Embracing Multi-Cloud Cassandra at Walmart
summary
Walmart runs 300 production clusters totaling multi‑petabytes on a mix of a million‑core private cloud plus Azure and GCP, 100% on VMs with most clusters on Cassandra 4.x and a few on 4.1/5 beta. Clusters map one Cassandra ring per provider across Regional triplets with single‑digit ms between providers; rings are added via oneops automation. Observability uses Prometheus + Grafana + M3 and the insta_cluster metric exporter. Performance parity relies on systematic benchmarking with cassandra‑stress and NoSQLBench, payload‑size tests, and real‑app validation.
AI-written summary · about
chapters (17)
- 0:00Multi-cloud Cassandra intro
- 1:05Current Cassandra footprint
- 2:09Why multi-cloud strategy
- 4:43Modernizing apps for multi-cloud
- 7:34Multi-cloud strategy in practice
- 10:05Choosing cloud providers
- 11:49Why Cassandra fits multi-cloud
- 13:13Benchmarking and automation
- 16:29Cloud provider challenges
- 18:16Q&A: Latency considerations
- 20:20Q&A: Observability stack
- 21:03Q&A: Benchmarking gaps
- 23:32Q&A: Scaling token rings
- 24:35Q&A: Relational to Cassandra
- 27:26Q&A: Data consistency
- 29:55Q&A: Use case successes
- 31:20Closing remarks
related talks
29mCassandra - an introduction
43mMulti-Cloud Magic: Leveraging Multi-Cloud Clusters in Real-World Scenarios
46mCassandra Serving Netflix @ Scale
40mMulti Cloud Magic: Leveraging Multi Cloud Clusters in Real World Scenarios
42mLeveraging Multi-Cloud Clusters in Real-World Scenarios
35mMission Critical Data, and Evaluating the Community in OSS Tech
01h 05mRoundtable: Cloud Databases as a Service
29mPre-cloud Cassandra to Multi-Cloud Kubernetes