Kubernetes Cassandra: How to Run HA Cassandra on Amazon EKS

This post is part of our ongoing series on running Cassandra on Kubernetes.  We’ve published a number of articles about running Cassandra on Kubernetes for specific platforms and for specific use cases.  If you are looking for a specific Kubernetes platform, check out these related articles.

Running HA Cassandra on Azure Kubernetes Service (AKS)

Running HA Cassandra on Google Kubernetes Engine (GKE)

Running HA Cassandra on Red Hat OpenShift

Running HA Cassandra on IBM Cloud Kubernetes Service (IKS)

Running HA Cassandra on Rancher Kubernetes Engine (RKE)

Running HA Cassandra on IBM Cloud Private

And now, onto the post…

 

Amazon EKS is a fully managed Kubernetes environment running in AWS. With Amazon EKS customers get a highly-available, secure Kubernetes control plane without needing to worry about provisioning, upgrades, or patching. Amazon EKS is certified by the Cloud Native Computing Foundation (CNCF) as Kubernetes conformant, which means it supports all existing plugins and tooling from the Kubernetes community, including Portworx.

Portworx, an EKS launch partner, is a cloud native storage platform to run persistent workloads deployed on a variety of orchestration engines including Kubernetes. With Portworx, customers can manage the database of their choice on any infrastructure using any container scheduler. It provides a single data management layer for all stateful services, no matter where they run.

This tutorial is a walk-through of the steps involved in deploying and managing a highly available Cassandra NoSQL database on Amazon EKS as a Kubernetes statefulset.

In summary, to run HA Cassandra on Amazon EKS you need to:

1. Install an EKS cluster by following instructions in the Amazon docs
2. Install a cloud native storage solution like Portworx as a daemon set on EKS
3. Create a storage class defining your storage requirements like replication factor, snapshot policy, and performance profile
4. Deploy Cassandra as a statefulset on Kubernetes
5. Test failover by killing or cordoning nodes in your cluster
6. Optional – Take an app consistent snapshot of Cassandra
7. Optional – Bootstrap a new Cassandra cluster from snapshot backup

How to set up an EKS cluster

Portworx is fully supported on Amazon EKS. Please follow the instructions to configure an Amazon EKS cluster.

You should have a three-node Kubernetes cluster deployed based on the default EKS configuration.

$ kubectl get nodes
NAME                                            STATUS    ROLES     AGE       VERSION
ip-192-168-146-135.us-west-2.compute.internal   Ready     none	    30m       v1.10.3
ip-192-168-208-187.us-west-2.compute.internal   Ready     none	    30m       v1.10.3
ip-192-168-99-43.us-west-2.compute.internal     Ready     none	    30m       v1.10.3

Installing Portworx in EKS

Installing Portworx on Amazon EKS is not very different from installing it on a Kubernetes cluster setup through Kops. Portworx EKS documentation has the steps involved in running the Portworx cluster in a Kubernetes environment deployed in AWS.

Portworx cluster needs to be up and running on EKS before proceeding to the next step. The kube-system namespace should have the Portworx pods in running state.

$ kubectl get pods -n=kube-system -l name=portworx
NAME             READY     STATUS    RESTARTS   AGE
portworx-rckrx   1/1       Running   0          20m
portworx-vbp7g   1/1       Running   0          20m
portworx-wllbd   1/1       Running   0          20m

Creating a storage class for Cassandra

Once the EKS cluster is up and running, and Portworx is installed and configured, we will deploy a highly available Cassandra database.

Through storage class objects, an admin can define different classes of Portworx volumes that are offered in a cluster. These classes will be used during the dynamic provisioning of volumes. The storage class defines the replication factor, I/O profile (e.g., for a database or a CMS), and priority (e.g., SSD or HDD). These parameters impact the availability and throughput of workloads and can be specified for each volume. This is important because a production database will have different requirements than a development Jenkins cluster.

In this example, the storage class that we deploy has a replication factor of 3s with I/O profile set to “db,” and priority set to “high.” This means that the storage will be optimized for low latency database workloads like Cassandra and automatically placed on the highest performance storage available in the cluster.

$ cat > px-cassandra-sc.yaml << EOF
kind: StorageClass
apiVersion: storage.k8s.io/v1
metadata:
  name: px-storageclass
provisioner: kubernetes.io/portworx-volume
parameters:
  repl: "3"
  io_profile: "db_remote"
  priority_io: "high"
  fg: "true"
EOF

Create the storage class and verify its available in the default namespace.

