Field notes from operating Cassandra in production.
September 21, 2026
Telemetry and event data are the workload Cassandra was built for — and the one teams most often model into a corner. Time bucketing math, TTL rules, and the TWCS settings that decide whether expired data actually leaves disk.
Read moreSeptember 21, 2026
A node dies at 03:00 and someone reaches for nodetool. What actually happens during bootstrap and replace, which settings decide whether streaming finishes, and the ordering mistakes that cost teams a consistency incident instead of an afternoon.
Read moreSeptember 21, 2026 · RapidMindLabs
Sooner or later something downstream needs every change your cluster accepts — a search index, a warehouse, a Kafka topic. Cassandra's CDC is real but narrow. What the commit-log feed actually gives you, where the dedup and ordering work lands, and when to stop and pick a different pattern.
Read moreSeptember 18, 2026 · RapidMindLabs
Most Cassandra latency outliers we chase are garbage collection pauses, not disks. What to set for heap and collector on JDK 17, why Cassandra 5's trie memtables change the arithmetic, and how to prove a change helped.
Read moreSeptember 17, 2026
IF NOT EXISTS is one line of CQL and four round trips of coordination. What an LWT actually does on the wire, the latency and contention numbers to expect, the modeling patterns that avoid them, and how Accord's transactions in the 5.x line change the calculus.
Read more