Breakout Session
summary
Claims a working replacement for Hbase OLAP by combining PLProxy and Postgres-HLL: a distributed Postgres cluster that uses HLL sketches for commutative, idempotent counts and plproxy to fan out writes. Prototype gets ~2k events/sec on three nodes, batching to ~11k and with more parallel shards to ~25k, runs on loaded clusters (60GB test data), and supports ad‑hoc SQL. Limits include deadlocks on unsorted batches, no cross‑table joins unless same partition key, operational complexity, and not yet productionized.
AI-written summary · about
chapters (6)
- 0:00Start
- 1:27POSTGRES AS DISTRIBUTED ONLINE PROCESSING ANALYTICS ENGINE 1. Problem Statement 2. Distributed Postgres 3. Probabalistic Datastructures 4. Benchmarking Solutions, Unloaded/Loaded.
- 2:49EXPLORING SOLUTIONS Postgres is pretty nice to work with. Makes adhoc analytics simple. Well known replication and backup story
- 4:33APPROACH Two phase commit Commutative, Idempotent data
- 4:54IDEMPOTENT WRITES WITH HYPERLOGLOG Postgres-hll extension Commutative, idempotent Fast, approximate, cardinality
- 8:09CLUSTER CONFIG Partition defs, cluster version, connection config elided Partition mapping is as follows
related talks
38mKeynote | Data Warehouse, Data Lakes, What's Next?
13mBuilding reliable infrastructure with Postgres
50mSupersized Postgres: Postgres-XL for OLTP and Big Data Analytics
27mBuilding a PostgreSQL data warehouse
51mScaling Postgres to the next level at OpenAI
43mJust Use Postgres.
38mPostgreSQL as a distributed computing platform
04h 15mTopical Edition on Databases - Track 2