Object Storaged backed Kubernetes cluster state with Netsy
Podplane uses Netsy as the Kubernetes API server’s etcd-compatible cluster state service.
Object storage is the durable system of record, while each control-plane node keeps a disposable local SQLite copy for fast reads and coordination.
This changes the traditional operational model of Kubernetes, as cluster state is not tied to a particular VM or persistent disk.
A replacement control-plane node can rebuild its local state from the object store, and single-node clusters do not need a separate etcd quorum or disk volume.
In a healthy multi-node cluster, Netsy can acknowledge a transaction after a durable replica quorum and flush records to object storage asynchronously. If there are not enough healthy replicas, including in a single-node cluster, it falls back to synchronous object-storage writes.
Netsy supports Amazon S3, Google Cloud Storage, and compatible S3 services with the conditional-write behavior it requires. Persistent local disks are optional and only improve restart speed; they are not the system of record.
Netsy Snapshots/Seed Mechanism
Podplane creates the initial Netsy snapshot from a versioned “seed” file. The snapshot is uploaded to AWS or Google Cloud object storage using a conditional write and never overwrites an existing cluster, protecting durable state when infrastructure is reapplied.