$ kubectl create -f px-cassandra-sc.yaml
storageclass.storage.k8s.io "px-storageclass" created

$ kubectl get sc
NAME                PROVISIONER                     AGE
px-storageclass     kubernetes.io/portworx-volume   24s
stork-snapshot-sc   stork-snapshot                  27m

Deploying Cassandra StatefulSet on EKS

Finally, let’s create a Cassandra cluster as a Kubernetes statefulset object. Like a Kubernetes deployment, a statefulset manages pods that are based on an identical container spec. Unlike a deployment, a statefulset maintains a sticky identity for each of their Pods. For more details on statefulsets, refer to Kubernetes documentation.

A statefulset in Kubernetes requires a headless service to provide network identity to the pods it creates. The following command and the spec will help you create a headless service for your Cassandra installation.

$ cat > px-cassandra-svc.yaml << EOF
apiVersion: v1
kind: Service
metadata:
  labels:
    app: cassandra
  name: cassandra
spec:
  clusterIP: None
  ports:
    - port: 9042
  selector:
    app: cassandra
EOF

$ kubectl create -f px-cassandra-svc.yaml
service "cassandra" created

Now, let’s go ahead and create a statefulset running Cassandra cluster based on the below spec.

cat > px-cassandra-app.yaml << EOF
apiVersion: apps/v1
kind: StatefulSet
metadata:
  name: cassandra
spec:
  selector:
    matchLabels:
      app: cassandra
  serviceName: cassandra
  replicas: 3
  template:
    metadata:
      labels:
        app: cassandra
    spec:
      schedulerName: stork
      containers:
      - name: cassandra
        image: cassandra:3
        ports:
          - containerPort: 7000
            name: intra-node
          - containerPort: 7001
            name: tls-intra-node
          - containerPort: 7199
            name: jmx
          - containerPort: 9042
            name: cql
        env:
          - name: CASSANDRA_SEEDS
            value: cassandra-0.cassandra.default.svc.cluster.local
          - name: MAX_HEAP_SIZE 
            value: 512M
          - name: HEAP_NEWSIZE
            value: 512M
          - name: CASSANDRA_CLUSTER_NAME
            value: "Cassandra"
          - name: CASSANDRA_DC
            value: "DC1"
          - name: CASSANDRA_RACK
            value: "Rack1"
          - name: CASSANDRA_AUTO_BOOTSTRAP
            value: "false"            
          - name: CASSANDRA_ENDPOINT_SNITCH
            value: GossipingPropertyFileSnitch
        volumeMounts:
        - name: cassandra-data
          mountPath: /var/lib/cassandra
  volumeClaimTemplates:
  - metadata:
      name: cassandra-data
      annotations:
        volume.beta.kubernetes.io/storage-class: px-storageclass
      labels:
         app: cassandra
    spec:
      accessModes: [ "ReadWriteOnce" ]
      resources:
        requests:
          storage: 1Gi
EOF

$ kubectl apply -f px-cassandra-app.yaml
statefulset.apps "cassandra" created

Verify that all the pods are in the Running state before proceeding further.

$ kubectl get statefulset
NAME        DESIRED   CURRENT   AGE
cassandra   3         2         45s

$ kubectl get pods
$ kubectl get pods
NAME          READY     STATUS    RESTARTS   AGE
cassandra-0   1/1       Running   0          2m
cassandra-1   1/1       Running   0          1m
cassandra-2   1/1       Running   1          43s

Let’s also check if persistent volume claims are bound to the volumes.

$ kubectl get pvc
NAME                         STATUS    VOLUME                                     CAPACITY   ACCESS MODES   STORAGECLASS      AGE
cassandra-data-cassandra-0   Bound     pvc-c79d0e58-b1b6-11e8-8c6b-027b3f29e3d8   1Gi        RWO            px-storageclass   4m
cassandra-data-cassandra-1   Bound     pvc-e173fdf6-b1b6-11e8-8c6b-027b3f29e3d8   1Gi        RWO            px-storageclass   3m
cassandra-data-cassandra-2   Bound     pvc-fa515912-b1b6-11e8-8c6b-027b3f29e3d8   1Gi        RWO            px-storageclass   2m

Notice the naming convention followed by Kubernetes for the pods and volume claims. The arbitrary number attached to each object indicates the association of pods and volumes.

