正在加载页面 · Loading page
Backend · Shenzhen
An on-call rota is a design defect.
Five years of backend, three of them spent on one thing: converting anything that needs a human at 3am into something that can wait until morning. The rest was documentation.
Approach
A page firing means two things went wrong.
The system broke, and nobody knew in advance that it would. The second is almost always the cheaper fix, but it produces no heroics, so it rarely gets made.
My approach is dull: every alert must answer "what happens if nobody acts". The ones that cannot are demoted to a daily digest; the ones that can get a written runbook.
01
Ask the consequence
An alert that cannot say what happens if ignored has not earned the right to wake anyone.
02
Then write the runbook
If the procedure is "ask me", the system has exactly one replica and it is a person.
03
Then delete it
An alert silent for six months is either wrong or obsolete. Both mean delete.
Experience
Where I have worked
Three roles, counting the internship.
Meridian Systems Senior backend engineer
17 → 2
Payment reconciliation. Cut nightly pages from seventeen a week to two, mostly by deleting eleven alert rules.
Feb 2025 — PresentShenzhen
Northgate Retail Backend engineer
2.5 years
Inventory service. Wrote the first runbook, after which a new hire could take on-call unaccompanied.
Jul 2022 — Jan 2025Guangzhou
Kestrel Studio Full-stack intern
3 tools
Three internal tools, two of which are still running.
Jun 2021 — Jun 2022Shenzhen
Tools
Things I have run in production
Production only. Things I have read about but never shipped are not on this list.
Education
Where I studied
Northgate Institute of Technology BEngComputer science
Final project on distributed log collection
Sep 2017 — Jun 2021