Storage
An ArangoDB cluster relies heavily on fast persistent storage. The ArangoDB Kubernetes Operator uses PersistentVolumeClaims
to deliver the storage to Pods that need them.
Storage configuration
In the ArangoDeployment
resource, one can specify the type of storage used by groups of servers using the spec.<group>.storageClassName
setting.
This is an example of a Cluster
deployment that stores its Agent & DB-Server data on PersistentVolumes
that use the my-local-ssd
StorageClass
apiVersion: "database.arangodb.com/v1alpha"
kind: "ArangoDeployment"
metadata:
name: "cluster-using-local-ssh"
spec:
mode: Cluster
agents:
storageClassName: my-local-ssd
dbservers:
storageClassName: my-local-ssd
The amount of storage needed is configured using the spec.<group>.resources.requests.storage
setting.
Note that configuring storage is done per group of servers. It is not possible to configure storage per individual server.
This is an example of a Cluster
deployment that requests volumes of 80GB for every DB-Server, resulting in a total storage capacity of 240GB (with 3 DB-Servers).
apiVersion: "database.arangodb.com/v1alpha"
kind: "ArangoDeployment"
metadata:
name: "cluster-using-local-ssh"
spec:
mode: Cluster
dbservers:
resources:
requests:
storage: 80Gi
Local storage
For optimal performance, ArangoDB should be configured with locally attached SSD storage.
The easiest way to accomplish this is to deploy an ArangoLocalStorage
resource. The ArangoDB Storage Operator will use it to provide PersistentVolumes
for you.
This is an example of an ArangoLocalStorage
resource that will result in PersistentVolumes
created on any node of the Kubernetes cluster under the directory /mnt/big-ssd-disk
.
apiVersion: "storage.arangodb.com/v1alpha"
kind: "ArangoLocalStorage"
metadata:
name: "example-arangodb-storage"
spec:
storageClass:
name: my-local-ssd
localPath:
- /mnt/big-ssd-disk
Note that using local storage required VolumeScheduling
to be enabled in your Kubernetes cluster. ON Kubernetes 1.10 this is enabled by default, on version 1.9 you have to enable it with a --feature-gate
setting.
Manually creating PersistentVolumes
The alternative is to create PersistentVolumes
manually, for all servers that need persistent storage (single, Agents & DB-Servers). E.g. for a Cluster
with 3 Agents and 5 DB-Servers, you must create 8 volumes.
Note that each volume must have a capacity that is equal to or higher than the capacity needed for each server.
To select the correct node, add a required node-affinity annotation as shown in the example below.
apiVersion: v1
kind: PersistentVolume
metadata:
name: volume-agent-1
annotations:
"volume.alpha.kubernetes.io/node-affinity": '{
"requiredDuringSchedulingIgnoredDuringExecution": {
"nodeSelectorTerms": [
{ "matchExpressions": [
{ "key": "kubernetes.io/hostname",
"operator": "In",
"values": ["node-1"]
}
]}
]}
}'
spec:
capacity:
storage: 100Gi
accessModes:
- ReadWriteOnce
persistentVolumeReclaimPolicy: Delete
storageClassName: local-ssd
local:
path: /mnt/disks/ssd1
For Kubernetes 1.9 and up, you should create a StorageClass
which is configured to bind volumes on their first use as shown in the example below. This ensures that the Kubernetes scheduler takes all constraints on a Pod
that into consideration before binding the volume to a claim.
kind: StorageClass
apiVersion: storage.k8s.io/v1
metadata:
name: local-ssd
provisioner: kubernetes.io/no-provisioner
volumeBindingMode: WaitForFirstConsumer