You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I tried to expose the cassandra service as a Nodeport using Kubernetes version 1.5.7. I ran the cassandra-replication controller as defined in this project.
I used the cassandra-stress tool for testing a five-node cassandra cluster using cql2
The cassandra-stress tool reported that it retrieved IP addresses for all cassandra nodes: one IP address was a virtual IP address of the openstack cloud. The other IP addresses where PodIP addresses, and therefore not reachable from outside the cluster.
The throughput was very low as a result.
When I added the VM, on which the cassandra-stress tool ran, as a Kubernetes worker node, the throughput was back to normal.
I experienced the same problem with the YCSB benchmark tool.
Of course returning the cassandra nodes as PodIP addresses is not ideal as all these IP addresses are transient. But this is how the headless cassandra-peer service works via DNS lookup. Is it possible to configure the headless service such that host ports are added as a DNS A record?
I would like to know why the cassandra-stress tool needs to have access to all cassandra nodes. Why isn't it enough to just connect to the VM on which I connect to the Nodeport.
The text was updated successfully, but these errors were encountered:
First of all thanks for this great project.
I tried to expose the cassandra service as a Nodeport using Kubernetes version 1.5.7. I ran the cassandra-replication controller as defined in this project.
I used the cassandra-stress tool for testing a five-node cassandra cluster using cql2
The cassandra-stress tool reported that it retrieved IP addresses for all cassandra nodes: one IP address was a virtual IP address of the openstack cloud. The other IP addresses where PodIP addresses, and therefore not reachable from outside the cluster.
The throughput was very low as a result.
When I added the VM, on which the cassandra-stress tool ran, as a Kubernetes worker node, the throughput was back to normal.
I experienced the same problem with the YCSB benchmark tool.
Of course returning the cassandra nodes as PodIP addresses is not ideal as all these IP addresses are transient. But this is how the headless cassandra-peer service works via DNS lookup. Is it possible to configure the headless service such that host ports are added as a DNS A record?
I would like to know why the cassandra-stress tool needs to have access to all cassandra nodes. Why isn't it enough to just connect to the VM on which I connect to the Nodeport.
The text was updated successfully, but these errors were encountered: