SLOs for Data-Intensive Services
Talkat SREcon19 Europe/Middle East/Africa · on Wednesday, 2019-10-02· at 11:00· Track 1: The Liffey B·
summary
Booking.com monitors a reservation pipeline that handles 1.5M reservations/day and 23 years of terabytes of data. SLOs evolve from naive HTTP success/latency to concrete data SLOs: replay sampled queries to two data nodes to measure consistency and freshness, per-query complexity scoring and bucketing for latency targets, and Hadoop MapReduce integrity jobs over historical data for durability and automated repair. Results: actionable percentages of consistent search results, automated mitigation (discard/repair nodes), daily validated snapshots, and fewer false alerts from complex queries.
AI-written summary · about
chapters (18)
- 0:00Intro
- 0:30Agenda
- 2:00Scalability
- 2:50Service Overview
- 3:55Service Monitoring
- 4:30Reservation Service
- 5:03Data Monitoring
- 6:05Consistency
- 7:12Traffic
- 7:56Search Results
- 8:21Time of Data
- 8:35Reservation
- 9:19Storage
- 11:31Brackets
- 12:45Scores
- 14:41Automated Mitigation
- 15:13Automated Repair
- 16:52Awareness
related talks
20mCase Study: Implementing SLOs for a New Service
32m9 Things You Should Do When Starting to Use SLOs
37mReal World SLOs and SLIs: A Deep Dive
28mEngineering Data Reliably Using SLO Theory
23mGoing from 30 to 30 Million SLOs
33mHow and why we lowered our SLO
18mSLICK: Driving SLO Culture
25mAn SLO-Driven Approach to Enhance Kubernetes Cluster Reliability