What Happens when Service Mesh Maintainers Get a Taste of Their Own Mesh?
summary
Linkerd is deployed to run a new SaaS (Dive) and maintainers become on‑call users, uncovering concrete failures: route metrics vanish because an nginx config strips the Linkerd header, GraphQL slowdowns push adoption of distributed tracing instead of unavailable tap bodies, weekly edge upgrades reveal upgrade and proxy‑version workflow pain leading to linkerd check --proxy now listing all out‑of‑date pods, checks hang on evicted pods fixed, an admin‑server timeout breaks Prometheus scrapes, and paired Wireshark debugging plus a staging cluster produce reproducible fixes and tests.
AI-written summary · about
chapters (13)
- 0:00Intro
- 0:15Risha Mars
- 3:00Why a Service Mesh?
- 3:44Investigate high latency
- 5:12Debugging slow requests
- 6:24Distributed tracing helps! Sweet!
- 7:29linkerd check is our friend
- 8:19Upgrading the proxies is tedious
- 9:27Can't upgrade linkerd due to a render error
- 10:05The case of the hanging linkerd check
- 11:52Dogfooding Dive
- 14:05Sharing (clusters) is caring
- 15:24Conclusion
