Data Science for Infrastructure: Observe, Understand, Automate
Talkat KubeCon + CloudNativeCon North America 2021 · on Friday, 2021-10-15· at 11:00· Concourse Hall 152 + Online·
summary
Pixie is an open-source observability agent that uses eBPF-based auto telemetry to capture full SQL requests and HTTP traces with under ~2% CPU overhead, compiles px scripts as Python/Pandas dataflows, and runs models or regexes in-cluster. A demo pipeline classifies suspicious SQL via regex and sends a Slack alert for SQL injection. Another demo exposes Pixie scripts (px) as metrics to the Kubernetes custom metrics API and drives an HPA that scales up to 10 pods targeting 20 req/s — and it works live in the demo.
AI-written summary · about
chapters (10)
- 0:00Start
- 1:11What we learned in the data space
- 2:16How to do data-driven automation?
- 4:58We built Pixie to solve these problems
- 7:08PxL provides a programmable API for Pixie
- 9:16PxL scripts use transforms to analyze data
- 9:56PxL provides an interface to work with data
- 14:31What is a SQL injection? SQL injection is a code injection technique used to attack applications, in which malicious S. statements are inserted into an entry field for execution
- 14:50Example SQL injection
- 25:20Very sophisticated demo app
related talks
14mFlexible, Open and Easy Observability for Developers
04mInstant, Auto-telemetry for K8s Observability
34mTotal Visibility: Harnessing Pixie for a Full Observability Landscape
26mTriaging Real-time Security Threats with eBPF-powered Observability
10mLightning Talks: Detecting Data Exfiltration on the Edge with Pixie
28mBpftrace Meets Pixie: Dynamic Monitoring of K8s Clusters Unleashed
13mKeynote: Debugging in Prod Without (Lots Of) Pain
26mFlux + Observability: Featuring Prometheus Operator and Pixie