Saturday, July 11, 2020

Scaling Up NDB Cluster on Kubernetes


On my first article about NDB Cluster on Kubernetes, I discussed steps how to deploy NDB Cluster on Kubernetes. You can find the article here https://mysqlsg.blogspot.com/2020/06/deploying-ndb-cluster-on-kubernetes-in_30.html. The second article, I discussed steps how to clone NDB Cluster here https://mysqlsg.blogspot.com/2020/07/clone-ndb-cluster-on-kubernetes-using.html.

This article is the continuation of previous two articles discussing steps how to scaling up data nodes and SQL nodes when running on Kubernetes.

Content:

A.     Start the NDB Cluster
B.     Edit config.ini of the management node
C.     Spin new data nodes on Kubernetes
D.     Restart all cluster nodes
E.     Adding New data nodes
F.     Adding SQL nodes
G.     Test it!

 
A.   START THE NDB CLUSTER

Start the Minikube if this hasn’t been started. Let’s check the pod’s status:

$ kubectl -n mysql-cluster get pod
NAME                             READY   STATUS    RESTARTS   AGE
dataa-0                          1/1     Running   3          5d1h
datab-0                          1/1     Running   3          5d1h
mgmt-0                           1/1     Running   3          5d1h
mysql-cluster-796d8b4d78-dwn9j   1/1     Running   5          10d
mysql-cluster-796d8b4d78-n6n4l   1/1     Running   5          10d
$

Pod “mgmt-0” is the management node, “dataa-0” and “datab-0” are data nodes of data node group 0 where data is stored (see below for brief explanation), and “mysql-cluster-…” are SQL nodes.
Now let’s check our cluster, if it is automatically started up.

$ kubectl -n mysql-cluster exec -it mgmt-0 -- ndb_mgm -c localhost -e show
Connected to Management Server at: localhost:1186
Cluster Configuration
---------------------
[ndbd(NDB)]           2 node(s)
id=2          @172.17.0.5  (mysql-5.7.30 ndb-7.6.14, Nodegroup: 0, *)
id=3          @172.17.0.7  (mysql-5.7.30 ndb-7.6.14, Nodegroup: 0)

[ndb_mgmd(MGM)]              1 node(s)
id=1           @172.17.0.8  (mysql-5.7.30 ndb-7.6.14)

[mysqld(API)]          2 node(s)
id=4          @172.17.0.4  (mysql-5.7.30 ndb-7.6.14)
id=5          @172.17.0.6  (mysql-5.7.30 ndb-7.6.14)
$

Perfect! It is up and running automatically.

B.    Edit Config.ini of the Management Node

Open and edit config_ini.yaml into the following, see the update in red below for adding 2 more data nodes and 2 more SQL nodes:

[ndbd default]
NoOfReplicas=2
DataMemory=98M   

[ndb_mgmd]
NodeId=1
HostName=mgmt-0                              
DataDir=/var/lib/mysql               

[ndbd]
HostName=dataa-0                               
NodeId=2                                         
DataDir=/var/lib/mysql                 
ServerPort=2202

[ndbd]
HostName=datab-0                               
NodeId=3                                
DataDir=/var/lib/mysql                 
ServerPort=2202

[ndbd]
HostName=dataa-1                       
NodeId=4                                         
DataDir=/var/lib/mysql                 
ServerPort=2202

[ndbd]
HostName=datab-1                       
NodeId=5                           
DataDir=/var/lib/mysql                 
ServerPort=2202


[mysqld]

[mysqld]

[mysqld]

[mysqld]

Apply the updated config_ini.yaml into Kubernetes as ConfigMap with this command:

$ kubectl -n mysql-cluster delete configmap config-ini
configmap "config-ini" deleted

$ kubectl -n mysql-cluster create configmap config-ini --from-file=config_ini.yaml
configmap/config-ini created

It is easy to understand how to scaling up the SQL nodes by looking at the above YAML file. To understand how to scale up data nodes requires combination between the number of replicas, node groups as well as Kubernetes StatefulSet. Long story short, the replicas is a concept of storing all data into up to four copies across multiple data nodes to avoid having just a single copy of the data for high availability. Number of replicas is determined by the value of parameter “NoOfReplicas” as seen at the above config. A node group is a group of data nodes that holds the same data, so add an additional node group with the same number of data nodes to expand / scale up the cluster’s overall data volume capacity. That’s what we are going to do now, but on Kubernetes!