We can now inspect the Portworx volume associated with one of the Cassandra pods by accessing the pxctl tool.

$  VOL=`kubectl get pvc | grep cassandra-0 | awk '{print $3}'`
$ PX_POD=$(kubectl get pods -l name=portworx -n kube-system -o jsonpath='{.items[0].metadata.name}')
$ kubectl exec -it $PX_POD -n kube-system -- /opt/pwx/bin/pxctl volume inspect ${VOL}
Volume	:  16315439828490113
	Name            	 :  pvc-c79d0e58-b1b6-11e8-8c6b-027b3f29e3d8
	Size            	 :  1.0 GiB
	Format          	 :  ext4
	HA              	 :  2
	IO Priority     	 :  LOW
	Creation time   	 :  Sep 6 09:25:27 UTC 2018
	Shared          	 :  no
	Status          	 :  up
	State           	 :  Attached: ip-192-168-208-187.us-west-2.compute.internal (192.168.208.187)
	Device Path     	 :  /dev/pxd/pxd16315439828490113
	Labels          	 :  namespace=default,pvc=cassandra-data-cassandra-0
	Reads           	 :  50
	Reads MS        	 :  476
	Bytes Read      	 :  1114112
	Writes          	 :  47
	Writes MS       	 :  1176
	Bytes Written   	 :  16883712
	IOs in progress 	 :  0
	Bytes used      	 :  48 MiB
	Replica sets on nodes:
		Set 0
		  Node 		 : 192.168.208.187 (Pool 0)
		  Node 		 : 192.168.146.135 (Pool 0)
	Replication Status	 :  Up
	Volume consumers	 :
		- Name           : cassandra-0 (c79daea1-b1b6-11e8-8c6b-027b3f29e3d8) (Pod)
		  Namespace      : default
		  Running on     : ip-192-168-208-187.us-west-2.compute.internal
		  Controlled by  : cassandra (StatefulSet)

The output from the above command confirms the creation of volumes that are backing Cassandra nodes.

We can also use Cassandra’s nodetool  to check the status of the cluster.

$ kubectl exec cassandra-0 -- nodetool status
Datacenter: DC1
===============
Status=Up/Down
|/ State=Normal/Leaving/Joining/Moving
--  Address         Load       Tokens       Owns (effective)  Host ID                               Rack
UN  192.168.196.73  108.62 KiB  256          67.1%             e567984d-0143-4d9b-9738-edea3b68c3f6  Rack1
UN  192.168.81.245  69.94 KiB  256          67.3%             b84b4537-61fe-41bc-9009-d881fcc38f46  Rack1
UN  192.168.172.87  88.86 KiB  256          65.6%             94ef766a-6100-464b-abcb-f9153aaf331a  Rack1

To get the pods and hosts associated with the Cassandra cluster, run the below command:

$ kubectl get pods -l app=cassandra -o json | jq '.items[] | {"name": .metadata.name,"hostname": .spec.nodeName, "hostIP": .status.hostIP, "PodIP": .status.podIP}'
{
  "name": "cassandra-0",
  "hostname": "ip-192-168-208-187.us-west-2.compute.internal",
  "hostIP": "192.168.208.187",
  "PodIP": "192.168.196.73"
}
{
  "name": "cassandra-1",
  "hostname": "ip-192-168-146-135.us-west-2.compute.internal",
  "hostIP": "192.168.146.135",
  "PodIP": "192.168.172.87"
}
{
  "name": "cassandra-2",
  "hostname": "ip-192-168-99-43.us-west-2.compute.internal",
  "hostIP": "192.168.99.43",
  "PodIP": "192.168.81.245"
}

Failing over Cassandra pod on Kubernetes

Populating sample data

Let’s populate the database with some sample data by accessing the first node of the Cassandra cluster. We will do this by invoking Cassandra shell, cqlsh in one of the pods.

$ kubectl exec -it cassandra-0 -- cqlsh
Connected to Cassandra at 127.0.0.1:9042.
[cqlsh 5.0.1 | Cassandra 3.11.3 | CQL spec 3.4.4 | Native protocol v4]
Use HELP for help.
cqlsh>

Now that we are inside the shell, we can create a keyspace and populate it.

