Introducing Icehouse LakeOps: Automated Maintenance, Full Observability, No Manual Ops

Apache Iceberg promised a simpler, more open lakehouse, but that promise comes with a catch. Engineers hand-roll compaction and cleanup scripts to keep queries fast. Platform leads watch storage costs creep up as metadata and orphaned files pile up unseen. Business leaders feel it in slower dashboards and rising infrastructure spend. The lakehouse meant to reduce operational burden has become another system to babysit.

Starburst built the Managed Icehouse to change that: a comprehensive platform for the Iceberg lakehouse architecture. Today, we’re introducing the newest piece of it: Icehouse LakeOps, a fully automated table maintenance and lake observability solution, built on a dedicated Starburst engine purpose built for Iceberg.

In this session, Ahmed Niyaz and Lucas Lemos walk through how LakeOps solves the Iceberg Ops problem, so engineering teams can focus on getting value from the lakehouse rather than maintaining it.

Hands-Off Maintenance, Built on a Dedicated Engine

• LakeOps runs on a dedicated engine built specifically for Managed Iceberg services, not a generic extension bolted onto a query engine.
• That dedicated engine gives LakeOps optionality. It maintains tables regardless of where the writes come from, whether Flink, Spark, dbt, or Starburst itself. No migration required.
• The goal is hands-off maintenance. Teams shouldn’t need to think about compaction schedules or cleanup runs at all.
• What teams do get visibility into is table health: is metadata under control, is storage being reclaimed, is the table optimized for queries.

Attendees will leave understanding:

• Where LakeOps fits in a modern Iceberg strategy
• Why observability matters for a healthy lakehouse, and how table health metrics make degradation visible before it becomes a problem
• Common Iceberg maintenance issues caused by everyday write patterns, and why they compound over time if left unaddressed
• How Starburst builds a comprehensive Iceberg lakehouse with a unified read and write engine, plus metastore optionality