Durability and Tiering
Start with memory when the application can refill its cache. Add local durability to restore resident state after a restart, or a storage tier when retention needs to extend beyond the local working set or disk.
| Storage choice | What it gives you | Boundary |
|---|---|---|
| Memory-only Swytch | Fast, disposable local state; shared keys may have other holders | Losing every holder loses the data |
| Local durability | Open-source disk mirror enabled with --data-dir | Asynchronous sync; eviction removes records; recovery needs sufficient memory and the same storage |
| Atmos | A commercial local-disk tier | Available to commercial customers; setup documentation is supplied separately |
| Swytch Cloud | A network storage tier and peer discovery | Only uploaded, acknowledged state has reached durable storage |
Atmos documentation is supplied separately to commercial customers.
A process restart, Pod replacement, and loss of a node are different failures. Memory disappears with the process. Local storage depends on the lifetime and availability of its disk. A network tier can recover acknowledged state after all local processes stop. Local durability mirrors resident effects; it does not keep evicted keys available as a disk tier.
For local recovery, the memory-pressure fallback is part of the operating contract: if recovery cannot fit, Swytch waits 30 seconds before deleting the mirror and continuing empty.
Adding another Swytch replica does not automatically make a backup of every key. Subscriptions follow use.
Enable local durability, configure Swytch Cloud, or follow Kubernetes tiering. For a rollout plan, read survive pod replacement.