Deploy DM on Kubernetes
TiDB Data Migration (DM) is an integrated data migration task management platform that supports the full data migration and the incremental data replication from MySQL/MariaDB into TiDB. This document describes how to deploy DM on Kubernetes using TiDB Operator and how to migrate MySQL data to TiDB cluster using DM.
Prerequisites
- Complete deploying TiDB Operator.
Configure DM deployment
To configure the DM deployment, you need to configure the DMCluster
Custom Resource (CR). For the complete configurations of the DMCluster
CR, refer to the DMCluster example and API documentation. Note that you need to choose the example and API of the current TiDB Operator version.
Cluster name
Configure the cluster name by changing the metadata.name
in the DMCluster
CR.
Version
Usually, components in a cluster are in the same version. It is recommended to configure only spec.<master/worker>.baseImage
and spec.version
. If you need to deploy different versions for different components, configure spec.<master/worker>.version
.
The formats of the related parameters are as follows:
spec.version
: the format isimageTag
, such asv8.1.0
.spec.<master/worker>.baseImage
: the format isimageName
, such aspingcap/dm
.spec.<master/worker>.version
: the format isimageTag
, such asv8.1.0
.
TiDB Operator only supports deploying DM 2.0 and later versions.
Cluster
Configure DM-master
DM-master is an indispensable component of the DM cluster. You need to deploy at least three DM-master Pods if you want to achieve high availability.
You can configure DM-master parameters by spec.master.config
in DMCluster
CR. For complete DM-master configuration parameters, refer to DM-master Configuration File.
apiVersion: pingcap.com/v1alpha1
kind: DMCluster
metadata:
name: ${dm_cluster_name}
namespace: ${namespace}
spec:
version: v8.1.0
configUpdateStrategy: RollingUpdate
pvReclaimPolicy: Retain
discovery: {}
master:
baseImage: pingcap/dm
maxFailoverCount: 0
imagePullPolicy: IfNotPresent
service:
type: NodePort
# Configures masterNodePort when you need to expose the DM-master service to a fixed NodePort
# masterNodePort: 30020
replicas: 1
storageSize: "10Gi"
requests:
cpu: 1
config: |
rpc-timeout = "40s"
Configure DM-worker
You can configure DM-worker parameters by spec.worker.config
in DMCluster
CR. For complete DM-worker configuration parameters,refer to DM-worker Configuration File.
apiVersion: pingcap.com/v1alpha1
kind: DMCluster
metadata:
name: ${dm_cluster_name}
namespace: ${namespace}
spec:
...
worker:
baseImage: pingcap/dm
maxFailoverCount: 0
replicas: 1
storageSize: "100Gi"
requests:
cpu: 1
config: |
keepalive-ttl = 15
Topology Spread Constraint
By configuring topologySpreadConstraints
, you can make pods evenly spread in different topologies. For instructions about configuring topologySpreadConstraints
, see Pod Topology Spread Constraints.
You can either configure topologySpreadConstraints
at a cluster level (spec.topologySpreadConstraints
) for all components or at a component level (such as spec.tidb.topologySpreadConstraints
) for specific components.
The following is an example configuration:
topologySpreadConstrains:
- topologyKey: kubernetes.io/hostname
- topologyKey: topology.kubernetes.io/zone
The example configuration can make pods of the same component evenly spread on different zones and nodes.
Currently, topologySpreadConstraints
only supports the configuration of the topologyKey
field. In the pod spec, the above example configuration will be automatically expanded as follows:
topologySpreadConstrains:
- topologyKey: kubernetes.io/hostname
maxSkew: 1
whenUnsatisfiable: DoNotSchedule
labelSelector: <object>
- topologyKey: topology.kubernetes.io/zone
maxSkew: 1
whenUnsatisfiable: DoNotSchedule
labelSelector: <object>
Deploy the DM cluster
After configuring the yaml file of the DM cluster in the above steps, execute the following command to deploy the DM cluster:
kubectl apply -f ${dm_cluster_name}.yaml -n ${namespace}
If the server does not have an external network, you need to download the Docker image used by the DM cluster and upload the image to the server, and then execute docker load
to install the Docker image on the server:
Deploy a DM cluster requires the following Docker image (assuming the version of the DM cluster is v8.1.0):
pingcap/dm:v8.1.0To download the image, execute the following command:
docker pull pingcap/dm:v8.1.0 docker save -o dm-v8.1.0.tar pingcap/dm:v8.1.0Upload the Docker image to the server, and execute
docker load
to install the image on the server:docker load -i dm-v8.1.0.tar
After deploying the DM cluster, execute the following command to view the Pod status:
kubectl get po -n ${namespace} -l app.kubernetes.io/instance=${dm_cluster_name}
You can use TiDB Operator to deploy and manage multiple DM clusters in a single Kubernetes cluster by repeating the above procedure and replacing ${dm_cluster_name}
with a different name.
Different clusters can be in the same or different namespace
, which is based on your actual needs.
Access the DM cluster on Kubernetes
To access DM-master in the pod within a Kubernetes cluster, use the DM-master service domain name ${cluster_name}-dm-master.${namespace}
.
To access the DM cluster outside a Kubernetes cluster, expose the DM-master port by editing the spec.master.service
field configuration in the DMCluster
CR.
spec:
...
master:
service:
type: NodePort
You can access the DM-master service via the address of ${kubernetes_node_ip}:${node_port}
.
For more service exposure methods, refer to Access the TiDB Cluster.
What’s next
- To migrate MySQL data to your TiDB cluster using DM on Kubernetes, see Migrate MySQL Data to TiDB Cluster Using DM.
- To enable TLS between components of the DM cluster on Kubernetes, see Enable TLS for DM.