Real User Metrics with Fastly and Prometheus

This deck covers turning Fastly logs into Prometheus metrics with a simple syslog server and then expanding on that by capturing data from client browsers and pulling that into Prometheus through the same mechanism.

A simple way to gain some insight into what your users are seeing on your site.

Transcript

1. Real User Metrics with Fastly and Prometheus Dewald Viljoen @dewald_v
3. 3 • App metrics • Ingress metrics • Edge proxy metrics • All query-able from Prometheus • Request logs in BigQuery • Stats endpoint • Real-time analytics endpoint • Separate systems
4. 4 How do we get a complete picture?
5. Getting the data out of Fastly
6. 6 • App metrics • Ingress metrics • Edge proxy metrics • All query-able from Prometheus • Request logs in BigQuery • Stats endpoint • Real-time analytics endpoint • Seperate systems
7. 7 Fastly log streaming options • Amazon S3 • Heroku's Logplex • OpenStack • Cloud Files • Honeycomb • Papertrail • Datadog • Kafka • Scalyr • DigitalOcean Spaces • Log Shuttle • SFTP • Elasticsearch • LogDNA • Splunk • FTP • Logentries • Sumo Logic • Google BigQuery • Loggly • Syslog • Google Cloud Storage • Microsoft Azure Blob Storage
8. 8 Fastly logger config
9. 9 Fastly custom log format
10. 10 The fastly-prometheus-exporter
11. 11 Querying the results
12. How can we see what our users are seeing?
13. 13 We need to measure in the browser
14. 14 The perf script
15. 15 And the perf endpoint in VCL
16. 16 Beacon request to /perf
17. 17 fastly-prometheus-exporter /perf
18. 18 Querying the results
19. Next Steps
20. 20 Some more interesting metrics • fastly_perf_first_contentful_paint_ms • fastly_perf_first_input_delay_ms • fastly_perf_dom_ready_ms • fastly_perf_lighthouse_score
21. 21 Performance-based SLOs
22. Thank you! We’re hiring!