Skip to content

Latest commit

 

History

History
73 lines (55 loc) · 2.65 KB

kubeclusters.md

File metadata and controls

73 lines (55 loc) · 2.65 KB

Access Distinct Clusters

ServiceCenter Aggregate Architecture

In the Multiple Datacenters article, we introduce the aggregation architecture of service center. In fact, this aggregation architecture of service center can be applied not only to the scene deployed in multiple datacenters, but also to the scene services data aggregation in multiple kubernetes clusters.

architecture

The service centers deployed in distinct kubernetes clusters can communicate with each other, sync the services data from other kubernetes clusters. Applications can discover services from different the kubernetes cluster through using the service center HTTP API. It solve the problem of isolation between kubernetes clusters.

Quick Start

Let's assume you want to install 2 clusters of Service-Center in different Kubernetes clusters with following details.

Cluster Kubernetes namespace Node
sc1 k1 default 10.12.0.1
sc2 k2 default 10.12.0.2

To facilitate deployment, we will publish the service address of the service center in [NodePort] mode.

Deploy the Service Center

Using helm to deploy the service center to kubernetes here, the instructions for specific values can be referred to here.

Take deployment to kubernetes cluster 1 as an example.

# login the k1 kubernetes master node to deploy sc1
git clone [email protected]:apache/servicecomb-service-center.git
cd examples/infrastructures/k8s
helm install --name k1 \
    --set sc.discovery.clusters="sc2=http://10.12.0.2:30100" \
    --set sc.discovery.aggregate="k8s\,servicecenter" \
    --set sc.registry.type="buildin" \
    --set sc.service.type=NodePort \
    service-center/

Notes: To deploy Service Center in kuberbetes cluster 2, you can repeat the above steps and just change the sc.discovery.clusters value to sc1=http://10.12.0.1:30100.

Confirm the service is OK

We recommend that you use scctl, and using cluster command which makes it very convenient to verify OK.

# check the sc1 cluster api
scctl --addr http://10.12.0.3:30100 get cluster
#   CLUSTER |        ENDPOINTS         
# +---------+-------------------------+
#   sc2     | http://10.12.0.2:30100

# check the sc2 cluster api
scctl --addr http://10.12.0.3:30100 get cluster
#   CLUSTER |        ENDPOINTS         
# +---------+-------------------------+
#   sc1     | http://10.12.0.1:30100