K8-scalar is a workbench for implementing and evaluating different self-adaptive approaches to autoscaling container-orchestrated services.
The k8-scalar artifact has been used and validated in the context of autoscalers for database clusters. Although autoscalers for database clusters or multi-tier applications have been researched, developing an effective autoscaler for databases is still an art, rather than a science.
To deploy cass-operator, run the following commands from the root directory of this repository.
- Install dependency of cass-operator:
helm install cert-manager jetstack/cert-manager --namespace cert-manager --create-namespace --set installCRDs=true - Install a CassandraDatacenter:
helm install cass-dc operations/cass-operator-dc --namespace cass-operator --create-namespace - For unknown reasons, the initial installation does not properly start the datacenter, but reapplying the CassandraDatacenter YAML solves the issue:
helm upgrade -n cass-operator cass-dc operations/cass-operator-dc
To deploy the monitoring core and k8-scalar experiment controller, run the following commands.
- Remove the existing clusterrole:
role=kubectl get clusterrole | grep heapster | head -n1 | awk '{print $1;}'
kubectl delete clusterrole $role
- Install monitoring core:
helm install operations/monitoring-core --generate-name --namespace=kube-system - Install experiment controller:
helm install experiment-controller operations/experiment-controller
To set up Cassandra for an experiment, create a keyspace and table with the desired replication factor.
kubectl exec -n cass-operator cluster1-dc1-default-sts-0 -- cqlsh -e "CREATE KEYSPACE IF NOT EXISTS scalar WITH replication = {'class':'SimpleStrategy', 'replication_factor':2};"kubectl exec -n cass-operator cluster1-dc1-default-sts-0 -- cqlsh -e "CREATE TABLE IF NOT EXISTS scalar.logs (id text PRIMARY KEY, timestamp text, message text);"
As an alternative to creating a keyspace and table manually, we experimented with creating a pre-start k8-scalar User called CassandraAdminInitUser.
To run an experiment, shell into the experiment controller with kubectl exec -it experiment-controller-0 -- bash.
Then configure experiment.properties, and start scalar.jar with java -jar target/scalar-1.0.0.jar etc/platform.properties etc/experiment.properties.
After completing an experiment, the keyspace should be removed before starting a new experiment: kubectl exec -n cass-operator cluster1-dc1-default-sts-2 -- cqlsh -e "DROP KEYSPACE scalar;".
- Eddy Truyen, Wito Delnat, Ansar Rafique, Dimitri Van Landuyt, Wouter Joosen, "K8-Scalar: A workbench to compare autoscalers for container-orchestrated database clusters", in Proceedings of the 13th International Symposium on Software Engineering for Adaptive and Self-Managing Systems (SEAMS 2018), Gothenburg, Sweden. (pdf).