AWS Bahrain Data Loss, a Google London Zone Outage and AWS's Sovereign Cloud Test: What Commentators Say About Who Owns Cloud Resilience

Background

Cloud providers describe regions and availability zones as the unit of resilience, and customers often read that as a guarantee. Several recent incidents tested that reading from different directions, and the commentary around them keeps returning to the same question: who owns resilience, the provider or the customer?

The most severe case is the loss of data in AWS’s Middle East regions. As InfoQ reported on September 21, AWS told customers in a status update that it was unable to restore access to resources and data hosted exclusively in the mec1-az2 availability zone of the UAE region (me-central-1). For Bahrain (me-south-1), InfoQ quotes AWS as saying the damage spanned multiple availability zones and “exceeded what our regional and multi-AZ services are designed to withstand”, and that it could not restore resources and data hosted exclusively in that region. InfoQ adds that AWS had earlier advised customers to replicate critical data to other regions. The Herald covered the AWS statement at the time.

Commenters on Hacker News, as summarised by InfoQ, divided over who should have known. Some pointed to the shared responsibility model; others argued that vendors’ marketing had given executives a different impression. One commenter, in InfoQ’s account, described AWS as a toolbox rather than a ready-made solution for redundancy against every catastrophe.

Current situation

Three later events differ in cause but touch the same assumptions.

Azure gateways, September 30 to October 1. The Register reported that Microsoft warned users of ExpressRoute Gateway, VPN Gateway and Azure VMware Service of degraded or interrupted connectivity from 20:30 UTC on September 30, with 18 Azure regions affected at the time of writing. Microsoft’s own update, quoted by The Register, said that some supporting network management components “have not recovered automatically” in a subset of regions, and that some affected VPN gateways had reduced redundancy rather than a complete loss of connectivity. The Herald’s earlier report covers the incident in detail.

Google Cloud, London, October 20. According to The Register, Google Cloud blamed a “power event” in the europe-west2-b zone, with unplanned downtime beginning at 0608 UTC. Affected services included BigQuery Data Transfer Service, Cloud Filestore, Cloud Run, Google App Engine, Google Cloud NetApp Volumes, Google Kubernetes Engine and Managed Service for Apache Kafka. Per The Register, Google said customers might not be able to create new workloads while it prioritised recovery of existing ones, and later said the issue was mitigated for most customers at 14:03 UTC, with an analysis to follow. In a companion report, The Register cited Down Detector as showing customers caught up including Lloyds Bank, HSBC, Royal Mail, Rightmove, DPD, BT, Screwfix and Revolut, and said Google’s guidance for a workaround was that an impacted customer “may try to migrate their workload impacted to another available location” if feasible. The same report quotes a Reddit user as saying that “the classic one zone takes a nap and your non-HA service is in it”.

The Register also quoted Dr Amro Al-Said Ahmad, a lecturer in computer science at Keele University, who estimated that 38 services were down. He told the outlet that “resilience is not achieved by simply moving to the cloud” and that true resilience “requires designing for regional failures through multi-region deployments, workload portability, and tested disaster recovery procedures”.

AWS European Sovereign Cloud, October 24. In a blog post, AWS announced that on Saturday, October 24, 2026, the AWS European Sovereign Cloud would run for several hours without a connection to the AWS Global Network backbone, and that its team would also disable a dedicated system used for limited transfers of operational data. AWS says neither is operation-critical and that the exercise would not affect service availability within the region, other AWS Regions or Direct Connect, though customers might see brief disruptions as traffic moved routes. AWS says it will share an update with regulators and customers after the exercise. The sources the Herald read on October 27 do not report the exercise’s results.

InfoQ’s coverage records divided reactions. Fadi Dasus, a senior cloud security architect, separated a technical question (does the service stop if the link to the US is cut?) from a legal one, and concluded: “So, this is real progress for resilience. It just isn’t sovereignty.” Lars Larsson Lundsten, field CTO at Elastisys, which sells a European alternative to US cloud platforms, wrote that “You can’t solve legal problems with technical solutions”, while crediting AWS with being careful about what it claims. Odd-Arne Haraldsen, who runs IT operations at Svenljunga Kommun in Sweden, asked how long the region could operate independently before authentication tokens, certificates or licences expire, concluding that “Redundancy is not the same as independence, and continued operation is not the same as sustained autonomy.” InfoQ itself notes that AWS’s post does not say what happens to global services while the backbone is down.

Implications

The following draws on what the sources say; the Herald makes no forecast.

Zone and region guarantees are described as building blocks. In an InfoWorld column published October 23, David S. Linthicum writes that “Resilient architecture must be designed, funded, tested, and owned”, and that the provider is responsible for the reliability of the platform while the enterprise is responsible for how its applications use it. He describes cloud services such as multiple regions and zones, managed databases and replication tools as “building blocks” rather than a finished resilience strategy. This is the same position Keele’s Ahmad took in The Register, and the one HN commenters took in the Bahrain thread, as InfoQ reports it.

Dependencies are the recurring theme. Linthicum lists hidden dependencies in a typical stack: a regional database, an identity provider, DNS that may need manual intervention, monitoring that goes dark. Ahmad describes failures propagating through “an entire digital supply chain”. Haraldsen’s questions about tokens, certificates and external control planes apply the same lens to a sovereign-cloud test. The Azure incident, as The Register reports it, involved supporting management components that did not recover automatically.

Multi-cloud is treated cautiously. Linthicum writes that the reflexive response to outages, “We need multicloud”, is sometimes right but “Often, it is an expensive distraction”, and argues for workload-specific resilience: some workloads need multi-region deployment within one provider, others active-passive failover or reliable backup and restore. Ahmad, by contrast, lists workload portability among the requirements.

Testing is the shared remedy, and the sources disagree about what a test proves. Linthicum writes that “A resilience plan that has not been tested is just a theory”, and recommends measuring actual recovery time. AWS’s planned exercise is a test of that kind at provider level. Dasus and Lundsten, in InfoQ’s account, argue it speaks to technical independence and not legal exposure, and Haraldsen questions whether a few hours is long enough to establish autonomy. A second InfoWorld column, published October 27, makes a related distinction: “data may be local, but control may not be”.

What the sources do not establish. None of the sources reviewed quantifies how many customers lost data in Bahrain, or how many Google customers had single-zone deployments. Google has said it will publish an analysis of the London incident and, as of the Herald’s reading, had not done so in the sources reviewed. The results of AWS’s October 24 exercise were also not available in those sources. Commentary quoted above comes from practitioners, an academic, a columnist and online commenters; one of them, Elastisys, sells a European alternative to US cloud platforms. The Herald has not verified their claims independently.