Lightning Talk: Hotspot detection and handling in FoundationDB
summary
Hotspots are detected by correlating FDB metrics and app context: SnowflakeDB maps transactions to FDB partitions, buffers per-transaction stats in memory (or samples at high throughput), computes a partition rank, and flags partitions where storage queue and non-durable versions rise together (write hotspot) or where only non-durable versions rise (read hotspot). Fixes include raising write-task priority, application-level throttling, caching or access-pattern fixes; sustained accesses near a gigabyte per minute cause system-wide degradation.
AI-written summary · about
chapters (4)
- 0:00Understanding hotspots
- 1:04Detecting data issues
- 1:57Mitigation strategies
- 2:51Analyzing access patterns
related talks
18mPerformance Profiling in FoundationDB
11mConsistent Caching in FoundationDB
46mFoundationDB: The Best-Kept Secret of New Distributed Architectures!
01h 04mHow We Build Firebolt
19mVMware Network Insight's Configuration-Store Layer
18mFoundationDB Record Layer: Open Source Structured Storage on FoundationDB
34mFoundationDB, the Black Knight
37mBeyond Hot Spots: Revolutionising OpenSearch Performance Through Heat-Aware Operations