We are pleased to announce that Cockroach Labs has a new paper appearing at SIGMOD 2026: Scalable Leader Leases For Multi Consensus Groups in CockroachDB (download the PDF here). This is the latest in a series of papers we've published at SIGMOD describing the core techniques that power CockroachDB, following our 2020 paper on CockroachDB's distributed SQL architecture, our 2022 paper on multi-region, and our 2025 paper on multi-tenant CockroachDB Cloud.
The new paper addresses a longstanding scaling problem in distributed SQL databases. Systems like CockroachDB shard data into many small, replicated data ranges, each of which needs to designate a single replica that holds a “lease” to serve reads efficiently. At the scale of modern clusters, with hundreds of thousands or millions of these ranges, coordinating those lease-management decisions can consume significant CPU and slow the cluster's recovery from failures.
The paper is co-authored by Ibrahim Kettaneh, Mira Radeva, Arul Ajmani, Sumeer Bhola, Nathan VanBenschoten, and Rebecca Taft.
What is SIGMOD and why does it matter for database research?
The ACM SIGMOD/PODS Conference is one of the premier venues for database research, bringing together academic and industrial researchers each year to share work on data management. This year's conference takes place from May 31 to June 5, 2026 at the Sheraton Grand Bengaluru Whitefield Hotel & Convention Center in Bengaluru, India. We are thrilled to be one of SIGMOD's sponsors this year.
RELATED
SIGMOD 2020 | CockroachDB: The Resilient Geo-Distributed SQL Database
How does CockroachDB scale lease management in distributed databases?
In CockroachDB, data is divided into many small chunks called Ranges, and each Range is replicated across multiple nodes for fault tolerance. To serve reads quickly without coordinating with the other replicas on every request, each Range designates one of its replicas, the leaseholder, as the authoritative source of truth. The leaseholder holds a temporary lease that it must periodically renew, and the system runs continuous health checks to detect when a leaseholder has failed and a new one needs to be chosen.
Historically, CockroachDB has used two different types of leases. The first, called expiration-based leases, is simple: each Range renews its lease on its own timer. They're reliable, but each renewal is a Raft write replicated across the Range's replicas, so as a cluster grows to hundreds of thousands or even millions of Ranges, the constant renewal traffic adds up to a substantial CPU cost. The second, called centralized leases, replaces all that per-Range bookkeeping with a single Range that holds the health status of every node in the cluster. All other leases consult this one Range instead of running their own checks. This is much cheaper, but it concentrates a critical dependency in one place: If that Range becomes slow or unavailable, every centralized lease in the cluster is affected.
In our new paper, we introduce Leader Leases, designed to give us the best of both: the low cost of centralized leases without the single point of failure, and the reliability of expiration leases without the CPU overhead of constant renewals. The result is a distributed lease-management approach that scales efficiently as clusters grow. It reduces coordination overhead and minimizes the operational risks associated with centralized coordination dependencies.
The protocol has three core components:
The Liveness Fabric: a decentralized health-check layer that detects failures at the node-to-node level rather than per Range. This avoids the per-Range health-check costs of expiration leases without introducing the single point of failure of centralized leases.
Leader Fortification: a change to the Raft consensus protocol that prevents leaders from losing leadership unexpectedly, so that the leaseholder can always be the Raft leader rather than drifting onto a different replica.
Leader Leases: the unified lease abstraction itself. By merging the previously separate "leader" and "leaseholder" roles into one replica, Leader Leases eliminate a layer of redundant coordination, simplify failure handling, and shorten outage windows.
The safety properties of the Liveness Fabric are formally verified in TLA+, and the verification model is publicly available on GitHub.
In our experiments, Leader Leases reduced CPU usage for lease maintenance by up to 85% at scale and sustained stable throughput as the number of Ranges grew. The system also recovered within seconds from a wide range of failure conditions including node crashes, network partitions, and disk stalls, improving fault tolerance and recovery behavior at scale. For organizations running globally distributed applications, these improvements improve cluster stability during failures and reduce infrastructure inefficiencies. The work also demonstrates how distributed SQL databases can continue scaling coordination mechanisms without introducing centralized operational bottlenecks or single points of failure.
How to learn more about CockroachDB’s SIGMOD 2026 paper
The paper will be presented in the Industry 3 session on Tuesday, June 2, 2026 at 3:30 PM at the Sheraton Grand Bengaluru Whitefield Hotel & Convention Center. The full paper will also be available in the SIGMOD '26 proceedings on the ACM Digital Library.
If you're attending SIGMOD 2026, please stop by the Cockroach Labs booth! We'll be there throughout the conference and would love to chat about the paper, distributed databases more broadly, or career opportunities at Cockroach Labs, including roles at our Bengaluru office.
We hope to see you in Bengaluru!
Rebecca Taft is Director, Engineering at Cockroach Labs, leading the SQL Queries and Specialized Indexing teams. Notable projects include helping to build the cost-based query optimizer from scratch, and adding support for features such as geospatial indexing and locality-optimized search in multi-region clusters.