That explanation is actually very brief and insufficient. For more detail information about NDB Cluster nodes, node groups, and replicas; please visit https://dev.mysql.com/doc/mysql-cluster-excerpt/8.0/en/mysql-cluster-nodes-groups.html. I also encourage to visit a blog by Mikael Ronstrom here http://mikaelronstrom.blogspot.com/2020/01/support-3-4-replicas-in-ndb-cluster-80.html for the updates on 3-4 replicas support on NDB Cluster 8.0.

On Kubernetes, as I mentioned in previous blog article, I recommend to use StatefulSet replicas to expand / scale up and use number of StatefulSets to determine the number of data node within a data node group. Thus, if we want to have two data nodes in one data node group, then we need two StatefulSets. If we want to scale up from one data node group into two data node group, then we need to scale up parameter “replicas” on each StatefulSets from “1” to “2”. This is what we are going to cover in this article!

 


Because data nodes are deployed using StatefulSets, things become easy to handle. If we scale up StatefulSet replicas being “1” to “2”, there will be new Kubernetes Pod created with Pod ordering number equal to “1”, (e.g. dataa-1, and datab-1 as shown in the diagram). Thus, we will have dataa-0 and datab-0 from current cluster setup, plus additional dataa-1 and datab-1. In this case, dataa-0 and dataa-1 belong to StatefulSet “dataa”, while datab-0 and datab-1 belong to StatefulSet “datab”.  But from NDB cluster perspective, dataa-0 and datab-0 belong to data node group 0, while datab-0 and datab-1 belong to data node group 1. It is indeed reversed, but this design may give flexibility to scale.  

Back to the YAML file, as you see there are definition for data node “dataa-1” and “datab-1” in addition to the original YAML file.


C.   Spin New Data Nodes on Kubernetes

This is the easiest part of this article. We know we have two additional data nodes (“dataa-1” and datab-1”), therefore we need one new Kubernetes services for each Pods for these new data nodes. To spin up additional two Pods for two data nodes are even easier. Just edit parameter “replicas” on each data node’s StatefulSets (“dataa” and “datab”) in the YAML file from “1” to “2”. See below the edited YAML file which includes the necessary changes (written in bold). - please note: for simplification in this article, I don't put PV/PVC into Pods definition for data nodes.

