Non-Blocking Concurrency Control
A concurrency model for lakehouse table formats that uses distributed timelines rather than locks or optimistic retries, allowing multiple writers to simultaneously mutate the same table and rows without blocking or failing commits. Native to Apache Hudi 1.0.
Summary
A concurrency model for lakehouse table formats that uses distributed timelines rather than locks or optimistic retries, allowing multiple writers to simultaneously mutate the same table and rows without blocking or failing commits. Native to Apache Hudi 1.0.
Replaces Optimistic Concurrency Control (OCC) as the write coordination mechanism for high-velocity streaming and CDC workloads on object storage. OCC forces constant retries under concurrent write contention, degrading throughput. NBCC assigns each writer an independent timeline, merging results asynchronously.
- Not the same as eventual consistency — writes are still serializable via timeline ordering and sequence number resolution.
- Does not eliminate the need for compaction — background merge of timelines is still required for read performance.
- Currently Hudi-specific; not available in Apache Iceberg or Delta Lake, which remain on OCC.
enablesApache Hudi — core concurrency model for Hudi 1.0 architectureenablesCDC into Lakehouse — allows high-frequency CDC ingestion without commit conflictssolvesRead / Write Amplification — eliminates retry-induced write duplication under contention
Definition
A concurrency model for lakehouse table formats that uses distributed timelines rather than locks or optimistic retries, allowing multiple writers to simultaneously mutate the same table and rows without blocking or failing commits.
Optimistic Concurrency Control (OCC) forces constant retries under high-velocity concurrent write workloads, degrading throughput and causing commit failures. Non-Blocking Concurrency Control assigns each writer an independent timeline, merging results asynchronously. Native to Apache Hudi 1.0, it enables high-frequency CDC ingestion without the commit contention that cripples OCC-based formats under streaming loads.
High-velocity CDC ingestion into lakehouse tables, concurrent multi-writer streaming pipelines, real-time lakehouse architectures requiring sub-minute data visibility.
Recent developments
- Hudi NBCC (RFC-66) merged: lock-free multi-writer concurrency control. Hudi's Non-Blocking Concurrency Control replaces lock-based OCC with a timeline-merge model — multiple writers persist updating Log Files freely to the same File Slice, with conflict resolution deferred to the compaction stage. Per GitHub — Hudi RFC-66 NBCC PR #7907.
- NBCC is recommended over OCC for high-throughput concurrent + streaming workloads. The 2026 guidance from the Hudi team: NBCC over OCC for any workload where writers contend on the same file slice. OCC's retry storms under streaming load were the load-bearing reason to ship NBCC. Per Apache Hudi — Concurrency Control docs.
- MVCC stays for table services (clustering, compaction, cleaning) running asynchronously. Hudi's MVCC is the substrate that lets table services run without blocking writers — orthogonal to OCC/NBCC, which arbitrate concurrent writers. The two concurrency mechanisms coexist + cover different concurrency dimensions. Per Apache Hudi — Concurrency Control docs.
- Multi-writer support has been in Hudi since 0.8.0; matured through OCC → NBCC. Hudi's multi-writer roadmap: 0.8.0 added basic multi-writer; OCC came in subsequent releases; NBCC is the production-default lock-free path. Per Medium — Multi-Writer Support with Apache Hudi (Sivabalan Narayanan).
- "Conflict resolution deferred to compaction" is the architectural insight. Instead of resolving conflicts at write time (OCC retries), NBCC writes freely + defers conflict resolution to a later compaction step — trades some compaction-time work for much higher write throughput. Streaming-CDC use cases that previously hit OCC commit-failure walls now scale linearly. Per GitHub — Hudi RFC-66 PR.
- Open Lakehouse 2026 framing: NBCC is what makes Hudi the table-format-of-choice for concurrent CDC + streaming. Hudi's concurrency-control story is the 2026 differentiator vs Iceberg V3 + Delta Lake — when many writers fight for the same table, NBCC's lock-free model wins; for batch-dominated or low-concurrency workloads, the differentiator doesn't matter and Iceberg/Delta compete on other axes. Per X — Apache Hudi on Concurrency Control in an Open Data Lakehouse.
Connections 5
Outbound 5
Resources 3
Deep technical comparison of streaming ingest mechanics across table formats, including analysis of NBCC in Hudi versus OCC in Iceberg and Delta.
Comprehensive guide covering concurrency models, write paths, and conflict resolution across all major open table formats.
Performance comparison of streaming concurrency models between Paimon LSM-trees and Hudi NBCC under CDC workloads.