CREATE KEYSPACE classicmodels WITH REPLICATION = { 'class' : 'SimpleStrategy', 'replication_factor' : 3 };
	
CONSISTENCY QUORUM;
Consistency level set to QUORUM.

use classicmodels;

CREATE TABLE offices (officeCode text PRIMARY KEY, city text, phone text, addressLine1 text, addressLine2 text, state text, country text, postalCode text, territory text);

INSERT into offices(officeCode, city, phone, addressLine1, addressLine2, state, country ,postalCode, territory) values 
	('1','San Francisco','+1 650 219 4782','100 Market Street','Suite 300','CA','USA','94080','NA');

INSERT into offices(officeCode, city, phone, addressLine1, addressLine2, state, country ,postalCode, territory) values 
	('2','Boston','+1 215 837 0825','1550 Court Place','Suite 102','MA','USA','02107','NA');

INSERT into offices(officeCode, city, phone, addressLine1, addressLine2, state, country ,postalCode, territory) values 	
	('3','NYC','+1 212 555 3000','523 East 53rd Street','apt. 5A','NY','USA','10022','NA');

INSERT into offices(officeCode, city, phone, addressLine1, addressLine2, state, country ,postalCode, territory) values 
	('4','Paris','+33 14 723 4404','43 Rue Jouffroy abbans', NULL ,NULL,'France','75017','EMEA');

INSERT into offices(officeCode, city, phone, addressLine1, addressLine2, state, country ,postalCode, territory) values 		
	('5','Tokyo','+81 33 224 5000','4-1 Kioicho',NULL,'Chiyoda-Ku','Japan','102-8578','Japan');

INSERT into offices(officeCode, city, phone, addressLine1, addressLine2, state, country ,postalCode, territory) values 
	('6','Sydney','+61 2 9264 2451','5-11 Wentworth Avenue','Floor #2',NULL,'Australia','NSW 2010','APAC');

INSERT into offices(officeCode, city, phone, addressLine1, addressLine2, state, country ,postalCode, territory) values 
	('7','London','+44 20 7877 2041','25 Old Broad Street','Level 7',NULL,'UK','EC2N 1HN','EMEA');

INSERT into offices(officeCode, city, phone, addressLine1, addressLine2, state, country ,postalCode, territory) values 
	('8','Mumbai','+91 22 8765434','BKC','Building 2',NULL,'MH','400051','APAC');

Let’s verify that the data is populated.

SELECT * FROM classicmodels.offices;

 officecode | addressline1           | addressline2 | city          | country   | phone            | postalcode | state      | territory
------------+------------------------+--------------+---------------+-----------+------------------+------------+------------+-----------
          6 |  5-11 Wentworth Avenue |     Floor #2 |        Sydney | Australia |  +61 2 9264 2451 |   NSW 2010 |       null |      APAC
          7 |    25 Old Broad Street |      Level 7 |        London |        UK | +44 20 7877 2041 |   EC2N 1HN |       null |      EMEA
          4 | 43 Rue Jouffroy abbans |         null |         Paris |    France |  +33 14 723 4404 |      75017 |       null |      EMEA
          3 |   523 East 53rd Street |      apt. 5A |           NYC |       USA |  +1 212 555 3000 |      10022 |         NY |        NA
          5 |            4-1 Kioicho |         null |         Tokyo |     Japan |  +81 33 224 5000 |   102-8578 | Chiyoda-Ku |     Japan
          8 |                    BKC |   Building 2 |        Mumbai |        MH |   +91 22 8765434 |     400051 |       null |      APAC
          2 |       1550 Court Place |    Suite 102 |        Boston |       USA |  +1 215 837 0825 |      02107 |         MA |        NA
          1 |      100 Market Street |    Suite 300 | San Francisco |       USA |  +1 650 219 4782 |      94080 |         CA |        NA

(8 rows)
cqlsh:classicmodels>

Exit from the client shell to return to the host.

You can run the select query by accessing cqlsh from any of the pods of the statefulset.

Run nodetool  again to check the replication of the data. The below command shows that the hosts on which the row with officecode=6 is available.

$ kubectl exec -it cassandra-0 -- nodetool getendpoints classicmodels offices 6
192.168.172.87
192.168.81.245
192.168.196.73

Simulating node failure

Let’s get the node name where the first pod of Cassandra statefulset is running.

