Explore a technique for building a performant configuration-store layer in this 19-minute conference talk by Guarav Agarwal from VMware. Learn how to solve write-hotspot issues in ordered-log models, such as change-log tables, when using FoundationDB (FDB). Discover a method to preserve desired ordering while avoiding severe key concentration, balancing the trade-offs of increased read-time overhead and potential write conflicts. Delve into topics including hot shards, unbalanced mutations, row structure approaches, ordering challenges, and strategies for minimizing latency. Gain insights on handling large values and understand the potential for enhancements in FDB core to address conflicts.
Building a Performant Configuration-Store Layer
Overview
Syllabus
Introduction
Change-Logs - Hot Shards Use-cases: Indexing, cache_sync etc.
Change-Logs: Unbalanced Mutations
Change-Logs: Row Structure Approach • Switch SS faster, small bursts are okay • Unique, non-contiguous prefix (buckets)
Change-Logs: Ordering Issue
Minimizing Latency Latency sources
Handling Large Values: Add()
Recap
Taught by
Linux Foundation
