Chasing unicorns : 0 dataloss and 99.999% availability
summary
Explains why mathematical zero data loss and 100% uptime are impossible and shows concrete tradeoffs: CAP and PACELC constraints, five nines (≈5 minutes/year) availability targets, RPO examples like 200 ms lag between primary and synchronous standby, manual failover ~30 minutes, automatic failover ≲5 minutes, and multi‑queen/BDR for multi‑writer scenarios. Lists real tools and practices: physical backups + WAL archiving, daily restore tests, pgBouncer, HAProxy, Patroni, logical replication, pg_rewind, config in git, and where each option raises cost or complexity.
AI-written summary · about
chapters (9)
- 0:00Start
- 1:38Who Am I
- 7:03The Cap Theorem
- 8:15Network Partitioning
- 11:31Package Theorem
- 13:46Vocabulary
- 15:53What Rpo and Rto Is
- 28:58Cold Standby
- 38:48Oracle to Postgres
related talks
46mDeploying Maximum HA Architecture With Postgres
36mDesigning Layered High‑Availability Architectures for PostgreSQL on a Budget
42mDon't Do High Availability, Do Right Availability
47mPostgreSQL High Availability: The Considerations and the Candidates
55mHigh Availability with PostgreSQL and Pacemaker
22mPostgreSQL Disaster Recovery Fundamentals
33mFun with Postgres High Availability Poker
25mWaiting for Commit