Marin Ashworth-Vale

I keep large, boring systems boring. My work sits between the platform team and the people who depend on it — capacity, failure modes, the pager, and the slow removal of whatever keeps waking everyone at four in the morning.

Ten years across payments infrastructure, logistics tooling and a mid-size hosting provider. I am most useful in the eighteen months after a company outgrows its first architecture: when the traffic is real, the on-call rotation is thin, and nobody has had time to write down how anything actually works. I write the runbook, then I make it unnecessary.

• Retiring the shared database

  Halverd Freight · 2024–2026

  Context

  Eleven services wrote to one Postgres primary. Every schema change needed a cross-team meeting, and a single slow query could stall dispatch for the whole fleet.

  What I did

  Mapped the real access patterns with query sampling rather than the architecture diagram, which turned out to be wrong in four places. Split the three heaviest domains onto their own instances behind a thin routing layer, moved them one table at a time with dual writes, and deleted the routing layer once the last reader moved.

  Outcome

  Dispatch latency stopped correlating with the billing batch job. Schema changes became a team-local decision. No migration window exceeded four minutes and none were rolled back.

• An on-call rotation people would join

  Halverd Freight · 2025

  Context

  Six engineers carried the pager for forty services. Two were close to leaving over it and the alert rules had not been reviewed since they were written.

  What I did

  Audited a quarter of alerts against what an engineer could actually do at three in the morning. Deleted the ones with no action, routed the rest to the team that owned the code, and wrote a one-page runbook per remaining page. Introduced a standing review where every page from the previous week either gets a fix or gets deleted.

  Outcome

  Overnight pages fell to roughly one a fortnight. The rotation widened to fourteen people because joining it stopped being a punishment.

• Deploy tooling for a team that feared deploys

  Corbel Pay · 2021–2023

  Context

  Releases were weekly, manual, and took two engineers an evening. Because they were rare they were large, and because they were large they were frightening.

  What I did

  Built a progressive rollout path — staged environments, health-gated promotion, one-command rollback — and, more importantly, spent six weeks pairing with each team through their first ten self-service deploys rather than announcing the tool and walking away.

  Outcome

  Releases moved from weekly evenings to daily, in working hours. Rollback became a normal event instead of an incident.

• BSc Computer Science

  Ostrey University — distributed systems and numerical methods

  2012 — 2016

• Certified Kubernetes Administrator

  Renewed 2025

  2019

• WritingAn occasional newsletter on operational practice, mostly about the parts of incidents that happen before the alert fires.
• SpeakingComfortable presenting internally and at regional meetups; have run a half-day workshop on incident command twice.
• MaintainingCo-maintainer of a small open-source Postgres migration helper used mainly by teams doing zero-downtime column changes.