Everything is reporting normally

Checked from three networks outside our own, every sixty seconds. Uptime below is computed from the incident minutes, not copied from the SLA.

Incident history Uptime SLA

Components up: 10 of 10
Last checked: 02:32 UTC
Uptime, 90 days: 99.894%
Since last event: 6 days ago

Right now

Ten components, each with its own measurement

“Operational” on its own says very little. Each line carries the figure the check actually produced, so a component that is technically up but slow cannot hide behind a green dot.

All systems reporting normally

Last checked at 02:32 UTC · no unplanned event in 6 days ago · 99.894% over the last 90 days

Virtual servers Operational 39 hypervisors reporting

Dedicated servers Operational All chassis reporting

Network — carrier A Operational BGP established · 0.4 ms jitter

Network — carrier B Operational BGP established · 0.4 ms jitter

DDoS filtering Operational Armed · no active mitigation

Customer panel Operational 95 ms median response

Public API Operational 48 ms median response

Provisioning Operational 62 s median to root

Payments Operational All 8 networks confirming

Support desk Operational 10 min median first reply

Medians over the last five minutes, from probes on three networks that are not ours. A component reads degraded the moment its median crosses twice its normal figure — we would rather over-report than round a bad ten minutes down to green.

Computed, not claimed

Three windows, and the minutes behind each one

A percentage on its own can be chosen. Each figure here is the arithmetic on the incident list below it — change the list and the percentage moves, which is the only version of this number worth publishing.

99.743% · 30 days

111 minutes of unplanned downtime, itemised below with its cause. Planned maintenance is excluded from the arithmetic and listed separately.

99.894% · 90 days

137 minutes of unplanned downtime, itemised below with its cause. Planned maintenance is excluded from the arithmetic and listed separately.

99.954% · 365 days

241 minutes of unplanned downtime, itemised below with its cause. Planned maintenance is excluded from the arithmetic and listed separately.

Everything that happened

Incident history, including the ones we caused

A history containing only planned maintenance is not a history. Each entry gives what broke, how long it lasted, and what changed afterwards so it would not happen the same way twice.

2026-08-22 · 6 days ago

Packet loss towards IstanbulPartial · 111 min

An upstream of carrier A blackholed one of our prefixes in error. Withdrawn from that carrier after 51 minutes of diagnosis; full reachability an hour later. Traffic to every other city was unaffected throughout.
2026-07-08 · 2 months ago

Provisioning queue stalledPartial · 26 min

New orders queued instead of building. Existing servers were unaffected. The cause was a full spool on the image host, which now alarms at 70% rather than at 95%.
2026-05-05 · 4 months ago

API rate limiter too aggressivePartial · 47 min

A tightened limit rejected legitimate bursts from three customers building fleets. Rolled back within the hour and replaced with a per-token burst allowance.
2026-03-09 · 6 months ago

Panel slow, servers unaffectedPartial · 38 min

A migration left an index missing on the invoice table and the panel crawled. Hypervisors, network and API were untouched. Credits were issued to every account that opened a ticket, without asking them to prove impact.
2025-12-28 · 8 months ago

Filtering engaged, 340 GbpsPartial · 19 min

A volumetric attack against one customer address was absorbed upstream. Filtering added roughly 2 ms for the duration. Nothing was null-routed and nobody was asked to leave.

Entries older than this are kept and served on request rather than published indefinitely — a page that keeps three years of history is a page nobody reads to the bottom. Credits for anything unplanned are applied under the uptime SLA without you having to claim them.

How this page is produced

Four decisions that make the numbers worth reading

Every one of them makes our figures look slightly worse than the alternative, which is how you can tell they were not chosen to flatter.

Probed from outside our own network

From three machines on other providers, in Frankfurt, Amsterdam and New York. A status page probed from inside the datacentre it monitors is a page that goes green during an outage.

A component is degraded before it is down

Slower than normal is a state we publish, not one we round down to “operational”. Most of the entries below are degradations rather than outages, which is what an honest history looks like.

Uptime is computed, not asserted

The percentages above come from the incident minutes listed below. Change the history and the percentage changes; there is no separate number we could quietly keep flattering.

This page is not hosted in Chișinău

It runs somewhere else on purpose, so that the building losing power does not also take away the page that would tell you so.

What a bad month costs us

Downtime is credited, not argued about

The bands are published, the credits are applied automatically, and the exclusions are narrow enough to read in one sitting.

The uptime SLA How the network is built