Migrations under Production Load: How to Switch...
summary
Four-phase checklist: justify the change, set up parallel writes, compare results, optimize rollout, then wait and clean up. Two real projects show the specifics: a feed rebuild that forced recreating a self-hosted Cassandra cluster and wasted work because delivery was actually a Kafka switch, and a high‑traffic MySQL to Cassandra migration for a service at tens of thousands of req/s that discovers missing rows, backfills from MySQL, and finds 12 mismatches out of ~6 million before accepting slightly higher read latency and faster writes.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 1:35What is a migration
- 2:30The 4 phases
- 4:37Feed Migration
- 11:47Database Migration
- 12:17Problems
- 18:47Latency
- 20:17Waiting Stage
- 21:27Conclusion
related talks
51mMigrations - The Hardest Actual Problem in Computer Science
24mZero Downtime Mastering Cassandra Live Migrations at Scale
32mMigrating Workloads from Mainframes to Cassandra - Problems, Experience and Learnings
33mDatabase Schema Migrations with Zero Downtime
40mLatency, Throughput & Fault Tolerance: Building a Modern Database
18mSolving Migration Pipelines
35mDatabase migrations without downtime
32mMigrating Applications Between Cassandra Clusters with No Downtime