$ NODE=`kubectl get pods cassandra-0 -o json | jq -r .spec.nodeName`

Now, let’s simulate the node failure by cordoning off the Kubernetes node.

$ kubectl cordon ${NODE}
node "ip-192-168-208-187.us-west-2.compute.internal" cordoned

The above command disabled scheduling on one of the nodes.

$ kubectl get nodes
NAME                                            STATUS                     ROLES     AGE       VERSION
ip-192-168-146-135.us-west-2.compute.internal   Ready                          1h        v1.10.3
ip-192-168-208-187.us-west-2.compute.internal   Ready,SchedulingDisabled       1h        v1.10.3
ip-192-168-99-43.us-west-2.compute.internal     Ready                          1h        v1.10.3

Let’s go ahead and delete the pod cassandra-0 running on the node that is cordoned off.

$ kubectl delete pod cassandra-0
pod "cassandra-0" deleted

Kubernetes controller now tries to create the pod on a different node.

$ kubectl get pods -o wide
NAME          READY     STATUS              RESTARTS   AGE
cassandra-0   0/1       ContainerCreating   0          2s
cassandra-1   1/1       Running             0          54m
cassandra-2   1/1       Running             1          53m

Wait for the pod to be in Running state on the node.

$ kubectl get pods -o wide
NAME          READY     STATUS              RESTARTS   AGE     
cassandra-0   1/1       Running              0          1m            
cassandra-1   1/1       Running             0          54m           
cassandra-2   1/1       Running             1          53m          

Finally, let’s verify that the data is still available.

Verifying that the data is intact

Let’s access the data in the first pod of the statefulset – cassandra-0.

$ kubectl exec cassandra-0 -- cqlsh -e 'select * from classicmodels.offices'

 officecode | addressline1           | addressline2 | city          | country   | phone            | postalcode | state      | territory
------------+------------------------+--------------+---------------+-----------+------------------+------------+------------+-----------
          6 |  5-11 Wentworth Avenue |     Floor #2 |        Sydney | Australia |  +61 2 9264 2451 |   NSW 2010 |       null |      APAC
          7 |    25 Old Broad Street |      Level 7 |        London |        UK | +44 20 7877 2041 |   EC2N 1HN |       null |      EMEA
          4 | 43 Rue Jouffroy abbans |         null |         Paris |    France |  +33 14 723 4404 |      75017 |       null |      EMEA
          3 |   523 East 53rd Street |      apt. 5A |           NYC |       USA |  +1 212 555 3000 |      10022 |         NY |        NA
          5 |            4-1 Kioicho |         null |         Tokyo |     Japan |  +81 33 224 5000 |   102-8578 | Chiyoda-Ku |     Japan
          8 |                    BKC |   Building 2 |        Mumbai |        MH |   +91 22 8765434 |     400051 |       null |      APAC
          2 |       1550 Court Place |    Suite 102 |        Boston |       USA |  +1 215 837 0825 |      02107 |         MA |        NA
          1 |      100 Market Street |    Suite 300 | San Francisco |       USA |  +1 650 219 4782 |      94080 |         CA |        NA

(8 rows)

Observe that the data is still there and all the content is intact! We can also run the nodetool  again to see that the new node is indeed a part of the statefulset.

$ kubectl exec cassandra-1 -- nodetool status
Datacenter: DC1
===============
Status=Up/Down
|/ State=Normal/Leaving/Joining/Moving
--  Address          Load       Tokens       Owns (effective)  Host ID                               Rack
UN  192.168.148.159  100.44 KiB  256          100.0%            fd1610c8-7745-49eb-b801-983cde4e1b85  Rack1
UN  192.168.81.245   186.62 KiB  256          100.0%            b84b4537-61fe-41bc-9009-d881fcc38f46  Rack1
UN  192.168.172.87   196.54 KiB  256          100.0%            94ef766a-6100-464b-abcb-f9153aaf331a  Rack1

Capturing Application Consistent Snapshots to Restore Data

Portworx enables storage admins to perform backup and restore operations through the snapshots. 3DSnap is a feature to capture consistent snapshots from multiple nodes of a database cluster. This is highly recommended when running a multi-node Cassandra cluster as a Kubernetes statefulset. 3DSnap will create the snapshot from each of the nodes in the cluster, which ensures that the state is accurately captured from the distributed cluster.

