.. | ||
dashboards | ||
templates | ||
.helmignore | ||
Chart.yaml | ||
README.md | ||
values.yaml |
SEAWEEDFS - helm chart (2.x)
info:
- master/filer/volume are stateful sets with anti-affinity on the hostname, so your deployment will be spread/HA.
- chart is using memsql(mysql) as the filer backend to enable HA (multiple filer instances) and backup/HA memsql can provide.
- mysql user/password are created in a k8s secret (secret-seaweedfs-db.yaml) and injected to the filer with ENV.
- cert config exists and can be enabled, but not been tested.
prerequisites
kubernetes node have labels which help to define which node(Host) will run which pod.
s3/filer/master needs the label sw-backend=true
volume need the label sw-volume=true
to label a node to be able to run all pod types in k8s:
kubectl label node YOUR_NODE_NAME sw-volume=true,sw-backend=true
on production k8s deployment you will want each pod to have a different host, especially the volume server & the masters, currently all pods (master/volume/filer) have anti-affinity rule to disallow running multiple pod type on the same host. if you still want to run multiple pods of the same type (master/volume/filer) on the same host please set/update the corresponding affinity rule in values.yaml to an empty one:
affinity: ""
PVC - storage class
on the volume stateful set added support for K8S PVC, currently example with the simple local-path-provisioner from Rancher (comes included with k3d / k3s) https://github.com/rancher/local-path-provisioner
you can use ANY storage class you like, just update the correct storage-class for your deployment.
current instances config (AIO):
1 instance for each type (master/filer+s3/volume)
you can update the replicas count for each node type in values.yaml, need to add more nodes with the corresponding labels.
most of the configuration are available through values.yaml