Skip to content

Bundled PostgreSQL runs on node-local storage: registry, audit log and durable-exec state at ~24h RPO #37

Description

@raisedadead

Summary

The bundled PostgreSQL StatefulSet stores its data on the cluster's default local-path StorageClass. That volume holds the site registry (source of truth since v1.4.0), the audit_log, the deploy index, the outbox, and the Hatchet tenant's entire durable-execution state.

Properties of the current binding:

  • local-path — node-local directory, not replicated and not network-attached, which pins the pod to one node
  • reclaim policy Delete
  • ALLOWVOLUMEEXPANSION: false

The only durability layer is the nightly logical backup to R2, so the recovery point objective for all of the above is up to 24 hours.

This is the artemis-specific instance of the ADR-008 gap already recorded in the platform state audit (docs/architecture/universe-state-2026-07-17.md §4-3: no network-attached storage layer exists on this galaxy).

Fix direction

Needs a design decision rather than a patch. Options include network-attached block storage for the PVC, a replicated Postgres (the CNPG operator already runs on the sibling galaxy), or continuous archiving to shrink the RPO. Worth scoring against ADR-008 and the DR runbooks before picking.

Not a code change in this repo; filed here to keep the artemis reliability picture in one tracker.

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or request

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions