Cluster Management

In many production environments, the backend needs to support distributed aggregation, high throughput and provide high availability (HA) to maintain robustness, so you always need to setup CLUSTER management in product env. Otherwise, you would face metrics inaccurate.

core/gRPCHost is listening on 0.0.0.0 for quick start as the single mode for most cases. Besides the Kubernetes coordinator, which is using the cloud-native mode to establish cluster, all other coordinators requires core/gRPCHost updated to real IP addresses or take reference of internalComHost and internalComPort in each coordinator doc.

NOTICE, cluster management doesn’t provide a service discovery mechanism for agents and probes. We recommend agents/probes using gateway to load balancer to access OAP clusters.

There are various ways to manage the cluster in the backend. Choose the one that best suits your needs.

  • Kubernetes. When the backend clusters are deployed inside Kubernetes, you could make use of this method by using k8s native APIs to manage clusters.
  • Zookeeper coordinator. Use Zookeeper to let the backend instances detect and communicate with each other.
  • Consul. Use Consul as the backend cluster management implementor and coordinate backend instances.
  • Etcd. Use Etcd to coordinate backend instances.
  • Nacos. Use Nacos to coordinate backend instances.

In the application.yml file, there are default configurations for the aforementioned coordinators under the section cluster. You can specify any of them in the selector property to enable it.

Cloud Native

Kubernetes

The required backend clusters are deployed inside Kubernetes. See the guides in Deploy in kubernetes. Set the selector to kubernetes.

cluster:
  selector: ${SW_CLUSTER:kubernetes}
  # other configurations

Meanwhile, the OAP cluster requires the pod’s UID which is laid at metadata.uid as the value of the system environment variable SKYWALKING_COLLECTOR_UID

containers:
  # Original configurations of OAP container
  - name: {{ .Values.oap.name }}
    image: {{ .Values.oap.image.repository }}:{{ required "oap.image.tag is required" .Values.oap.image.tag }}
    # ...
    # ...
    env:
    # Add metadata.uid as the system environment variable, SKYWALKING_COLLECTOR_UID 
    - name: SKYWALKING_COLLECTOR_UID
      valueFrom:
        fieldRef:
          fieldPath: metadata.uid

Read the complete helm for more details.

Traditional Coordinator

NOTICE In all the following coordinators, oap.internal.comm.host:oap.internal.comm.port is registered as the ID and address for the current OAP node. By default, because they are same in all OAP nodes, the registrations are conflicted, and (may) show as one registered node, which actually would be the node itself. In this case, the cluster mode is NOT working.

Please check the registered nodes on your coordinator servers, to make the registration information unique for every node. You could have two options

  1. Change core/gRPCHost(oap.internal.comm.host) and core/gRPCPort(oap.internal.comm.port) for internal, and setup external communication channels for data reporting and query.
  2. Use internalComHost and internalComPort in the config to provide a unique host and port for every OAP node. This host name port should be accessible for other OAP nodes.

Zookeeper coordinator

Zookeeper is a very common and widely used cluster coordinator. Set the cluster/selector to zookeeper in the yml to enable it.

Required Zookeeper version: 3.5+

cluster:
  selector: ${SW_CLUSTER:zookeeper}
  # other configurations
  • hostPort is the list of zookeeper servers. Format is IP1:PORT1,IP2:PORT2,...,IPn:PORTn
  • enableACL enable Zookeeper ACL to control access to its znode.
  • schema is Zookeeper ACL schemas.
  • expression is a expression of ACL. The format of the expression is specific to the schema.
  • hostPort, baseSleepTimeMs and maxRetries are settings of Zookeeper curator client.

Note:

  • If Zookeeper ACL is enabled and /skywalking exists, you must ensure that SkyWalking has CREATE, READ and WRITE permissions. If /skywalking does not exist, it will be created by SkyWalking, and all permissions to the specified user will be granted. Simultaneously, znode grants READ permission to anyone.
  • If you set schema as digest, the password of the expression is set in clear text.

