Scheduled Deletions At Scale
summary
Meta builds a Schedule Deletion system that computes TTLs with a decision engine whose outputs are delete immediately / cancel / reschedule, logs trillions of TTL rows in immutable data warehouse tables, and compacts them via an index-bit logarithmic compaction aggregation pipeline that copies rows only O(log n) times. Before handing items to a central deletion framework the engine is re-evaluated, deletions run as synchronous hide + async DFS jobs, failed deletes enter a retry pipeline with alerts; future work moves to mutable storage and split pipelines.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 1:12Agenda and infrastructure
- 3:42TTL decision engine
- 7:17Object scanning pipeline
- 11:11Deletion framework
- 12:39Challenges and data model
- 14:49Deletion flow
- 15:53Key takeaways
- 16:13Future work
related talks
24mGraceful Deletes: Queues, Tasks & Distributed State Management
43mWhat Timescale's Approach to Dropping Chunks Tells Us About Concurrent DDL
36mTime-based domain events - how to deal with time
17mDynamic Scheduling of Approximate Telemetry Queries
22mAsynchronous Computing at Scale: Overview and Learning
34mPatterns for building Highly Available Cluster Schedulers
15mFrom Chaos to Clarity: Deciphering Cache Inconsistencies in a Distributed...
32mCan local-first really scale at the edge?