Open in app

Sign in

Write

Sign in

Aerospike Kubernetes — Talking Cluster to Cluster using XDR-Proxy

Naresh Maharaj
Aerospike Developer Blog
Published in
11 min readFeb 8, 2023

In this article we focus on establishing connectivity between 2 Aerospike clusters. The goal is to use Aerospike’s Cross Data Center Replication feature ( XDR ) to seamlessly send data from a source cluster to a destination cluster. The source cluster needs network visibility of all Aerospike service ports in the remote cluster, and this can present problems, particularly in a Kubernetes environment. Placing a proxy server in front of the private Kubernetes destination cluster can overcome this problem and achieve the desired goal. To demonstrate the solution we start by installing the Kubernetes Operator that will schedule our source and destination databases. In this example, we set up our replication in one direction. Aerospike is capable of supporting ‘master/master’ replication and provides a conflict resolution mechanism in the event of update clashes. This too could be supported using the XDR proxy.

Aerospike Kubernetes Operator

The following Kubernetes nodes have been created using EKS. You can display the following private and public IP addresses from listing the nodes with the kubectl command.

kubectl get nodes -o wide
NAME                             STATUS   ROLES    AGE     VERSION                INTERNAL-IP      EXTERNAL-IP      OS-IMAGE         KERNEL-VERSION                 CONTAINER-RUNTIME
ip-192-168-11-132.ec2.internal   Ready    <none>   2m53s   v1.22.15-eks-fb459a0   192.168.11.132   44.201.67.177    Amazon Linux 2   5.4.219-126.411.amzn2.x86_64   docker://20.10.17
ip-192-168-31-131.ec2.internal   Ready    <none>   2m52s   v1.22.15-eks-fb459a0   192.168.31.131   44.192.83.79     Amazon Linux 2   5.4.219-126.411.amzn2.x86_64   docker://20.10.17
ip-192-168-41-140.ec2.internal   Ready    <none>   2m51s   v1.22.15-eks-fb459a0   192.168.41.140   18.208.222.35    Amazon Linux 2   5.4.219-126.411.amzn2.x86_64   docker://20.10.17
ip-192-168-41-63.ec2.internal    Ready    <none>   2m51s   v1.22.15-eks-fb459a0   192.168.41.63    54.173.138.131   Amazon Linux 2   5.4.219-126.411.amzn2.x86_64   docker://20.10.17
ip-192-168-59-220.ec2.internal   Ready    <none>   2m52s   v1.22.15-eks-fb459a0   192.168.59.220   54.227.122.222   Amazon Linux 2   5.4.219-126.411.amzn2.x86_64   docker://20.10.17
ip-192-168-6-124.ec2.internal    Ready    <none>   2m51s   v1.22.15-eks-fb459a0   192.168.6.124    35.174.60.1      Amazon Linux 2   5.4.219-126.411.amzn2.x86_64   docker://20.10.1

Start by getting a copy of the Aerospike git repo for the Kubernetes Operator.

git clone https://github.com/aerospike/aerospike-kubernetes-operator.git
cp features.conf aerospike-kubernetes-operator/config/samples/secrets/.

Setup

Run the following commands in the order specified. Wait for the csv “Succeeded phase” to show up after running this line. Initially it might take between 30 seconds and a minute to show up. kubectl get csv -n operators -w

cd aerospike-kubernetes-operator/
kubectl apply -f config/samples/storage/eks_ssd_storage_class.yaml
curl -sL https://github.com/operator-framework/operator-lifecycle-manager/releases/download/v0.22.0/install.sh | bash -s v0.22.0
kubectl create -f https://operatorhub.io/install/aerospike-kubernetes-operator.yaml
kubectl get csv -n operators -w
cd..
git clone https://github.com/nareshmaharaj-consultant/kubernetes-anything
cd kubernetes-anything
./createNamespace.sh aerospike
cd ../aerospike-kubernetes-operator/
kubectl -n aerospike create secret generic aerospike-secret --from-file=config/samples/secrets
kubectl -n aerospike create secret generic auth-secret --from-literal=password='admin123'

