Jepsen: ParticleDB 1.1.4
- ID
- 41686
- Status
- summarized
- Published
- 31 Oct 2026, 9:39 PM
- Fetched
- 01 Nov 2026, 8:26 PM
- Provider
- Hacker News
- Category
- dev-community
- Original URL
- https://jepsen.io/analyses/particledb-1.1.4
- Source URL
- https://hnrss.org/best
Summary
- Score
- 7.5
- Created
- 01 Nov 2026, 8:27 PM
- Tags
- Audience
- developersdatabase_learnersai_agent_userssaas_startup_founders
What happened
Jepsen tested ParticleDB 1.1.4 through 1.1.18.11.rc-22, a Rust multi-model database for AI workloads from Orkes, and reported 96 issues—most reproducible with one server, one client, and no faults. By rc-22 all but ten were apparently fixed, but Jepsen still observed lost committed writes in the default occ transaction mode and internal anomalies plus Lost Update under serializable; Orkes has since shipped 1.1.19-rc.13 claiming all remaining issues are resolved. The work was funded by Orkes.
Why it matters
If you are considering ParticleDB for AI-agent state, vector search, or Postgres/Redis-compatible workloads, do not treat “Full ACID” as validated: the report documents lost committed writes under the default occ mode and Lost Update under serializable up to 1.1.18.11.rc-22, so keep durable workflows and payments off it until 1.1.19-rc.13 is independently verified or you run your own fault/correctness tests.
Discussion angle
What would you require before trusting a vendor-funded Jepsen report’s “resolved in rc” claim—and would you run a single-node lost-write test before putting ParticleDB behind agent memory or payments?