Bringing Apache Cassandra Closer to Kubernetes
summary
A Cassandra-based serverless system is built to run only on Kubernetes and is already running in production as Astra, decoupling compute and storage by using S3 as source of truth with local NVMe caches and miss latency around 10–20 ms. Metadata lives in Etcd, compaction is extracted and scales independently as compaction runs as a separate Lambda-like service, and multi-tenancy is achieved by making the ring a keyspace-level construct, with RF=3 and compaction-driven deduplication.
AI-written summary · about
chapters (17)
- 0:00Introduction
- 1:48Why is Apache Cassandra hard to operate
- 5:32Elasticity is a problem
- 7:01Ideal database as a service
- 7:35Ideal databases service
- 8:23Cloud databases
- 9:34Cassandra as a shared library
- 11:22The Logical description
- 12:01The Physical description
- 14:10Scaling
- 14:42Benefits
- 17:54Multitenant data platform
- 21:12Cash miss latency
- 25:04Availability zones
- 26:09Dealing with lost data
- 28:21Cash Coherency
- 32:31Compaction
related talks
42mDistributed Data Stores on Kubernetes
26mTaking Your Database Beyond the Border of a Single Kubernetes Cluster
29mPre-cloud Cassandra to Multi-Cloud Kubernetes
35mKubernetes Persistent Data Challenges – AZ, Region and Multi-Clou...
01h 00mKubernetes and cloud-native data management
39mServerless on Kubernetes with Kubeless [A]
30mScaling Time-Series Data to Infinity: A Kubernetes-Powered Solution with Envoy
33mBuild Serverless with K8s, Kata Containers and Bare Mental Cloud...