Migrating off a fragile stack in three weeks
JournalTorus
Notes from a move we helped plan
Written by Owen Strand, Product Lead
Tuesday, 15 July 2026
A fragile stack is rarely one big problem. It is a hundred small ones that nobody has time to fix. The trick of a calm migration is to move in slices, and to let the new setup prove itself before the old one goes.
Slice by slice
We start with the service that is easiest to move and least risky to break, then the next one. Each slice runs on both sides for a few days while Torus compares their behaviour.
When to switch off the old
Only when the new side has handled a full normal week, including the busy hours, with no surprises. That rule makes the timeline a little longer and the nights a lot shorter.
Our checklist per slice
- Same traffic on both sides
- Same error rate for a full week
- A tested way back
- Costs compared line by line
Key takeaways
- Small slices, fast feedback
- Run both until the new one is boring
- Measure the cost before and after
Notes from a move we helped plan.
