Jepsen finds data loss and isolation failures in ParticleDB
Original title: Jepsen: ParticleDB 1.1.4
Tests from 1.1.4 through 1.1.18.11.rc-22 found 96 issues; default occ mode still lost committed writes.
Jepsen, funded by Orkes, tested ParticleDB 1.1.4 through 1.1.18.11.rc-22 and reported 96 issues, including crashes, deadlocks, data loss, and isolation anomalies. Most required one server, one client, and no faults. By 1.1.18.11.rc-22 all but ten appeared resolved, but default occ mode still lost committed writes and serializable still allowed lost updates. Provided builds lacked Raft, so only single-node behavior was tested. Orkes says 1.1.19-rc.13 fixes the rest.
Unconfirmed: Orkes claims 1.1.19-rc.13 resolves the remaining issues; that release was not verified in this report.
- Home-page “Full ACID” claims sat next to text describing snapshot isolation.
- Isolation-level SQL was ignored or rejected; behavior depended on startup --txn-mode.
- Documentation disagreed on whether groupsync matched sync durability.
- No TLA+ specification was available despite a “formally proven” claim.
- ParticleDB 1.0.0 already backed some Conductor clusters and a few customer deployments.
Impact: Committed writes lost under the default transaction mode
| Product | Affected | Fixed in |
|---|---|---|
| Orkes ParticleDB | 1.1.4 through 1.1.18.11.rc-22 | 1.1.19-rc.13 |
- Vendors
- Orkes
- Products
- ParticleDBConductor
- Organizations
- Jepsen
Indicators of compromise
auto-extracted · verify before use · export allAll| Type | Indicator | Context |
|---|---|---|
| ipv4 | 1.1.18.11 | fault timeout. These deadlocks ( #71 ) appeared resolved by 1.1.18.11.rc-7. Broken Features Orkes’ public documentation claimed b |
From the original article
13 min read at jepsen.iodatabase designed for AI workloads. We tested versions 1. 1.
By aphyr · jepsen.io
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.