Good judgment, stuck in one place
The logic that scored providers worked well, but it lived inside one product and forgot everything on restart.
| What | Before | After |
|---|---|---|
| Provider health | Known in one product | Shared by every product |
| After a restart | Forgotten | Remembered |
| Two copies running | Disagree | Agree |
| Seeing what happens | Digging through logs | Live dashboards |
| Security | Not needed yet | Locked down from day one |
How we did it
- 01
Move it, do not change it
The existing logic was lifted out as it was, so nothing behaved differently.
- 02
Keep decisions local
Each product still decides instantly and shares what it learns in the background.
- 03
Lock it down
Sign-in, a closed network, and restricted permissions from the first release.
- 04
Live the same day
Test environment to production in one day, with no new servers.
One moment from the work
Designed for failure“What if the warning system goes down?”
Every product keeps deciding with what it already knows and catches up when the service returns. The shared view is a bonus, never a dependency.
- Idea to live
- 1 day
- New servers
- 0
- Waits per decision
- 0
What any team can take from this
Separate the move from the fix
Move things as they are first. Improve them one decision at a time.
Keep the fast path local
Coordination belongs in the background, not between a customer and an answer.
Put a number on every limit
Every buffer and timeout has a written size.