---
apiVersion: v1
kind: Service
metadata:
  name: mgmt-0
  namespace: mysql-cluster
spec:
  ports:
  - name: mgmtport
    port: 1186
    targetPort: 1186
  - name: pingport
    port: 7
    targetPort: 7
  selector:
    statefulset.kubernetes.io/pod-name: mgmt-0
  clusterIP: None
---
apiVersion: v1
kind: Service
metadata:
  name: dataa-1
  namespace: mysql-cluster
spec:
  ports:
  - name: dataport
    port: 2202
    targetPort: 2202
  - name: pingport
    port: 7
    targetPort: 7
  selector:
    statefulset.kubernetes.io/pod-name: dataa-1
  clusterIP: None
---
apiVersion: v1
kind: Service
metadata:
  name: dataa-0
  namespace: mysql-cluster
spec:
  ports:
  - name: dataport
    port: 2202
    targetPort: 2202
  - name: pingport
    port: 7
    targetPort: 7
  selector:
    statefulset.kubernetes.io/pod-name: dataa-0
  clusterIP: None
---
apiVersion: v1
kind: Service
metadata:
  name: datab-1
  namespace: mysql-cluster
spec:
  ports:
  - name: dataport
    port: 2202
    targetPort: 2202
  - name: pingport
    port: 7
    targetPort: 7
  selector:
    statefulset.kubernetes.io/pod-name: datab-1
  clusterIP: None
---
apiVersion: v1
kind: Service
metadata:
  name: datab-0
  namespace: mysql-cluster
spec:
  ports:
  - name: dataport
    port: 2202
    targetPort: 2202
  - name: pingport
    port: 7
    targetPort: 7
  selector:
    statefulset.kubernetes.io/pod-name: datab-0
  clusterIP: None
---
apiVersion: v1
kind: Service
metadata:
  name: mysql-cluster
  namespace: mysql-cluster
  labels:
    app: mysql-cluster
spec:
  ports:
    - name: tcp-rw
      port: 3306
      targetPort: 3306
  selector:
    app: mysql-cluster
  type: LoadBalancer
---
apiVersion: apps/v1
kind: StatefulSet
metadata:
  name: mgmt
  namespace: mysql-cluster
spec:
  serviceName: "mgmt" 
  replicas: 1
  selector:
    matchLabels:
      app: mgmt
  template:
    metadata:
      labels:
        app: mgmt
    spec:
     containers:
     - image: mysql/mysql-cluster:latest
       name: mysql
       volumeMounts:
         - name: mysql-configmap-volume
           mountPath: /etc/config.ini
           subPath: config.ini
       command: ["/bin/sh"]                           # do not modify
       args: ["-c", "sleep 30; ndb_mgmd --config-file=/etc/config.ini --config-dir=/home; while true; do sleep 1; done;"] # do not modify 
     volumes:
      - name: mysql-configmap-volume
        configMap:
          name: config-ini
          items:
           - key: config_ini.yaml
             path: config.ini
---
apiVersion: apps/v1
kind: StatefulSet
metadata:
  name: dataa
  namespace: mysql-cluster
spec:
  serviceName: "dataa" 
  replicas: 2
  selector:
    matchLabels:
      app: dataa
  template:
    metadata:
      labels:
        app: dataa
    spec:
     containers:
     - image: mysql/mysql-cluster:latest
       name: mysql
       volumeMounts:
       - name: mysql-configmap-volume
         mountPath: /etc/datanode.cnf
         subPath: datanode.cnf
       command: ["/bin/sh", "-c", "sleep 60; ndbd --defaults-file=/etc/datanode.cnf --ndb-connectstring=mgmt-0; while true; do sleep 1; done;"]  
     volumes:
      - name: mysql-configmap-volume
        configMap:
          name: datanode
          items:
           - key: datanode_ini.yaml
             path: datanode.cnf
---
apiVersion: apps/v1
kind: StatefulSet
metadata:
  name: datab
  namespace: mysql-cluster
spec:
  serviceName: "datab" 
  replicas: 2
  selector:
    matchLabels:
      app: datab
  template:
    metadata:
      labels:
        app: datab
    spec:
     containers:
     - image: mysql/mysql-cluster:latest
       name: mysql
       volumeMounts:
       - name: mysql-configmap-volume
         mountPath: /etc/datanode.cnf
         subPath: datanode.cnf
       command: ["/bin/sh", "-c", "sleep 60; ndbd --defaults-file=/etc/datanode.cnf --ndb-connectstring=mgmt-0; while true; do sleep 1; done;"]  
     volumes:
      - name: mysql-configmap-volume
        configMap:
          name: datanode
          items:
           - key: datanode_ini.yaml
             path: datanode.cnf
---
apiVersion: apps/v1
kind: Deployment
metadata:
  name: mysql-cluster
  namespace: mysql-cluster
spec:
  replicas: 2
  selector:
    matchLabels:
      app: mysql-cluster
  strategy:
    type: Recreate
  template:
    metadata:
      labels:
        app: mysql-cluster
    spec:
     hostname: mysql-cluster
     containers:
     - image: mysql/mysql-cluster:latest
       name: mysql
       env:
         - name: MYSQL_ROOT_PASSWORD
           valueFrom:
             secretKeyRef:
                name: mysql-root-password
                key: password
       volumeMounts:
         - name: mysql-configmap-volume
           mountPath: /etc/my.cnf
           subPath: my.cnf
     volumes:
      - name: mysql-configmap-volume
        configMap:
          name: my-cnf
          items:
           - key: my_cnf.yaml
             path: my.cnf

