search

k8ssandra / k8ssandra-operator

The Kubernetes operator for K8ssandra
https://k8ssandra.io/
Apache License 2.0
176 stars 79 forks source link

Do we support on-demand Cassandra setup via PV? #1044

Open chenweisomebody126 opened 1 year ago

chenweisomebody126 commented 1 year ago

What is missing? My end goal is to have Cassandra turn on and off in an on-demand fashion and all data are persisted in a long-standing Persistent Volume. Do we support it now?

Why do we need it? Our usage case is that we expect to set up Cassandra when someone pays for it.

┆Issue is synchronized with this Jira Story by Unito ┆Issue Number: K8OP-77

adejanovski commented 1 year ago

Hi @chenweisomebody126,

yes, Cassandra pods are controlled by statefulsets. This means the data gets persisted between pod restarts. We expose the capability of stopping datacenters through the datacenters[].stopped field: https://docs.k8ssandra.io/reference/crd/k8ssandra-operator-crds-latest/#k8ssandraclusterspeccassandradatacentersindex

So you can create a cluster, let it run as long as needed, then stop its Cassandra pods by setting the stopped flag to true. Then remove the flag to get it running again. It will re-attach the PVs if they are still available (which depends on your storage classes and PV management).

chenweisomebody126 commented 1 year ago

Hi @chenweisomebody126,

yes, Cassandra pods are controlled by statefulsets. This means the data gets persisted between pod restarts. We expose the capability of stopping datacenters through the datacenters[].stopped field: https://docs.k8ssandra.io/reference/crd/k8ssandra-operator-crds-latest/#k8ssandraclusterspeccassandradatacentersindex

So you can create a cluster, let it run as long as needed, then stop its Cassandra pods by setting the stopped flag to true. Then remove the flag to get it running again. It will re-attach the PVs if they are still available (which depends on your storage classes and PV management).

If I understand correctly, we can turn on and off the Cassandra pods but need to leave the cluster running, right? I was wondering if we allow turnning down the cluster to save instance cost and then spin up the cluster on-demand and mount the long-existing PV, e.g. azure blob storage to the stateful set.

adejanovski commented 1 year ago

Sure, as long as your etcd is persisted between restarts of the cluster. There could be some implementation specific details, so that's something to experiment first.

chenweisomebody126 commented 1 year ago

@adejanovski Thanks! Another follow-up question would be, do we have an answer for this https://forum.k8ssandra.io/t/k8ssandra-cluster-horizontal-autoscaling/441?

chenweisomebody126 commented 1 year ago

Thanks for answering these. One last question. @adejanovski Do we have a trigger-based approach to set the stopped flag to true for Cassandra pods? Imagine, we have a kafka event or even an api call as a trigger, How to make K8s set the stopped flag?

adejanovski commented 1 year ago

You have to build the automation interact with the k8s API to patch the K8ssandraCluster spec and set the stopped flag to true.