RC RANDOM CHAOS

Jepsen Analysis Uncovers Safety Issues in ParticleDB 1.1.4

· via Hacker News

Original source

Jepsen: ParticleDB 1.1.4

Hacker News →

ParticleDB is a Rust-based database engine built by Orkes to consolidate multiple workload types—SQL, key-value storage, vector search, and analytics—into a single system. Originally developed to support their Conductor workflow orchestration platform, the company refactored and prepared it for broader release, eventually bringing in Jepsen for independent safety testing.

The analysis revealed critical gaps between ParticleDB’s marketing claims and actual behavior. The system advertises full ACID compliance and formal TLA+ verification, but no TLA+ specification was ever made available. Documentation conflicts on basic guarantees: different sources contradict whether specific write-ahead log sync modes provide equivalent durability. ParticleDB ignores or silently rejects the standard SQL SET TRANSACTION ISOLATION LEVEL statement, instead requiring startup configuration flags with vaguely defined consistency models. The company claims “Full ACID” while simultaneously describing “snapshot isolation,” which is not fully isolated. When questioned, Orkes provided contradictory internal documentation that was difficult to interpret.

Testing covered versions 1.1.4 through 1.1.18 using custom Jepsen workloads designed to detect consistency violations across different transaction modes and sync settings. The combination of unclear safety guarantees, contradictory documentation, and absent formal verification raises significant questions about whether ParticleDB delivers the strong consistency needed for mission-critical systems.

Read the full article

Continue reading at Hacker News →

This is an AI-generated summary. Read the original for the full story.