The fundamental idea of the project is described here
Compared to GlusterD there are various points which we consider worth changing:
- Too many layers, hard to debug
- Duplication of task is bad, and can cause in-consistency
- K8s can provide added infrastructure like process management, cluster authentication, monitoring and centralized logging
- Running more than one process in a container defeats the purpose of microservices
- Currently, no ideal solution with Gluster for storage in k8s
Therefore kadalu does some things different:
| GlusterD | kaDalu |
|---|---|
| Clustering / Peer Management | k8s |
| Volume Management | ConfigMap, kubectl apply |
| Brick process management | K8s’s pod management |
| Portmap for Bricks | Not required in new model |
| Service Management (brick, self-heal, etc) | Runs as another container in same pod. So, managed by k8s as any other pod |
| Volfile for Bricks, self-heal etc | ConfigMap |
| Quota, Snapshot, Geo-Replication | CSI / SideCar containers |
- There is no need to have a
glusterdrunning on the host.