Beyond Sequential: A Recipe for Async Pipeline Observability and Alerting
summary
eBay shows a concrete recipe: instrument async pipelines with Prometheus counters and histograms, mark each event state as success / retry / abandoned and compute SLIs as good/valid event ratios. A real example: ML recommendations lift revenue by 2.1%, so SLOs matter — e.g. SLO: 99% over 30 days within 10s. Alerts use multi-window burn‑rate rules (critical 2h/1d, warning 6h/3d) and cook up dashboards for backlog, percentiles and monthly error‑budget burn.
AI-written summary · about
chapters (13)
- 0:00Introduction and recipe overview
- 2:19Preparing the ingredients
- 4:21Understanding async systems
- 10:23Defining SLIs and SLOs
- 13:42Measuring availability
- 17:16Measuring latency
- 20:11Building observability dashboards
- 21:43Alerting and error budgets
- 25:14Seasoning alerts
- 28:20Harnessing SLO metrics
- 30:47Advanced serving suggestions
- 32:16Case studies and real-world impact
- 34:40Wrap up and conclusion
related talks
39mActionable Observability
36mHow to Include Latency in SLO-based Alerting
50mApplied Alerting Philosophy
19mObservability for Data Pipelines: Monitoring, Alerting & Tracing Lineage
29mObservability; a gentle introduction
41mA Theory and Practice of Alerting with Service Level Objectives
31mServerless Observability: a case study of SLOs
30mFrom Monitoring to Observability: Left Shift your SLOs with Chaos