Save that file as “all_nodes_scaling.yaml” and run the following command to apply the change into Kubernetes:

$ kubectl apply -f all_nodes_scaling.yaml

Check if “dataa-1” and “datab-1” Pods are running:

$ kubectl -n mysql-cluster get pod
NAME                             READY   STATUS    RESTARTS   AGE
dataa-0                          1/1     Running   3          5d10h
dataa-1                          1/1     Running   0          62s
datab-0                          1/1     Running   3          5d10h
datab-1                          1/1     Running   0          62s
mgmt-0                           1/1     Running   3          5d10h
mysql-cluster-796d8b4d78-dwn9j   1/1     Running   5          11d
mysql-cluster-796d8b4d78-n6n4l   1/1     Running   5          11d

Check if Kubernetes services for dataa-1 and datab-1 are available:

$ kubectl -n mysql-cluster get svc
NAME               TYPE           CLUSTER-IP       EXTERNAL-IP   PORT(S)          AGE
dataa-0            ClusterIP      None             <none>        2202/TCP,7/TCP   11d
dataa-1            ClusterIP      None             <none>        2202/TCP,7/TCP   2m40s
datab-0            ClusterIP      None             <none>        2202/TCP,7/TCP   11d
datab-1            ClusterIP      None             <none>        2202/TCP,7/TCP   2m40s
mgmt-0             ClusterIP      None             <none>        1186/TCP,7/TCP   11d
mysql-cluster      LoadBalancer   10.107.196.107   <pending>     3306:32418/TCP   11d

Cool, we are ready!

D.    Restart All Cluster Nodes

D.1. Restart Management Node “mgmt.-0”

Remember that we recreated config.ini at section B of this article. But the one which currently running on management node “mgmt-0” is the old config.ini which does not have “dataa-1” and “datab-1”.
Thus, we need to restart Pod “mgmt-0”, as follow:

$ kubectl -n mysql-cluster delete pod mgmt-0

Wait for 1 minute, and check cluster status, as follow:

$ kubectl -n mysql-cluster exec -it mgmt-0 -- ndb_mgm -c localhost -e show
Connected to Management Server at: localhost:1186
Cluster Configuration
---------------------
[ndbd(NDB)]           4 node(s)
id=2          @172.17.0.5  (mysql-5.7.30 ndb-7.6.14, Nodegroup: 0, *)
id=3          @172.17.0.7  (mysql-5.7.30 ndb-7.6.14, Nodegroup: 0)
id=4 (not connected, accepting connect from dataa-1)
id=5 (not connected, accepting connect from datab-1)

[ndb_mgmd(MGM)]              1 node(s)
id=1           @172.17.0.8  (mysql-5.7.30 ndb-7.6.14)

[mysqld(API)]          4 node(s)
id=6 (not connected, accepting connect from any host)
id=7 (not connected, accepting connect from any host)
id=8 (not connected, accepting connect from any host)
id=9 (not connected, accepting connect from any host)

As you see the above output, now the cluster consists of “dataa-1” and “datab-1” as well as 4 SQL Nodes.

D.2. Restart Data Node “dataa-0” and “datab-0”

Restart node id 2:
$ kubectl -n mysql-cluster exec -it mgmt-0 -- ndb_mgm -c localhost -e "2 restart;"

Once done, restart node id 3:
$ kubectl -n mysql-cluster exec -it mgmt-0 -- ndb_mgm -c localhost -e "3 restart;"

D.3. Restart All SQL Nodes

$ kubectl -n mysql-cluster delete pod mysql-cluster-796d8b4d78-dwn9j
$ kubectl -n mysql-cluster delete pod mysql-cluster-796d8b4d78-n6n4l


E.    Adding New Data Nodes

$ kubectl -n mysql-cluster exec -it dataa-1 -- ndbd --defaults-file=/etc/datanode.cnf --ndb-connectstring=mgmt-0
$ kubectl -n mysql-cluster exec -it datab-1 -- ndbd --defaults-file=/etc/datanode.cnf --ndb-connectstring=mgmt-0

