[PM] Automatic Heap Dump Analysis for Snowflake’s Cloud Services Layer

The cloud services layer is Snowflake’s coordination and control plane. It is responsible for core system functions such as authentication, access control, metadata management, query parsing and optimization, transaction management, and overall orchestration of compute and storage resources.

This project improves application and platform stability by reducing errors driven by memory limits within the cloud services layer, with emphasis on lowering the frequency of out-of-memory (OOM) incidents. The expected impact is to cut the frequency of OOM-related incidents by an order of magnitude (relative to an agreed baseline), and to meaningfully reduce infrastructure cost – largely by reducing memory consumption across the workloads in scope.

The work centers on designing, building and deploying software that automates triage and notification for memory-related events across staged environments (pre-production through production). The goal is to route each event to a clear owner and attach structured, actionable diagnostic context at the time of alert, so teams can fix root causes faster and avoid repeat incidents.

Betreuer: Sebastian Götz