Apache Cassandra® at Community Over Code 2026

Scott Andreas — Engineering Manager, Core Databases, Apple, Inc.

Cassandra’s Dynamo roots produced a database with tradeoffs that weren’t always familiar to users. This presentation puts together the puzzle pieces of years of work in the community: Transactional Metadata, distributed transactions, mutation tracking, flexible placements and tablets, storage-attached indexes. Together they show what Apache Cassandra is becoming, a normal database that works the way users expect.

Bernardo Botella Corbi — Software Engineer, Apple

CASSANDRA-20854 introduced a native CPU profiler to the Cassandra toolset. This session covers how to use it and how to read its results, load-testing the features 6.0 is set to introduce, including Accord, change data capture and constraints.

Dmitry Konstantinov — System Architect, NetCracker Technology

A year of source-level optimisation work on the write path: request decoding, data validation, coordination, memtable allocation and flushing, compaction, and metrics collection. Includes the benchmarking setup and tooling used to find each bottleneck, and the optimisation patterns that generalise.

Fedor Ryabinin — Postdoctoral Researcher, IMDEA Software Institute

Cassandra 6.0 introduces strictly serializable multi-shard transactions that often complete in a single WAN round trip, enabled by Accord (CEP-15), which replaces Paxos. Rather than covering the implementation, this talk explains the protocol itself and traces the path from Paxos to Accord, the first leaderless protocol deployed in a widely used production system.

Branimir Lambov — Software Engineer, IBM/DataStax
Kamil Leśniak — Software Engineer, IBM

Memtables are the single largest contributor to Cassandra’s garbage collection overhead. This talk describes the final stage of a long progression: a fully off-heap memtable implementation building on the trie memtable introduced in CEP-19, and a cornerstone of the flat key infrastructure required for CEP-57. Covers configuration, measured performance impact, and the engineering behind it.

Joel Shepherd — Principal Software Engineer, Amazon.com

Cassandra currently supports a single authenticator per node, which makes migrating between authentication methods hard and diverse client types harder. CEP-50 lets clients advertise the methods they support while the server retains the decision. Covers the current protocol flow, how negotiation extends it, and real implementation problems, including mapping external identities such as AWS IAM principals to database roles.

Francisco Guerrero and Saranya Krishnakumar — Software Engineers, Apple, Inc.

Analytical workloads mean full table scans and long-running pipelines that compete with latency-sensitive production traffic. The Cassandra Analytics library sidesteps the read/write path entirely, interacting instead with Cassandra Sidecar for direct SSTable access and cluster metadata. This talk covers the integration and how both projects will evolve alongside Cassandra’s emerging features.

Hayato Shimizu — CTO, AxonOps

cqlsh has long been the primary interface to Cassandra. CQLAI is an open source alternative, Apache 2.0 licensed, aiming at fast startup, distribution as a single binary, and no Python runtime dependency. Includes a live demo, import and export in CSV, JSON and Parquet, and lessons from building open source tooling around Cassandra.

Youki Shiraishi — Researcher, NTT, Inc.

A code-informed benchmark study of Accord using NoSQLBench, with workloads inspired by YCSB-style account-transfer and balance-check patterns. Preliminary findings on how contention affects fast-path success, whether read-modify-write transactions carry substantial overhead over read-only ones, and how much the Ephemeral Read optimisation delivers.

Branimir Lambov — Software Engineer, IBM/DataStax
Christophe Bornet — Software Engineer, IBM

Manual partitioning strategy selection and the inability to sort between partitions cause a great deal of user frustration, and are a stumbling block to efficient materialized views and global secondary indexes. Tombstone handling is a challenge to even experienced users. CEP-57 addresses both through byte-comparable key translations that flatten the key hierarchy into a single byte sequence, combined with trie-based access structures.

Francisco Guerrero — Software Engineer, Apple, Inc.

Zero Trust rejects the assumption that traffic inside a network boundary can be trusted. This talk covers running Cassandra, Cassandra Sidecar and Cassandra Analytics under a strict Zero Trust model with mutual TLS: configuration across internode and client-to-node communication, certificate identities as the basis for authentication and authorization, certificate rotation, and migration from password-based authentication.

Maxim Muzafarov — Open-source Software Engineer

Cassandra management has always been fragmented across JMX, REST and nodetool, each requiring its own implementation of every operation. CEP-38, accepted in October 2025, introduces a unified CQL-based management API. This talk walks through the port decision: why sharing port 9042 proved impossible, what a dedicated port costs operationally, and how the execution model enforces what is permitted. Ends with a live demo of registering and executing a new command.

Jane He — IBM

Rolling restarts are routine, but they can still look disruptive from the client side: drivers ignore topology DOWN events, and existing mechanisms only signal per request, so drivers keep sending until connections fail. CEP-59 adds an in-band protocol event letting a node tell clients it intends to close, turning abrupt failure into controlled draining. Includes a demonstration of the behaviour before and after.