Leaving the WarehouseLakehouse · hard · ~16 minThe company has decided: the Redshift warehouse (creaking at concurrency limits, bill up 60% YoY) is being replaced by an Iceberg lakehouse on S3 with Trino for interactive queries and Spark for pipelines.Still running on Redshift today: 40 dashboards, 9 dbt projects, and the finance close process. The CTO's constraint is blunt: "migrate with zero broken Mondays."An engineer proposes: "freeze writes Friday night, bulk-export everything, point the BI tool at Trino on Monday." Explain why that's a trap, and design the migration that actually works.Evidence

Estate

Notes

Topology

Checkpoints

Why is 'freeze Friday, bulk-export, repoint Monday' a trap for THIS estate?

The PoC found aggregate diffs in 8 of 20 tables. What must parity validation compare before ANY consumer moves?

For consumer cutover, which sequencing respects 'zero broken Mondays'?

Write the quarter's migration runbook: the mirror architecture (including catalog choice and its consequences), parity gates, wave plan, rollback posture, and when Redshift actually dies.