Replication strategies for cloud apps

[lorenzo-w]

Databases

Always use full replication except for maybe large MongoDB databases, which could required sharding.

• MariaDB
  • default chart primary-replica
  • active-active via Galera chart
• Postgres
  • default chart primary-replica
  • active-active via HA chart
  • RAFT-based consensus via YugabyteDB
• MongoDB: default chart, with sharding if possible
• Redis: default chart

Backend storage

• S3, in-cluster, slow failover
  • Longhorn replicated RWO volume + Single-replica Minio
• NFS, in-cluster, slow failover
  • see #115 (Longhorn RWO + NFS server deployment)
• S3, in-cluster, fast failover
  • Multi-replica, multi-node Minio deployment with erasure coding based on Longhorn local RWO
  • Ceph Object Storage based on Longhorn local RWO PVC-Cluster
  • SeaweedFS with S3 enabled (questionable long-term support of project)
• NFS, in-cluster, fast failover
  • CephFS based on Longhorn local RWO PVC-Cluster
  • Longhorn RWX volume (after upstream issue is solved)
• S3, external
  • Storj

Storage mount

• NFS, non-cached: Directly create PV
• S3, cached:
  • Alluxio CSI with S3 understore (Explanation on Caching)
  • SeaweedFS CloudDrive, if it can be used with the Seaweed CSI
  • JuiceFS CSI (Explanation on Caching)

[lorenzo-w]

Example use case: Nextcloud with replicas both in datacenter and on the edge