LogDevice At Scale
summary
Meta runs LogDevice at roughly nine exabytes per month and uses Flexible Paxos with 3x replication for durability. A leader-side mechanism writes a special reconfiguration step into the log recording which acceptors are currently reachable so a new leader can obtain promises without blocking reads during large maintenance or regional outages. This reduces failover-induced read blackouts, sharply improves the six‑nines write SLA window in production, and adds prioritized background re-replication for records with only one or two copies.
AI-written summary · about
chapters (7)
- 0:00LogDevice at Meta scale
- 7:32Flexible Paxos trade-offs
- 12:54Flexible Paxos read availability
- 15:43Fixing read unavailability
- 17:31Reconfiguration fix results
- 18:15Replication priorities future work
- 19:17Membership reconfiguration future work
related talks
45mOK Log: Distributed and Coördination-Free Logging
39mTrust Deterministic Execution to Scale & Simplify Your Systems
20mFine Grained Replicated State Machines for a Cluster Storage System
19mScalog: Seamless Reconfiguration and Total Order in a Scalable Shared Log
23mProtecting your resources with sanlock
18mLeveraging Multiple Regions to Improve Site Reliability:
26mLocal-first
52mThe Computer Science behind a modern distributed data store