Destination Cluster

Use the following yaml configuration file for our destination cluster. Save the file and name it ssd1_xdr_dest_6.1_cluster_cr.yaml:

apiVersion: asdb.aerospike.com/v1beta1
kind: AerospikeCluster
metadata:
  name: aerocluster-dest-xdr
  namespace: aerospike

spec:
  size: 1
  image: aerospike/aerospike-server-enterprise:6.1.0.2

  storage:
    filesystemVolumePolicy:
      initMethod: deleteFiles
      cascadeDelete: true
    blockVolumePolicy:
      cascadeDelete: true
    volumes:
      - name: workdir
        aerospike:
          path: /opt/aerospike
        source:
          persistentVolume:
            storageClass: ssd
            volumeMode: Filesystem
            size: 1Gi
      - name: ns
        aerospike:
          path: /opt/aerospike/data/
        source:
          persistentVolume:
            storageClass: ssd
            volumeMode: Filesystem
            size: 1Gi
      - name: aerospike-config-secret
        source:
          secret:
            secretName: aerospike-secret
        aerospike:
          path: /etc/aerospike/secret

  podSpec:
    multiPodPerHost: true

  aerospikeAccessControl:
    roles:
      - name: writer
        privileges:
        - read-write
      - name: reader
        privileges:
        - read
    users:
      - name: admin
        secretName: auth-secret
        roles:
          - sys-admin
          - user-admin
          - read-write
      - name: xdr-writer
        secretName: xdr-user-auth-secret
        roles:
          - writer

  aerospikeConfig:
    service:
      feature-key-file: /etc/aerospike/secret/features.conf
    security: {}
    network:
      service:
        port: 3000
      fabric:
        port: 3001
      heartbeat:
        port: 3002
    namespaces:
      - name: test
        memory-size: 134217728
        replication-factor: 1
        storage-engine:
          type: device
          files:
            - /opt/aerospike/data/test.dat
          filesize: 1073741824
          data-in-memory: true

Create the following Kubernetes resources for our Aerospike destination cluster:

  • xdr destination database user login credentials, as a Kubernetes secret
  • destination database cluster using our yaml file named ssd1_xdr_dest_6.1_cluster_cr.yaml
export secret_auth_name=xdr-user-auth-secret
export password_secret=admin123
kubectl -n aerospike create secret generic $secret_auth_name --from-literal=password=$password_secret
kubectl create -f ssd1_xdr_dest_6.1_cluster_cr.yaml
kubectl -n aerospike get po -w

You should see the database pods running successfully.

kubectl get po -n aerospike -w
NAME                       READY   STATUS     RESTARTS   AGE
aerocluster-dest-xdr-0-0   0/1     Init:0/1   0          13s
aerocluster-dest-xdr-0-0   0/1     Init:0/1   0          18s
aerocluster-dest-xdr-0-0   0/1     PodInitializing   0          19s
aerocluster-dest-xdr-0-0   1/1     Running           0          24s

XDR-Proxy

Next we set up the xdr-proxy. If we look back at the main title image above, you will notice that we are working from the RIGHT hand side to the LEFT hand side in that order.

Configuration

Create the following xdr-proxy configuration file. Replace the seeds address with a FQN DNS name for the destination database pod(s) you created earlier. Multiple seed addresses may be added (recommended in production).

cd ..
mkdir -p xdr-cfg/etc/auth
cd xdr-cfg/etc/

cat <<EOF> aerospike-xdr-proxy.yml
# Change the configuration for your use case.
# Naresh Maharaj
# Refer to https://www.aerospike.com/docs/connectors/enterprise/xdr-proxy/configuration/index.html
# for details.

