Databricks explains LTAP: the lake is the database's storage, killing the CDC pipeline
matei_zaharia · x · 2026-11-19
Databricks published a deep dive on LTAP in Lakebase Postgres, arguing 'the lake is the database's storage, not a replica' — eliminating the need for a CDC pipeline. It debunks four misconceptions:
- Just replication? LTAP maintains one logical table, not two synced databases; the path is Postgres (compute cache) → Pageservers (hot pages) → object storage (Delta/Iceberg), with the Lakehouse reading object storage directly.
- One copy means slow queries? Postgres uses heap pages for fast lookups while analytics scan Parquet.
- Analytics must be stale? Queries see consistent snapshots, including recent commits not yet materialized in the lake.
- Logical replication is enough? CDC needs a separate pipeline and a second copy; LTAP doesn't.
More from Infra
- Qualcomm demos Qwen3-VL-32B running locally in FP16 via vLLM on Dragonfly — HowDevelop · 2026-11-19
- Swapping two used RTX 3090s for two Radeon R9700s: is ROCm ready for local inference? — Pear_Virtual · 2026-11-19
- Hugging Face's open-source nudge: 100GB private storage vs ~7TB for public data — RisingSayak · 2026-11-19
- Dev ports MiniMax H3 to Swift, cuts 480p video gen to ~4 min on a 16GB M1 MacBook — Lokoto123 · 2026-11-19
- Solo dev cuts sandbox cold-start to 0.45s, sleeping VMs cost 1¢/month — emilahlback · 2026-11-19
- Team cuts agent sandbox cost 88% by moving to GKE, open-sources gke-sandboxes — emilahlback · 2026-11-19