What problem does it solve? Operational data stores degrade over time: streaming writes create thousands of small files in Iceberg and Delta tables, unmanaged snapshots explode metadata, and fragmented B-tree or pgvector indexes slow queries and inflate cloud bills. This Skill provides structured runbooks and guardrails to perform maintenance safely without locking production tables or deleting in-flight data. ## Core Features & Use Cases - Lakehouse Compaction & Cleanup: Bin-pack compaction merging Deletion Vectors into 256MB Parquet blocks, manifest rewriting, 7-day snapshot expiration with a 50-snapshot floor, and orphan vacuuming with a 72-hour grace window. - Relational & Vector Index Maintenance: Concurrent reindexing of HNSW/IVFFlat pgvector indexes, bloat detection, VACUUM ANALYZE, and mandatory 2-second lock timeouts for safe DDL. - Data FinOps Attribution: Calculates file reduction ratio, storage reclaimed, and projected S3 GET savings, tagging runs with CostCenter, Environment, and TableOwner. - Use Case: A data platform engineer notices query latency rising on a streaming Iceberg table. The Skill guides them through assessing small-file distribution, running rewrite_data_files, expiring stale snapshots, and emitting a FinOps report showing reclaimed storage and cost savings. ## Quick Start Use the database-maintenance skill to plan a compaction and snapshot expiration run for my Iceberg table lakehouse.silver.orders with a rollback plan and FinOps report.
