JupiteX Get the app
Science & Technology17 Nov 2026 · about 6 min

Microsoft blames major outage on a July update defect – and two other factors

The brief

Microsoft traced the outage to a dormant defect in an update deployed in July. Under certain loads, lock contention inside an internal process could make a server unavailable. The defect had gone unnoticed for more than three months because the update initially deployed without issues. Two conditions raised the risk of triggering it. A surge in traffic followed a Teams failover between gateway systems, changing traffic patterns. At the same time, 14% of networking-gateway capacity was out of service because of an unrelated problem from the previous day. Together, those conditions exposed the defect and overloaded servers. At the outage’s low point, over 80% of servers were not responding. Microsoft’s traffic management then redirected users to remaining healthy endpoints, adding load there and helping the failure spread. The outage affected Microsoft 365 services in 74 countries.

01

What combination of problems caused Microsoft’s outage?

Microsoft traced the outage to a dormant defect in an update deployed in July. Under certain loads, lock contention inside an internal process could make a server unavailable. The defect had gone unnoticed for more than three months because the update initially deployed without issues.

Two conditions raised the risk of triggering it. A surge in traffic followed a Teams failover between gateway systems, changing traffic patterns. At the same time, 14% of networking-gateway capacity was out of service because of an unrelated problem from the previous day.

Together, those conditions exposed the defect and overloaded servers. At the outage’s low point, over 80% of servers were not responding. Microsoft’s traffic management then redirected users to remaining healthy endpoints, adding load there and helping the failure spread. The outage affected Microsoft 365 services in 74 countries.

02

What is lock contention, and how can it make a server stop responding?

Lock contention happens when multiple operations need the same protected resource at once. A lock controls access so operations do not interfere with each other. As general computing knowledge, when one operation holds a lock, others may have to wait; heavy or prolonged waiting can create a bottleneck.

Microsoft said its dormant defect involved lock contention inside an internal process. Under certain load conditions, the contention caused a machine to become unavailable. The article does not describe the specific lock or code path, so it does not establish whether operations deadlocked or exactly how long they waited.

The issue had been present since a July update, but Microsoft said it deployed worldwide without problems for more than three months. A traffic surge and reduced gateway capacity later made the risky load conditions more likely. Servers eventually stopped responding, and Microsoft said over 80% were unresponsive at the outage’s low point.

03

How widespread was the failure, and how much gateway capacity was already unavailable?

The outage affected access to Microsoft 365, Copilot, and Outlook in 74 countries, according to Downdetector. Reports began Thursday morning, and the disruption continued into Friday for some customers. The article does not give a total count of affected people.

At the outage’s low point, more than 80% of the servers were not responding. That figure describes server response, not the share of customers or countries affected. Microsoft also said that nearly 14% of networking-gateway server capacity was unavailable before the main outage began.

The missing gateway capacity came from an unrelated issue on the previous day. Along with a traffic surge and changed traffic patterns after Teams failed over between gateway systems, it increased the chance that the dormant defect would be triggered. Microsoft later deployed a fix gradually across groups of servers.

04

How did redirected traffic turn a problem on some servers into a cascading failure?

Redirecting traffic is a way to keep users connected when some servers become unavailable. Microsoft said its traffic management systems automatically sent users to other healthy endpoints after servers lost connectivity. That approach can help when enough capacity remains to take the extra requests.

During this outage, the servers that remained available received redirected traffic on top of the existing demand. Microsoft said this increased the likelihood of the same problem occurring on those endpoints. As a result, more servers became unresponsive, leaving still fewer places to handle requests.

Microsoft described the result as a cascading failure, with the same amount of load spread across fewer servers. At the outage’s low point, over 80% of servers were not responding. The added traffic also slowed recovery: waiting requests hit servers as soon as they came back, and made them unresponsive again quickly.

05

Why did Microsoft’s testing fail to catch the dormant defect before it affected customers?

Microsoft said its testing did not reveal the dormant defect because its labs used different traffic and SKU mixes from those seen in the outage. A test that does not reproduce the relevant combination of load and systems may not expose a defect that appears only under certain conditions.

Lamanna said Microsoft’s load testing tended to focus on weaker hardware SKUs with fewer cores. The July update, meanwhile, had deployed worldwide without issues for more than three months. Later, a surge in load and changed gateway traffic patterns made the latent lock-contention problem more likely to occur.

This gap matters because Microsoft’s usual safeguards did not quickly reverse the issue: the defect was months old, and normal rollback, flighting, and feature flags could not quickly undo it. Microsoft said it would publish a full post-incident review later that week. The article does not report additional testing changes or what that review would conclude.

06

Why could Teams and some other services recover sooner than Exchange Online?

Microsoft said Teams, OneDrive, SharePoint, and Office apps could fail out of the faulty network-gateway infrastructure. Exchange Online could not. That difference helps explain why some services recovered sooner: they could operate outside the affected gateway infrastructure, while Exchange Online could not take that route.

Teams was among the services that returned earlier. Lamanna said Teams chats and meetings functioned without issue for most of the outage. The article does not say that every Teams feature or user remained unaffected; some people reported trouble with Teams when the outage began.

Recovery was still difficult. Waiting traffic hit servers as soon as they came back, making them unresponsive again quickly. Microsoft said its automated recovery processes were not designed for such a broad unresponsive state. The company applied a fix gradually to groups of servers, and said a fuller post-incident review would follow.

07

How do cloud systems distribute traffic across servers, and why can that approach fail when too many servers become unavailable?

Cloud systems commonly distribute requests across multiple servers so no single machine has to handle all the work. They can also direct users to other available servers when one becomes unhealthy. This is established computing practice; the article specifically says Microsoft’s traffic management automatically routed users to healthy endpoints after connectivity was lost.

That approach can fail when too many servers become unavailable or the remaining machines lack room for the added requests. Microsoft said redirected traffic raised the chance of the same defect occurring on other endpoints. The original problem therefore spread as servers became unresponsive and the same amount of demand was shared by fewer machines.

At the outage’s low point, over 80% of servers were not responding. Microsoft also said waiting traffic quickly overwhelmed servers as they returned, slowing recovery. It deployed a fix gradually across groups of servers, but the article does not describe any later changes to its traffic-distribution system.

This brief was written by AI from the original reporting and checked by other models. Names, figures and quotes come from the source; read it for full context.

Read more in the JupiteX app

Pulse is free. New stories every 4 hours, each one broken into the questions that explain it.

Or read more news on the web