The 2026 Unified Data Architecture: Reconciling Multi-Cloud Data Lakehouses
Multi-cloud data lakehouses in 2026 run on Apache Iceberg, open catalogs, and zero-ETL federation. Here's what a composable, unified architecture looks like.
The archive
Writing, page 29
652 posts on Apache Iceberg, lakehouse architecture, data engineering and applied AI.
Why Traditional Lakehouses Fail AI Agents: The Mathematical Case for the Agentic Lakehouse
Traditional lakehouses expose raw directories and ambiguous schemas to AI agents, causing hallucination.
The Era of Zero-ETL Federation: Fueling AI Agents with Real-Time Cross-Enterprise Data
Zero-ETL federation lets AI agents join real-time CRM data with historical lakehouse tables instantly.
Use Hermes Agent for Free With DeepSeek V4 and Slack
Install Hermes Agent, choose a current DeepSeek provider, and connect Slack through Socket Mode with the tokens, scopes, events, and allowlist Hermes requires.
Concurrency, Isolation, and MVCC: How Engines Handle Contention
Databases handle concurrent access using locks, MVCC, or optimistic concurrency control. Here is how each approach works and what tradeoffs each creates.
Hash, Sort-Merge, Broadcast: How Distributed Joins Work
Distributed joins move data across the network using shuffle, broadcast, or co-location strategies. Here is how each works and when engines choose which.
Partitioning, Sharding, and Data Distribution Strategies
Hash partitioning distributes data evenly. Range partitioning enables fast range scans. Both create tradeoffs.
Buffer Pools, Caches, and the Memory Hierarchy
Databases use buffer pools, column caches, and result caches to keep hot data in RAM. Here is how each caching strategy works and what happens when data.
Volcano, Vectorized, Compiled: How Engines Execute Your Query
The Volcano model processes one row at a time. Vectorized execution processes batches with SIMD. Code generation fuses operators into compiled code.
Inside the Query Optimizer: How Engines Pick a Plan
Query optimizers transform SQL into execution plans using rule-based rewrites, cost-based search, and adaptive runtime adjustments.