# The connector's listening ports, manage service, TLS and network interface.
service:
  port: 8901
  # Aerospike Enterprise Server >= 5.0
  manage:
    address: 0.0.0.0
    port: 8902

# The destination aerospike cluster.
aerospike:
  seeds:
    - aerocluster-dest-xdr-0-0.aerospike.svc.cluster.local:
        port: 3000
  credentials:
    username: xdr-writer
    password-file: /etc/aerospike-xdr-proxy/auth/password_DC1.txt
    auth-mode: internal

# The logging config
logging:
  enable-console-logging: true
  file: /var/log/aerospike-xdr-proxy/aerospike-xdr-proxy.log
  levels:
    root: debug
    record-parser: debug
    server: debug
    com.aerospike.connect: debug
  # Ticker log interval in seconds
  ticker-interval: 3600
EOF

sudo tee auth/password_DC1.txt <<EOF
admin123
EOF
cd ..

kubectl -n aerospike create configmap xdr-proxy-cfg --from-file=etc/
kubectl -n aerospike create secret generic xdr-proxy-auth-secret --from-file=etc/auth

Deployment

Now that we have our xdr-proxy config file created we can now produce the kubernetes deployment yaml file for the xdr-proxy itself. The following yaml file is used to deploy our xdr-proxy ideally in the same data centre or location where the destination databases will be hosted.

cat <<EOF> xdr-proxy-deployment.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
  name: xdr-proxy
  namespace: aerospike
  labels:
    app: xdr-proxy
spec:
  replicas: 2
  selector:
    matchLabels:
      app: xdr-proxy
  template:
    metadata:
      labels:
        app: xdr-proxy
    spec:
      containers:
      - name: xdr-proxy
        image: aerospike/aerospike-xdr-proxy:2.1.0
        volumeMounts:
        - name: xdr-proxy-dir
          mountPath: "/etc/aerospike-xdr-proxy/"
          readOnly: true
        - name: xdr-auth-dir
          mountPath: "/etc/aerospike-xdr-proxy/auth"
          readOnly: true
        ports:
          - name: xdr-proxy-main
            containerPort: 8901
          - name: xdr-proxy-mng
            containerPort: 8902
      volumes:
      - name: xdr-proxy-dir
        configMap:
          name: xdr-proxy-cfg
          optional: false
      - name: xdr-auth-dir
        secret:
          secretName: xdr-proxy-auth-secret
          optional: false
---
apiVersion: v1
kind: Service
metadata:
  name: xdr-proxy
  namespace: aerospike
spec:
  selector:
    app: xdr-proxy
  ports:
  - name: main
    protocol: TCP
    port: 8901
    targetPort: xdr-proxy-main
  - name: manage
    protocol: TCP
    port: 8902
    targetPort: xdr-proxy-mng
EOF

kubectl create -f xdr-proxy-deployment.yaml
kubectl get po -n aerospike -w

The following shows the current scheduled pods. So far, so good.

kubectl get po -n aerospike -w
NAME                         READY   STATUS    RESTARTS   AGE
aerocluster-dest-xdr-0-0     1/1     Running   0          77m
xdr-proxy-7d9fccd6c8-g5mjt   1/1     Running   0          2m26s
xdr-proxy-7d9fccd6c8-mjxp4   1/1     Running   0          2m26s

Source Cluster

Create the source Aerospike cluster using the following configuration

cd ../aerospike-kubernetes-operator/

cat <<EOF> ssd1_xdr_src_6.1_cluster_cr.yaml
apiVersion: asdb.aerospike.com/v1beta1
kind: AerospikeCluster
metadata:
  name: aerocluster-source-xdr
  namespace: aerospike

spec:
  size: 1
  image: aerospike/aerospike-server-enterprise:6.1.0.2

  storage:
    filesystemVolumePolicy:
      initMethod: deleteFiles
      cascadeDelete: true
    blockVolumePolicy:
      cascadeDelete: true
    volumes:
      - name: workdir
        aerospike:
          path: /opt/aerospike
        source:
          persistentVolume:
            storageClass: ssd
            volumeMode: Filesystem
            size: 1Gi
      - name: ns
        aerospike:
          path: /opt/aerospike/data/
        source:
          persistentVolume:
            storageClass: ssd
            volumeMode: Filesystem
            size: 1Gi
      - name: aerospike-config-secret
        source:
          secret:
            secretName: aerospike-secret
        aerospike:
          path: /etc/aerospike/secret

