Revisiting XTM: A Practical Case Study Highlighting Its Needs
summary
Implements a concrete XTM-style change by adding an XTPC field to snapshots and altering heap visibility checks so local shard reads wait for prepared transactions instead of returning stale rows, applied on a postgres_fdw_plus two‑phase commit setup; uses the SmallBank benchmark on a 1 coordinator + 4 shard cluster and shows throughput and 95th‑percentile latency comparable to vanilla PostgreSQL at 1%, 10% and 50% distributed‑transaction mixes; highlights a modified snapshot architecture and the need for postgres_fdw_plus with 2PC APIs.
AI-written summary · about
chapters (10)
- 0:00Introduction and agenda
- 4:05Case study overview banking system
- 8:43FDW-based sharding overview
- 13:01Two-phase commit optimization
- 17:19XTM overview and need
- 23:21Snapshot architecture in PostgreSQL
- 25:29Session consistency problem
- 28:03XTM implementation details
- 31:45Small bank benchmark results
- 35:41Summary and future of XTM
related talks
54mScaling out by distributing and replicating data in Postgres XC
56mA Shared nothing Cluster System Postgres XC
50mSupersized Postgres: Postgres-XL for OLTP and Big Data Analytics
32m"State of XTDB"
46mLeveraging ETS Effectively
01h 06mReconstructing History with XTDB
27mImplementing Strict Serializability with pg_xact
36mRicher SQL — Steering SQL's Future Towards Clojure's Philosophy