3DSnap allows administrators to execute commands just before taking the snapshot and right after completing the task of taking a snapshot. These triggers will ensure that the data is fully committed to the disk before the snapshot. Similarly, it is possible to run a workload-specific command to refresh or force a sync immediately after restoring the snapshot.

This section will walk you through the steps involved in creating and restoring a 3DSnap for the Cassandra statefulset.

Creating a 3DSnap

It’s a good idea to flush the data to the disk before initiating the snapshot creation. This is defined through a rule, which is a Custom Resource Definition created by Stork.

$ cat > px-cassandra-rule.yaml << EOF
apiVersion: stork.libopenstorage.org/v1alpha1
kind: Rule
metadata:
  name: px-cassandra-rule
spec:
  - podSelector:
      app: cassandra
    actions:
    - type: command
      value: nodetool flush
EOF

Create the rule from the above YAML file.

$ kubectl create -f px-cassandra-rule.yaml
rule.stork.libopenstorage.org "px-cassandra-rule" created

We will now initiate a 3DSnap task to backup all the PVCs associated with the Cassandra pods belonging to the statefulset.

$ cat > px-cassandra-snap.yaml << EOF
apiVersion: volumesnapshot.external-storage.k8s.io/v1
kind: VolumeSnapshot
metadata:
  name: cassandra-3d-snapshot
  annotations:
    portworx.selector/app: cassandra
    stork.rule/pre-snapshot: px-cassandra-rule
spec:
  persistentVolumeClaimName: cassandra-data-cassandra-0
EOF

$ kubectl create -f px-cassandra-snap.yaml
volumesnapshot.volumesnapshot.external-storage.k8s.io "cassandra-3d-snapshot" created

Let’s now verify that the snapshot creation is successful.

$ kubectl get volumesnapshot
NAME                                                                                    AGE
cassandra-3d-snapshot                                                                   7s
cassandra-3d-snapshot-cassandra-data-cassandra-0-f7ffa638-cdda-11e8-a2f0-061f808edbd0   2s
cassandra-3d-snapshot-cassandra-data-cassandra-1-f7ffa638-cdda-11e8-a2f0-061f808edbd0   1s
cassandra-3d-snapshot-cassandra-data-cassandra-2-f7ffa638-cdda-11e8-a2f0-061f808edbd0   2s

$ kubectl get volumesnapshotdatas
NAME                                                                                    AGE
cassandra-3d-snapshot-cassandra-data-cassandra-0-f7ffa638-cdda-11e8-a2f0-061f808edbd0   7s
cassandra-3d-snapshot-cassandra-data-cassandra-1-f7ffa638-cdda-11e8-a2f0-061f808edbd0   6s
cassandra-3d-snapshot-cassandra-data-cassandra-2-f7ffa638-cdda-11e8-a2f0-061f808edbd0   7s
k8s-volume-snapshot-fb802623-cdda-11e8-ad5b-7e7e4ed5cafb                                6s

Restoring from a 3DSnap

Let’s now restore from the 3DSnap. Before that, we will simulate the database crash by deleting the statefulset and associated PVCs.

$ kubectl delete sts cassandra
statefulset.apps "cassandra" deleted

$ kubectl delete pvc -l app=cassandra
persistentvolumeclaim "cassandra-data-cassandra-0" deleted
persistentvolumeclaim "cassandra-data-cassandra-1" deleted
persistentvolumeclaim "cassandra-data-cassandra-2" deleted

Now our Kubernetes cluster has no database running. Let’s go ahead and restore the data from the snapshot before re-launching Cassandra statefulset.

We will now create three Persistent Volume Claims (PVCs) from existing 3DSnap with exactly the same volume name that the statefulset expects. When the pods are created as a part of the statefulset, they point to the existing PVCs which are already populated with the data restored from the snapshots.

Let’s create three PVCs from the 3DSnap snapshots. Notice how the annotation points to the snapshot in each PVC manifest.