  podSpec:
    multiPodPerHost: true

  aerospikeAccessControl:
    roles:
      - name: writer
        privileges:
        - read-write
      - name: reader
        privileges:
        - read
    users:
      - name: admin
        secretName: auth-secret
        roles:
          - sys-admin
          - user-admin
          - read-write
      
  aerospikeConfig:
    service:
      feature-key-file: /etc/aerospike/secret/features.conf
    security: {}
    network:
      service:
        port: 3000
      fabric:
        port: 3001
      heartbeat:
        port: 3002
    xdr:
      dcs:
        - name: DC2
          connector: true
          node-address-ports:
            - xdr-proxy.aerospike.svc.cluster.local 8901
          namespaces:
            - name: test
    namespaces:
      - name: test
        memory-size: 134217728
        replication-factor: 1
        storage-engine:
          type: device
          files:
            - /opt/aerospike/data/test.dat
          filesize: 1073741824
          data-in-memory: true
EOF

kubectl create -f ssd1_xdr_src_6.1_cluster_cr.yaml
kubectl get po -n aerospike -w

From the source cluster, confirm the xdr component has made a connection to the xdr-proxy by filtering the Kubernetes log file as shown in the following kubectl command.

kubectl -n aerospike logs aerocluster-source-xdr-0-0 -c aerospike-server | grep xdr | grep conn
Dec 08 2022 13:49:21 GMT: INFO (xdr): (dc.c:581) DC DC2 connectedOct 10 2022 13:57:53 GMT: INFO (xdr): (dc.c:581) DC DC2 connected

Simple Message Test

Add some data to the source database and confirm it is being received in the destination cluster. Start by getting the source database service address and connect using the Aerospike’s command line tool called aql.

kubectl get svc -n aerospike
NAME                         TYPE        CLUSTER-IP       EXTERNAL-IP   PORT(S)             AGE
aerocluster-dest-xdr         ClusterIP   None             <none>        3000/TCP            3h38m
aerocluster-dest-xdr-0-0     NodePort    10.100.226.179   <none>        3000:30168/TCP      3h38m
aerocluster-source-xdr       ClusterIP   None             <none>        3000/TCP            33m
aerocluster-source-xdr-0-0   NodePort    10.100.116.173   <none>        3000:31999/TCP      33m
xdr-proxy                    ClusterIP   10.100.44.96     <none>        8901/TCP,8902/TCP   41mkubectl run -it --rm --restart=Never aerospike-tool -n aerospike --image=aerospike/aerospike-tools:latest -- aql -U admin -P admin123 -h aerocluster-source-xdr-0-0
Insert a source record using the following command in aql
insert into test.a1 (PK,a,b,c,d) values(1,"A","B","C","D")
OK, 1 record affected.

aql> select * from test
+----+-----+-----+-----+-----+
| PK | a   | b   | c   | d   |
+----+-----+-----+-----+-----+
| 1  | "A" | "B" | "C" | "D" |
+----+-----+-----+-----+-----+
1 row in set (0.023 secs)

Connect to the destination cluster in the same way and confirm data has successfully arrived.

kubectl run -it --rm --restart=Never aerospike-tool -n aerospike --image=aerospike/aerospike-tools:latest -- aql -U admin -P admin123 -h aerocluster-dest-xdr-0-0
Run the following select query in the destination cluster.
aql> select * from test
+----+-----+-----+-----+-----+
| PK | a   | b   | c   | d   |
+----+-----+-----+-----+-----+
| 1  | "A" | "B" | "C" | "D" |
+----+-----+-----+-----+-----+
1 row in set (0.031 secs)

OK

Interim summary

So far, at this point it’s confirmed that xdr-proxy is doing exactly what we need it to do.

