply from agriffith
• This topic is not resolved

Tags:

No tags yet.

1. 

   agriffith
   Member

   I am having trouble with keyspace creation/replication when adding a brisk node to an existing cassandra cluster.

   I moved the brisk.jar into the lib folder of the cassandra nodes, and modified the yaml file to use the brisk simple snitch.

   The cluster comes up fine with all nodes in the cluster and schemas in agreement. The brisk version of nodetool, shows the correct nodes in brisk dc and cassandra dc.

   When creating a new keyspace, it says it can't reach the other node and then after a short pause a message about schemas agree. However, when you describe the cluster there is now a different schema version for each node.

   I don't have this problem when I start the vanilla cassandra node as the 0.8.1 version that comes with the brisk package.

   Are there configuration changes I have missed to add a brisk node running 0.8.1 to an existing cassandra cluster running 0.8.0.

   Here is the output from the system.log on the Brisk node. This keyspace was created on the vanilla cassandra node.

   ERROR [Thread-47] 2011-06-24 11:00:09,002 AbstractCassandraDaemon.java (line 113) Fatal exception in thread Thread[Thread-47,5,main]
   java.lang.AssertionError: No stage for message type null
   at org.apache.cassandra.net.MessagingService.receive(MessagingService.java:445)
   at org.apache.cassandra.net.IncomingTcpConnection.run(IncomingTcpConnection.java:117)

   Aaron

   Posted 1 year ago #
2. 

   joaquin
   Administrator

   Hello,

   This form of adding Brisk won't work. You should install the Brisk binaries in their own folder, stop the Cassandra process, then start the Brisk process. Then in a rolling fashion, do the same procedure to each machine around the ring.

   Beta2 contains a version of Cassandra that wont allow a mixed version cluster.

   Thanks,
   Joaquin

   Posted 1 year ago #
3. 

   agriffith
   Member

   Is it possible to use cassandra 0.8.0 with the brisk beta 2?

   Also, when Brisk has been released, is it going to come packaged with more than one version of cassandra, or will you be forced to use the cassandra version that brisk is packaged with?

   Posted 1 year ago #
4. 

   joaquin
   Administrator

   Brisk Beta 2 already contains a version of Cassandra 0.8.0+, but it doesn't allow you to run a mixed cluster due to the included patches, so all nodes should be Brisk Beta 2.

   Brisk will be packaged with only the one version of Cassandra. The reason for this is due to multiple patches that are going into Cassandra 0.8's branch to make Brisk better with some of it's analytical and file system components.

   That being said, it is very likely that Brisk Beta 3 will release with Cassandra 0.8.1 stable.

   Posted 1 year ago #
5. 

   agriffith
   Member

   I removed Brisk Beta 2 and used Brisk Beta 1 instead and it seemed to work. However, when I added a new 0.8.0 node to the vanilla DC, the brisk node would not see it with a describe cluster using the cassandra-cli. It shows up with the nodetool ring command, but has a status of UNREACHABLE with describe cluster. Doing a describe cluster from the cli connected to either of the vanilla nodes works fine.

   Posted 1 year ago #
6. 

   joaquin
   Administrator

   What kind of snitches are your Cassandra nodes running? How about your Brisk nodes?

   Posted 1 year ago #
7. 

   agriffith
   Member

   all are running the simplebrisksnitch. i moved the brisk.jar file into the /lib folder on the vanilla cassandra nodes.

   the 0.8.1 was released today, so i will try that with the beta2 and they should all be same version, hopefully that will work everything out.

   Posted 1 year ago #
8. 

   joaquin
   Administrator

   Also keep in mind that backwards-compatibility is never been advised or tested. Forward compatibility is advised and used during our upgrade process.

   Again, what we would recommend is installing the same version of Brisk on all machines and starting only the node that you want with HADOOP_ENABLED. Is this a dev cluster or a production cluster?

   Posted 1 year ago #
9. 

   agriffith
   Member

   development cluster. it seems to be running good now with 2 vanilla 0.8.1 nodes and one brisk cassandra node with the hadoop trackers on.

   Posted 1 year ago #

RSS feed for this topic

Reply

You must log in to post.