حالة الشبكة

2026-07-19

Madrid

A peer withdrew a prefix without notice and traffic fell back to transit. Median RTT rose from 3 ms to 19 ms for Iberian eyeballs. Resolved by pinning a backup session; no packet loss at any point.

2026-05-02

Singapore

Announced eleven days in advance. Instances on the affected rack lost network for under two minutes each during the LACP failover; no reboots were required.

2026-02-11

Frankfurt

A firmware bug in a Gen4 NVMe batch caused two drives in the same RAID-10 mirror to drop within four minutes. 218 instances were affected. All were restored from the cluster's own replica with no data loss; the batch has since been replaced fleet-wide. Full write-up published on 13 February.

2025-11-30

Amsterdam

The largest attack we have absorbed. Scrubbing engaged in 3.1 seconds. The targeted instance saw 40 seconds of degradation; no neighbouring instance was affected. Capacity was doubled the following month.

2025-09-08

New York

The facility lost both utility feeds during a regional event. Generators carried the load for 4 h 12 min with no customer impact. Logged here because we log everything.

2025-06-21

Tokyo

Live migration used where possible; 6% of instances required a short pause. Zero rollbacks.

2025-03-14

Global

A migration on the billing database took a table lock during business hours. The console and API were slow; running instances were unaffected. Migrations are now gated to a maintenance window.

2024-12-02

Bucharest

Construction work severed both paths of a supposedly diverse pair. We now buy diversity with a documented route, not a promise, at every site.

2023-08-17

Amsterdam

The last Milan-generation hosts in Amsterdam were retired over four nights. 94% of instances live-migrated with no reboot; the remainder saw a pause under ninety seconds. Nobody was repriced.

2022-07-19

Frankfurt

Outside air reached 39.3 °C and one of three chillers derated. We shed clock speed on 1,400 instances rather than shed load, so nothing went offline but CPU-bound workloads ran roughly 12% slower for three hours. Free-cooling capacity in FRA6 was uprated the following spring.

2021-04-06

Warsaw

A transit provider leaked our prefixes to a peer that re-announced them with a shorter path. Warsaw traffic took a 180 ms detour and a third of it blackholed. Fixed by withdrawing that session. Every session has carried RPKI origin validation and a strict prefix limit since.

2019-11-12

London

A fabric fault at the exchange caused intermittent loss on peered routes. Traffic was moved to transit within nine minutes of detection; the residual impact was latency, not loss. We have peered at both LINX LON1 and LONAP since.

2017-05-03

Singapore

Singapore opened with one transit provider while the second contract was still in legal review. That provider had a card failure and the site went dark. It was avoidable and it is the reason no site has opened with fewer than three upstreams since.

2016-02-24

Amsterdam

Repeated session resets against a large eyeball network caused traffic to oscillate between peering and transit. Resolved with the peer by raising hold timers on both sides.

2014-12-19

Frankfurt

Six weeks after the first instance was sold, a rack PDU failed and took two of our four racks with it. We had installed it, we had not load-tested it, and we had no spare on site. Eleven customers were affected and all eleven were refunded in full. Every rack has carried A+B feeds from separate PDUs since that night.