If you now review the logs for the initial 2 xdr-proxies that were scheduled you should indeed see userKey=1.

kubectl logs xdr-proxy-7d9fccd6c8-5q7gn -n aerospike | grep record-parser
2022-12-08 14:53:50.607 GMT DEBUG record-parser - parsed message fields: key=Key(namespace='test', set=a1, digest=[120, 48, -23, -90, 110, 126, 84, -1, 114, -116, -9, -21, 28, 75, 126, -68, -51, 83, 31, -117], userKey=1, lastUpdateTimeMs=1670511230565, userKeyString=1, digestString='eDDppm5+VP9yjPfrHEt+vM1TH4s=')

kubectl logs xdr-proxy-7d9fccd6c8-f5zkt -n aerospike | grep record-parser
(none)

Scaling the XDR-Proxies

Go ahead and scale up the xdr-proxy to 6 pods by editing the file xdr-proxy-deployment.yaml and then applying the changes.

apiVersion: apps/v1
kind: Deployment
metadata:
  name: xdr-proxy
  namespace: aerospike
  labels:
    app: xdr-proxy
spec:
  replicas: 6
  selector:
    matchLabels:
      app: xdr-proxy
...
kubectl apply -f xdr-proxy-deployment.yaml

You can also achieve the same by running the following command

kubectl scale deploy xdr-proxy -n aerospike --replicas=6

You should now have 6 instances of the xdr-proxies running.

kubectl get po -n aerospike -w
NAME                         READY   STATUS    RESTARTS   AGE
aerocluster-dest-xdr-0-0     1/1     Running   0          3h50m
aerocluster-source-xdr-0-0   1/1     Running   0          75m
xdr-proxy-7d9fccd6c8-49ttl   1/1     Running   0          7s
xdr-proxy-7d9fccd6c8-5q7gn   1/1     Running   0          83m
xdr-proxy-7d9fccd6c8-c4j7k   1/1     Running   0          7s
xdr-proxy-7d9fccd6c8-f5zkt   1/1     Running   0          83m
xdr-proxy-7d9fccd6c8-lscbg   1/1     Running   0          7s
xdr-proxy-7d9fccd6c8-r56vs   1/1     Running   0          7s

Lets add some sample messages from our source xdr cluster with primary keys 5,6,7

aql> insert into test (PK,a,b) values (5,"A","B")
OK, 1 record affected.
aql> insert into test (PK,a,b) values (6,"A","B")
OK, 1 record affected.
aql> insert into test (PK,a,b) values (7,"A","B")
OK, 1 record affected.

Notice how we have userKey=5, userKey=6 and userKey=7 across the newly scaled xdr-proxies.

kubectl logs xdr-proxy-7d9fccd6c8-5q7gn -n aerospike | grep record-parser
2022-12-08 14:53:50.607 GMT DEBUG record-parser - parsed message fields: key=Key(namespace='test', set=a1, digest=[120, 48, -23, -90, 110, 126, 84, -1, 114, -116, -9, -21, 28, 75, 126, -68, -51, 83, 31, -117], userKey=1, lastUpdateTimeMs=1670511230565, userKeyString=1, digestString='eDDppm5+VP9yjPfrHEt+vM1TH4s=')

kubectl logs xdr-proxy-7d9fccd6c8-f5zkt -n aerospike | grep record-parser
(none)

kubectl logs xdr-proxy-7d9fccd6c8-49ttl -n aerospike | grep record-parser
(none)

kubectl logs xdr-proxy-7d9fccd6c8-c4j7k -n aerospike | grep record-parser
2022-12-08 15:05:37.511 GMT DEBUG record-parser - parsed message fields: key=Key(namespace='test', set=null, digest=[-88, 104, 104, -114, 19, -44, -19, 29, -15, 18, 118, 72, -117, -106, -28, 21, -48, 50, 26, 113], userKey=7, lastUpdateTimeMs=1670511937250, userKeyString=7, digestString='qGhojhPU7R3xEnZIi5bkFdAyGnE=')