In some cases, the OAP default gRPC host and port in the core are not suitable for internal communication among the OAP nodes, such as the default host(0.0.0.0) should not be used in cluster mode. The following settings are provided to set the host and port manually, based on your own LAN env.

  • internalComHost: The exposed host name for other OAP nodes in the cluster internal communication.
  • internalComPort: the exposed port for other OAP nodes in the cluster internal communication.
zookeeper:
  namespace: ${SW_NAMESPACE:""}
  hostPort: ${SW_CLUSTER_ZK_HOST_PORT:localhost:2181}
  #Retry Policy
  baseSleepTimeMs: ${SW_CLUSTER_ZK_SLEEP_TIME:1000} # initial amount of time to wait between retries
  maxRetries: ${SW_CLUSTER_ZK_MAX_RETRIES:3} # max number of times to retry
  internalComHost: ${SW_CLUSTER_INTERNAL_COM_HOST:172.10.4.10}
  internalComPort: ${SW_CLUSTER_INTERNAL_COM_PORT:11800}
  # Enable ACL
  enableACL: ${SW_ZK_ENABLE_ACL:false} # disable ACL in default
  schema: ${SW_ZK_SCHEMA:digest} # only support digest schema
  expression: ${SW_ZK_EXPRESSION:skywalking:skywalking}

Consul

Recently, the Consul system has become more and more popular, and many companies and developers now use Consul as their service discovery solution. Set the cluster/selector to consul in the yml to enable it.

cluster:
  selector: ${SW_CLUSTER:consul}
  # other configurations

Same as the Zookeeper coordinator, in some cases, the OAP default gRPC host and port in the core are not suitable for internal communication among the OAP nodes, such as the default host(0.0.0.0) should not be used in cluster mode. The following settings are provided to set the host and port manually, based on your own LAN env.

  • internalComHost: The exposed host name for other OAP nodes in the cluster internal communication.
  • internalComPort: the exposed port for other OAP nodes in the cluster internal communication.

Etcd

Set the cluster/selector to etcd in the yml to enable it. The Etcd client has upgraded to v3 protocol and changed to the CoreOS official library. Since 8.7.0, only the v3 protocol is supported for Etcd.

cluster:
  selector: ${SW_CLUSTER:etcd}
  # other configurations
  etcd:
    # etcd cluster nodes, example: 10.0.0.1:2379,10.0.0.2:2379,10.0.0.3:2379
    endpoints: ${SW_CLUSTER_ETCD_ENDPOINTS:localhost:2379}
    namespace: ${SW_CLUSTER_ETCD_NAMESPACE:/skywalking}
    serviceName: ${SW_CLUSTER_ETCD_SERVICE_NAME:"SkyWalking_OAP_Cluster"}
    authentication: ${SW_CLUSTER_ETCD_AUTHENTICATION:false}
    user: ${SW_CLUSTER_ETCD_USER:}
    password: ${SW_CLUSTER_ETCD_PASSWORD:}

Same as the Zookeeper coordinator, in some cases, the OAP default gRPC host and port in the core are not suitable for internal communication among the OAP nodes, such as the default host(0.0.0.0) should not be used in cluster mode. The following settings are provided to set the host and port manually, based on your own LAN env.

  • internalComHost: The exposed host name for other OAP nodes in the cluster internal communication.
  • internalComPort: the exposed port for other OAP nodes in the cluster internal communication.

Nacos

Set the cluster/selector to nacos in the yml to enable it.

cluster:
  selector: ${SW_CLUSTER:nacos}
  # other configurations

Nacos supports authentication by username or accessKey. Empty means that there is no need for authentication. Extra config is as follows:

nacos:
  username:
  password:
  accessKey:
  secretKey:

Same as the Zookeeper coordinator, in some cases, the OAP default gRPC host and port in the core are not suitable for internal communication among the OAP nodes, such as the default host(0.0.0.0) should not be used in cluster mode. The following settings are provided to set the host and port manually, based on your own LAN env.

  • internalComHost: The exposed host name for other OAP nodes in the cluster internal communication.
  • internalComPort: the exposed port for other OAP nodes in the cluster internal communication.