We opened Milan and Prague and closed Riga. Closing a zone taught us more than opening one: about export, about telling people early, about the quiet cost of running something half-used. We would rather run nine good zones than twelve lonely ones.
Routine firmware updates on the storage volumes ran overnight across zones. Uploads were briefly slower in a couple of zones during each window. Nothing to do on your side.
Prague is open. We are back to a steady eight-then-nine count after retiring Riga.
We are closing the Riga trial. It never found enough steady use to justify a full zone, and a near-empty zone is harder to keep healthy than you would think. If you have an account there, you have until 30 September to export; after that the zone is retired. There is a short guide on moving your data to another zone. We are sorry to the handful of you who liked it there.
One of us spent two days at a small infrastructure meetup. The useful takeaway was not a talk but a hallway conversation about how other small teams handle disk-full: most refuse writes well before the disk is actually full, with a wide margin, because the failure modes near a full disk are nasty. We already refuse below 3 GB free; the conversation made us comfortable keeping that margin generous.
On 2 October last year a bad release made share pages return errors for eighteen minutes. We never wrote it up properly, so here it is, late: a template change shipped without the share route's data, every /s/ link 500'd, we rolled back in eighteen minutes. Accounts and files were untouched. We added a check that a release serves a real share page before it goes out.
Milan is open.