AI
Longhorn Architecture and Project Growth
Longhorn runs as a hyperconverged storage layer on Kubernetes nodes, exposing volumes through a CSI-compatible interface. The control plane is the Longhorn manager; the instance manager handles engines, replicas, and downstream disks. A REST API coordinates internally, and a dedicated longhorn-ctl CLI covers day-zero through day-two operations. Public metrics at longhorn.io show more than 2,000 nodes tracked, with node count growing 35% year over year and cluster count growing more than 50%.
V2 Data Engine: SPDK Design and Performance Results
The V2 engine replaces Longhorn’s in-house iSCSI stack with SPDK. The front end defaults to NVMe-oF over TCP, with VFIO-user available and tunable queue-depth parameters added in 1.11. SPDK normally requires dedicated CPU cores (poll mode), so the team is adding an interrupt mode to allow shared-resource operation like V1. Performance tests on Oracle OCI IO-intensive instances show significant IOPS gains as CPU count rises, with V1 results in row one for direct comparison.
Longhorn 1.11 Key Features
Release 1.11, shipped in January, promoted V2 to technical preview. To qualify, the team required two criteria: feature parity with V1 and sufficient test coverage. Of 378 existing pytest end-to-end cases, 62% now run against V2. An additional 330 robot-framework cases cover both engines. Other 1.11 additions include parallel replica rebuilding, a smarter replica scheduler that accounts for concurrent rebuilds, dedicated storage-network support for RWX share-manager traffic, and per-disk active monitoring for both V1 and V2.
Longhorn 1.12 Roadmap: V2 GA and Erasure Coding
Version 1.12 targets V2 general availability, closing the remaining feature-parity gaps and wrapping automation coverage. Live engine upgrades will ship, though behavior differs from V1 to accommodate the SPDK daemon. IPv6 support, already in V1, extends to V2. Engine migration, where the client initiator and engine run on separate nodes, lays the groundwork for V2 live upgrades. The most significant new feature is erasure-coding sharding via BDF on SPDK, which breaks the current replication-only constraint where every replica must be the same size.
CNCF Graduation Push and Community Health
Divya Mohan took over Longhorn community management to drive the project toward CNCF graduation. In the 1.11.0 release, more than 50% of contributors were non-SUSE. Monthly community calls launched in June 2025, alternating between APAC and Americas/EU time zones, with recordings on the official YouTube channel. The team is dividing the project into areas to staff roles from outside SUSE. Current contributor count sits at 72 people globally, which Mohan called a good start but not a great one.
Notable Quotes
we have over 600 test case end to end it’s not integration testing or or unit test case this end to end test case including the native test case like involuntary no down clust up this kind of or stress testing for IO David Ko · ▶ 13:20
72 people across the globe is not really a great start. Um it’s good but it’s not a great start. So, we want more of you all to come and join us. Divya Mohan · ▶ 29:27
you can write beautiful code uh but if it’s not actually solving a problem it’s not really useful to anyone. Divya Mohan · ▶ 27:33
Key Takeaways
- Longhorn node deployments grew 35% year over year and cluster count grew more than 50%.
- V2 SPDK engine reached technical preview in 1.11 with 62% of V1 test cases passing against it.
- Erasure-coding sharding in 1.12 will remove the per-replica size constraint that replication mode imposes.
About the Speaker(s)
David Ko is Engineering Director at SUSE with over 20 years of software development experience across microservices, distributed systems, Kubernetes, and cloud-native data processing.
Divya Mohan is Principal Technology Advocate at SUSE and a maintainer for both the Kubernetes and Longhorn CNCF projects. She leads Longhorn community management and previously worked extensively in systems engineering.