Now check our cluster status:

$ kubectl -n mysql-cluster exec -it mgmt-0 -- ndb_mgm -c localhost -e show
Connected to Management Server at: localhost:1186
Cluster Configuration
---------------------
[ndbd(NDB)]           4 node(s)
id=2          @172.17.0.5  (mysql-5.7.30 ndb-7.6.14, Nodegroup: 0, *)
id=3          @172.17.0.7  (mysql-5.7.30 ndb-7.6.14, Nodegroup: 0)
id=4     @172.17.0.3  (mysql-5.7.30 ndb-7.6.14, Nodegroup: 1)
id=5     @172.17.0.4  (mysql-5.7.30 ndb-7.6.14, Nodegroup: 1)

[ndb_mgmd(MGM)]              1 node(s)
id=1           @172.17.0.8  (mysql-5.7.30 ndb-7.6.14)

[mysqld(API)]          4 node(s)
id=6          @172.17.0.19  (mysql-5.7.30 ndb-7.6.14)
id=7          @172.17.0.6  (mysql-5.7.30 ndb-7.6.14)
id=8 (not connected, accepting connect from any host)
id=9 (not connected, accepting connect from any host)

As you see, now our NDB Cluster already has 4 data nodes with 2 data node groups!

F.   Adding SQL Nodes

Simple! Just open all_nodes_scaling.yaml again, and edit the SQL Nodes section to change replicas=2 into replicas=4, as highlighted below.

---
apiVersion: apps/v1
kind: Deployment
metadata:
  name: mysql-cluster
  namespace: mysql-cluster
spec:
  replicas: 4
  selector:
    matchLabels:
      app: mysql-cluster

Apply the YAML file again.

$ kubectl apply -f all_nodes_scaling.yaml


G.    TEST IT !

Show cluster status:

$ kubectl -n mysql-cluster exec -it mgmt-0 -- ndb_mgm -c localhost -e show
Connected to Management Server at: localhost:1186
Cluster Configuration
---------------------
[ndbd(NDB)]           4 node(s)
id=2          @172.17.0.5  (mysql-5.7.30 ndb-7.6.14, Nodegroup: 0, *)
id=3          @172.17.0.7  (mysql-5.7.30 ndb-7.6.14, Nodegroup: 0)
id=4     @172.17.0.3  (mysql-5.7.30 ndb-7.6.14, Nodegroup: 1)
id=5     @172.17.0.4  (mysql-5.7.30 ndb-7.6.14, Nodegroup: 1)

[ndb_mgmd(MGM)]              1 node(s)
id=1           @172.17.0.8  (mysql-5.7.30 ndb-7.6.14)

[mysqld(API)]          4 node(s)
id=6          @172.17.0.19  (mysql-5.7.30 ndb-7.6.14)
id=7          @172.17.0.6  (mysql-5.7.30 ndb-7.6.14)
id=8     @172.17.0.20  (mysql-5.7.30 ndb-7.6.14)
id=9     @172.17.0.18  (mysql-5.7.30 ndb-7.6.14)

Done! The cluster now is running with 4 SQL Nodes !

H.    How about Cluster Manager

As you see in this article, scaling up NDB Cluster are quite manual without MySQL Cluster Manager (mcm). The more Data Nodes and SQL Nodes to handle, the task will get complicated. Using the Cluster Manager (mcm), DBA can use a lot of handy commands to simplify cluster management, including backup and recovery. See this URL for detail: https://dev.mysql.com/doc/mysql-cluster-manager/1.4/en/


Disclaimer:
The method and tricks presented here are experimental only and it’s your own responsibility to test, implement, and support in case of issues. It is only showing example and not for a production deployment, as this is not formally a supported configuration. I encourage more testing to be done, including by development team. Implementation with less layer as possible is recommended, and with having support from Oracle Support for real world implementation serving business application.

No comments:

Post a Comment