VMDistributed is the Custom Resource Definition for orchestrated updates of several VictoriaMetrics clusters. It allows you to define and manage cluster components of a distributed VictoriaMetrics setup and apply changes to them sequentially, ensuring high availability and minimal disruption.
Note: VMDistributed is an experimental feature and may not be suitable for production environments. API is not yet stabilized and may change in future releases.
For a high-level overview of VictoriaMetrics distributed cluster architecture, refer to the official VictoriaMetrics documentation .
Specification #
The VMDistributed resource allows you to configure various aspects of your VictoriaMetrics distributed cluster. The specification includes:
retain: If set totrue, the operator will keep managed components during VMDistributed CR removal.paused: If set totrue, the operator will not perform any actions on the underlying managed objects. Useful for temporarily halting reconciliation.vmauth: Configuration for aVMAuthinstance that acts as a proxy/load-balancer for thevmselectandvmagentcomponents of the managed clusters. It acts as a global read and write path entry point.zoneCommon: DefinesVMDistributedZoneCommoncommon zone configuration, which is default for each zone resource.zones: Defines the list ofVMDistributedZonethat form the distributed setup.license: Configures the license key for enterprise features. If provided, it is automatically passed to the managedVMAuth, allVMAgent, andVMClusterinstances.
VMDistributedZoneCommon
#
readyTimeout: The readiness timeout for each zone update. Default is5m.updatePause: Time the operator should wait between zone updates to ensure a smooth transition. Default is1m.
VMDistributedZone
#
Each entry in the zones array includes:
namedefines name of zone, which is used as for default per zone resource name or can be used inzoneCommontemplate using%ZONE%placeholder.vmclusterdefinesVMDistributedZoneCluster, that allows either reference an existingVMClusterresource or creates a new one.vmagentdefinesVMDistributedZoneAgent, that allows either reference an existingVMAgentresource or creates a one. It receives write traffic fromvmauthand sends tovmclusterinstances.
VMDistributedZoneCluster
#
name: Optional name of an existingVMClusteror one to be created; when omitted it defaults tozoneCommon.vmcluster.name(where%ZONE%is used to template the zone name) or${cr.Name}-${zone.name}when unset.spec: Avmv1beta1.VMClusterSpecobject that defines the desired state of a new or referencedVMClustermanaged by this resource.
Example: Defining a VMDistributed resource:
apiVersion: operator.victoriametrics.com/v1alpha1
kind: VMDistributed
metadata:
name: my-distributed-cluster
spec:
vmauth:
name: my-distributed-vmauth
spec:
replicaCount: 1
unauthorizedUserAccessSpec:
targetRefs:
- name: read
- name: write
zoneCommon:
vmagent:
spec:
replicaCount: 1
vmcluster:
spec:
vmstorage:
replicaCount: 1
storage:
volumeClaimTemplate:
spec:
resources:
requests:
storage: 10Gi
vmselect:
replicaCount: 1
vminsert:
replicaCount: 1
zones:
- name: zone-a
- name: zone-b
VMDistributed and distributed chart #
VMDistributed can be used alongside the resources created by the distributed chart. The distributed chart provides a convenient way to create multiple VMCluster objects and surrounding resources.
In order to update VMClusters in a coordinated manner, add VMCluster resources to the zones list:
spec:
vmauth:
spec:
replicaCount: 1
unauthorizedUserAccessSpec:
targetRefs:
- name: read
- name: write
zoneCommon:
vmagent:
name: vmagent-%ZONE%
spec:
replicaCount: 1
vmcluster:
name: vmcluster-%ZONE%
spec:
replicationFactor: 1
requestsLoadBalancer:
enabled: true
spec:
nodeSelector:
kubernetes.io/os: linux
replicaCount: 2
topologySpreadConstraints:
- maxSkew: 1
topologyKey: kubernetes.io/hostname
whenUnsatisfiable: ScheduleAnyway
retentionPeriod: "14"
vminsert:
nodeSelector:
kubernetes.io/os: linux
port: "8480"
replicaCount: 2
topologySpreadConstraints:
- maxSkew: 1
topologyKey: kubernetes.io/hostname
whenUnsatisfiable: ScheduleAnyway
vmselect:
nodeSelector:
kubernetes.io/os: linux
port: "8481"
replicaCount: 2
topologySpreadConstraints:
- maxSkew: 1
topologyKey: kubernetes.io/hostname
whenUnsatisfiable: ScheduleAnyway
vmstorage:
nodeSelector:
kubernetes.io/os: linux
replicaCount: 2
storageDataPath: /vm-data
topologySpreadConstraints:
- maxSkew: 1
topologyKey: kubernetes.io/hostname
whenUnsatisfiable: ScheduleAnyway
zones:
- name: central
- name: east
- name: west
Note that VMDistributed will create a single VMAuth, which encapsulates rules for both read and write access to the zones, unlike the distributed chart which creates separate VMAuths for read and write paths. This provides a convenient way to migrate the traffic between VMDistributed and distributed helm chart implementations.
Ownership and references #
VMDistributed creates or becomes an owner of existing VMAuth, VMAgent and VMCluster resources and by default it removes all resources it owns on CR deletion. To remove VMDistributed but keep all owned resources set spec.retain: true before VMDistributed removal.
Current shortcomings #
- Only
VMClusterobjects are supported for distributed management. - Only one
VMAuthcan be managed perVMDistributed. - All objects must belong to the same namespace as the
VMDistributed. - Objects must be referred to by name; label selectors are not supported for cluster selection.
Authorization #
By default VMDistributed provides no access to read/write endpoints. Instead it provides preconfigured named backends read and write, which can be referenced in spec.vmauth.spec.unauthorizedUserAccessSpec section for provide unauthorized access or in VMUser spec for authorized access.
To provide unauthorized access to the distributed setup need to add spec.vmauth.spec.unauthorizedUserAccessSpec section, where read, write backends should be referenced:
spec:
vmauth:
spec:
unauthorizedUserAccessSpec:
targetRefs:
- name: read
- name: write
With configuration above both read and write endpoints become available for unauthorized access
In order to configure authorization spec.vmauth.spec should have
user selectors
properly configured and VMUser CRs should reference read and write backend in targetRefs:
apiVersion: operator.victoriametrics.com/v1beta1
kind: VMUser
metadata:
name: example
spec:
username: simple-user
password: simple-password
targetRefs:
- name: write
- name: read