Smooth Sailing...
summary
Migration moves ~900 clusters from CentOS to Ubuntu for Azure Cosmos DB for PostgreSQL and finds glibc-driven collation changes force reindex text indexes (char/text/ citext) and a PostGIS upgrade. Validation uses a custom validator tool that forks clusters with PITR and runs dry‑runs; pg_prewarm is used to preload tables. Performance fixes include increasing vCPU (sometimes to 32–64), tuning maintenance_work_mem and VM overcommit. Result: about 400 under 5 minutes downtime, others up to ~16 minutes.
AI-written summary · about
chapters (14)
- 0:00Introduction and speaker welcome
- 1:27About Azure Cosmos DB for PostgreSQL
- 3:38Why migrate from CentOS to Ubuntu
- 5:01Initial migration plan and collation issues
- 8:22Impact of collation changes on sorting
- 10:10Detecting affected indexes for reindexing
- 11:57Optimizing memory with pg_prewarm
- 13:11Ensuring smooth migration with validator tool
- 17:13Overcoming out of memory challenges
- 23:02PostGIS extension version upgrade
- 24:52Handling range partition changes
- 26:37Summary of migration challenges
- 27:40Customer communication and global coordination
- 28:48Q&A session
related talks
28mCitus Cluster: our zero-downtime migration story
26mMastering Large-Scale PostgreSQL Migration
50mUnleashing the Power of Azure for PostgreSQL
24mMigrating from Oracle to Azure Database for PostgreSQL, Seamlessly
24mCitus from the Customer eyes
32mAuto scaling Azure Cosmos DB for PostgreSQL with Citus, Grafana & Azure Serverless
24mLessons learned: Migrating from AWS-Hosted PostgreSQL RDS to Self-Hosted Citus
46mBig databases, many servers, data center or cloud - it's all here!