kubectl logs xdr-proxy-7d9fccd6c8-lscbg -n aerospike | grep record-parser
2022-12-08 15:05:27.548 GMT DEBUG record-parser - parsed message fields: key=Key(namespace='test', set=null, digest=[68, 4, -94, -44, -75, 112, -102, 73, -120, 41, -101, -120, 33, -111, 15, -114, -85, 46, -2, 80], userKey=5, lastUpdateTimeMs=1670511927465, userKeyString=5, digestString='RASi1LVwmkmIKZuIIZEPjqsu/lA=')
2022-12-08 15:05:32.300 GMT DEBUG record-parser - parsed message fields: key=Key(namespace='test', set=null, digest=[33, -100, 127, 120, 17, 45, -79, 115, -40, 53, -70, -57, 120, 73, 20, -50, -99, -98, -104, 85], userKey=6, lastUpdateTimeMs=1670511932288, userKeyString=6, digestString='IZx/eBEtsXPYNbrHeEkUzp2emFU=')

kubectl logs xdr-proxy-7d9fccd6c8-r56vs -n aerospike | grep record-parser
(none)

When data is actively flowing between source and destination clusters, the existing cached list of xdr-proxy connections from the source cluster is not refreshed, so the newly scaled xdr-proxies we have just scheduled will not be utilized immediately.

To demonstrate, let’s add data to the source cluster using Aerospike’s benchmark tool. At the same time, we will scale the xdr-proxies on the destination side and observe the results. Before we start, we reduce the xdr-proxy server count to 1 to make the observations clear.

I use my own local machine which has the benchmark tool already installed to send data to the source EC2 instances. In order to do this, we need to get the external IP address of the service.

kubectl get AerospikeCluster aerocluster-source-xdr -n aerospike  -o yaml

...
pods:
    aerocluster-source-xdr-0-0:
      aerospike:
        accessEndpoints:
        - 192.168.41.63:31999
        alternateAccessEndpoints:
        - 54.173.138.131:31999
        clusterName: aerocluster-source-xdr
        nodeID: 0a0
        tlsAccessEndpoints: []
        tlsAlternateAccessEndpoints: []
        tlsName: ""
      aerospikeConfigHash: 4aacb9809beaa01d99a9f00293c9f7dc141845f8
      hostExternalIP: 54.173.138.131
      hostInternalIP: 192.168.41.63
      image: aerospike/aerospike-server-enterprise:6.1.0.2
      initializedVolumes:
      - workdir
      - ns
      networkPolicyHash: acbbfab3668e1fceeed201139d1173f00095667e
      podIP: 192.168.50.203
      podPort: 3000
      podSpecHash: 972dc2a779fe9ab407212b547d54d3a72ecef259
      servicePort: 31999
...

Add the AWS firewall rule to allow traffic into the Kubernetes service. Connect the asbenchmark tool to start writing traffic using the public IP address for the NodePort Service.

asbenchmark -h 54.173.138.131:31999 -Uadmin -Padmin123 -z 10 -servicesAlternate -w RU,0 -o B256

Scale up the xdr-proxy servers from 1 to 2, and check the logs of both proxies to see what messages they received. In a production environment, you should disable the logging.

Notice that no data has passed through the new xdr-proxy instance xdr-proxy-7d9fccd6c8-s2tzt.

Conclusion

Dependable resiliency is a core part of the Aerospike value proposition. Aerospike’s XDR feature ensures that users can mitigate the risk of a cluster becoming unavailable by replicating data asynchronously from one data center to another. We show here how you can achieve this in a Kubernetes environment by using the xdr-proxy, allowing you to avoid network complications. All this has been achieved with minimum effort thanks to the Aerospike Kubernetes Operator.

Aerospike
Kubernetes
Xdr
Xdr Proxy
K8s

--

--

Naresh Maharaj
Aerospike Developer Blog

Written by Naresh Maharaj

12 Followers
Writer for

Help

Status

About

Careers

Press

Blog

Privacy

Terms

Text to speech

Teams