Reducing Client Latency and Timeouts When Running Cassandra on Public Cloud
summary
Public cloud maintenance and VM freezes produce multi‑second tail latencies (observed ~8–30s) that make local-one reads and local quorum fail or return stale data; mitigation combines speculative execution (send reads to multiple replicas, requires idempotent queries, trades predictability for extra load), automated node rotation using Azure Scheduled Events, active inter-node probes to detect network hiccups, and a read‑through local SSD cache (writes remote + local, rebuilds on loss) — these reduce timeouts but cost extra resources.
AI-written summary · about
chapters (6)
- 0:00Introduction and context
- 1:12Latency and speculative exec
- 14:19Infrastructure and maintenance
- 14:58Storage and disk challenges
- 15:51Diagnosis and mitigation
- 26:40Closing remarks
related talks
40mLatency, Throughput & Fault Tolerance: Building a Modern Database
23mDeciphering Cassandra’s Prophecy: Preventing the Fall of Your Application
45m8ms/99th write percentile latency - is it fast? (...)
48m8ms/99th write percentile latency - is it fast?
39mMaking Cassandra Faster in Cloud-native architecture
01h 05mRoundtable: Cloud Databases as a Service
57mDSS Keynote — Back to the Future: Prepare Your Data Infrastructure for Anything
54mEvolving Transactions in Apache Cassandra