Storage modes

Choose unlogged or logged mode for durability and throughput trade-offs.

PostRESP supports two storage modes. The same modes work for both the in-process gateway and the standalone service.

Mode	Backing	Restart	Crash	Role
unlogged (default)	Postgres `UNLOGGED` tables	Preserved after clean restart	Table may be truncated; not replicated	Single-instance only
logged	Normal Postgres tables	Preserved	Recovered through WAL; replicable	Crash recovery / HA via Postgres

HA and clustering

The Docker image and default unlogged mode are for a single Postgres instance. They are not an HA topology.

unlogged does not work in a replicated cluster. Postgres UNLOGGED tables skip table WAL, so changes are not shipped by physical streaming replication or logical replication. After failover or on a standby, replicas will not have those keys (and a crash on the primary can truncate the unlogged heaps). Use logged whenever you need Postgres-level HA, standbys, or logical subscribers.

Goal	Mode	Notes
Local / Docker / demos	`unlogged` (default)	Fast shared SQL storage; no replica story
Crash recovery, streaming HA, logical replication	`logged`	Data follows normal Postgres WAL / publications

HA for PostRESP means Postgres is the cluster (primary + standbys, operator failover, one write endpoint) — not Redis Cluster hash slots or multi-primary RESP. Pub/Sub delivery is still process-local today: multiple RESP gateway processes against one DB do not share live subscriptions until a cross-process bus lands (see Commands › Pub/Sub). Practical constraints when you do run logged on a primary/standby setup:

Serve RESP (and schedule pg_cron TTL jobs) on the primary only; standbys are recovery and do not run those writers until promoted.
Point clients at a stable primary endpoint (k8s Service, Patroni / CloudNativePG VIP, etc.).
Do not put cron.* job metadata into a logical publication that would make two writable nodes both execute the same sweeper.

Configuring the mode

Set via StorageMode in SetupConfiguration.json or the PG_RESP_STORAGE_MODE environment variable:

{
  "StorageMode": "unlogged"
}

PG_RESP_STORAGE_MODE=logged

SQL helpers pgresp.storage_mode() / pgresp.set_storage_mode('logged'|'unlogged') change table persistence.

Memory cache after restart

After a Postgres restart, its in-memory table cache starts empty. Docker / compose enable pg_prewarm autoprewarm by default so recently used pgresp.* data is loaded back after restart. Details, disable switch, and manual pgresp.prewarm(): Buffer cache prewarm.

TTL / expiry

TTL uses both lazy and active purge:

Lazy — expiry checked on read/write paths.
Active — a pg_cron job periodically deletes expired keys.

Active expiry is intentionally naive today (not Redis-style random sampling). Details and tunable GUCs live under Commands › TTL and Configuration.