$ cat > px-cassandra-pvc-0.yaml << EOF
apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: cassandra-data-cassandra-0
  annotations:
    snapshot.alpha.kubernetes.io/snapshot: "cassandra-3d-snapshot-cassandra-data-cassandra-0-f7ffa638-cdda-11e8-a2f0-061f808edbd0"
spec:
  accessModes:
     - ReadWriteOnce
  storageClassName: stork-snapshot-sc
  resources:
    requests:
      storage: 5Gi
EOF

$ cat > px-cassandra-pvc-1.yaml << EOF
apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: cassandra-data-cassandra-1
  annotations:
    snapshot.alpha.kubernetes.io/snapshot: "cassandra-3d-snapshot-cassandra-data-cassandra-0-f7ffa638-cdda-11e8-a2f0-061f808edbd0"
spec:
  accessModes:
     - ReadWriteOnce
  storageClassName: stork-snapshot-sc
  resources:
    requests:
      storage: 5Gi
EOF

$ cat > px-cassandra-pvc-2.yaml << EOF
apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: cassandra-data-cassandra-0
  annotations:
    snapshot.alpha.kubernetes.io/snapshot: "cassandra-3d-snapshot-cassandra-data-cassandra-2-f7ffa638-cdda-11e8-a2f0-061f808edbd0"
spec:
  accessModes:
     - ReadWriteOnce
  storageClassName: stork-snapshot-sc
  resources:
    requests:
      storage: 5Gi
EOF

Create the PVCs from the above definitions.

$ kubectl create -f px-cassandra-snap-pvc-0.yaml
persistentvolumeclaim "cassandra-data-cassandra-0" created

$ kubectl create -f px-cassandra-snap-pvc-1.yaml
persistentvolumeclaim "cassandra-data-cassandra-1" created

$ kubectl create -f px-cassandra-snap-pvc-2.yaml
persistentvolumeclaim "cassandra-data-cassandra-2" created

$ kubectl create -f px-cassandra-app.yaml
statefulset.apps "cassandra" created

Verify that the new PVCs are ready and bound.

$ kubectl get pvc
NAME                         STATUS    VOLUME                                     CAPACITY   ACCESS MODES   STORAGECLASS        AGE
cassandra-data-cassandra-0   Bound     pvc-90218646-cddb-11e8-be72-02e1e4a8c0ea   5Gi        RWO            stork-snapshot-sc   12s
cassandra-data-cassandra-1   Bound     pvc-92e74a65-cddb-11e8-a2f0-061f808edbd0   5Gi        RWO            stork-snapshot-sc   7s
cassandra-data-cassandra-2   Bound     pvc-95731baa-cddb-11e8-a2f0-061f808edbd0   5Gi        RWO            stork-snapshot-sc   3s

With the PVCs in place, we are ready to launch the statefulset with no changes to the YAML file. Everything remains exactly the same while the data is already restored from the snapshots.

$ kubectl create -f px-cassandra-app.yaml
statefulset.apps "cassandra" created

Check the data through the cqlsh from one the Cassandra pods.

$ kubectl exec cassandra-0 -- cqlsh -e 'select * from classicmodels.offices'

 officecode | addressline1           | addressline2 | city          | country   | phone            | postalcode | state      | territory
------------+------------------------+--------------+---------------+-----------+------------------+------------+------------+-----------
          6 |  5-11 Wentworth Avenue |     Floor #2 |        Sydney | Australia |  +61 2 9264 2451 |   NSW 2010 |       null |      APAC
          7 |    25 Old Broad Street |      Level 7 |        London |        UK | +44 20 7877 2041 |   EC2N 1HN |       null |      EMEA
          4 | 43 Rue Jouffroy abbans |         null |         Paris |    France |  +33 14 723 4404 |      75017 |       null |      EMEA
          3 |   523 East 53rd Street |      apt. 5A |           NYC |       USA |  +1 212 555 3000 |      10022 |         NY |        NA
          5 |            4-1 Kioicho |         null |         Tokyo |     Japan |  +81 33 224 5000 |   102-8578 | Chiyoda-Ku |     Japan
          8 |                    BKC |   Building 2 |        Mumbai |        MH |   +91 22 8765434 |     400051 |       null |      APAC
          2 |       1550 Court Place |    Suite 102 |        Boston |       USA |  +1 215 837 0825 |      02107 |         MA |        NA
          1 |      100 Market Street |    Suite 300 | San Francisco |       USA |  +1 650 219 4782 |      94080 |         CA |        NA

(8 rows)

Congratulations! You have successfully restored an application consistent snapshot for Cassandra.

Summary

Portworx can easily be deployed on Amazon EKS to run stateful workloads in production. It integrates well with K8s statefulsets by providing dynamic provisioning. Additional operations such as expanding the volumes and performing backups stored as snapshots on object storage can be performed while managing production workloads.