How three failures became one major collapse
On Monday, Microsoft published its initial analysis of the outage that hit Outlook, Teams and other Microsoft 365 services. The conclusion: no single factor would have brought the system down on its own, but the three together triggered a chain reaction, as reported by iefimerida and Digital Life.
Charles Lamanna, the company's executive vice president for Copilot, Agents and Platform, explained in a post on the X platform that the origin lay in a latent bug in the code. The flaw had existed since July, without having caused a problem until then.
The second factor was the network gateways, the points through which data passes between services. Microsoft has two. One was not working properly for Teams and almost all traffic went to the other, which had not been designed for such a load.
The third factor was capacity. More than 14% of servers had been taken out of operation, while iefimerida writes that the cause was a different problem that appeared the previous day. The increased load and the lost capacity activated the old bug, requests piled up and the servers stopped responding. The incident fell during peak hours for the US.
Techblog.gr adds more technical detail. According to the report, the bug was in the second gateway system and had arrived with a fix built specifically for Microsoft 365's gateway servers. Under certain load conditions, an internal process got stuck in "lock contention", that is, competition for locks, and the server did not respond. At the worst point, more than 80% of servers were not responding, and in less than two hours the problem had spread everywhere, according to Lamanna.
Why recovery took so long and why testing missed it
Recovery was difficult because every server that came back up was immediately hit by the volume of requests that were waiting. As Lamanna said, the servers were "immediately overwhelmed by the traffic waiting to be served" and went down again. The automatic recovery mechanisms had not been designed for such widespread unresponsiveness.
Because the bug was not recent, Microsoft could not simply undo a last change. The vice president said that the usual rollback procedures, staged rollouts and feature switches did not deliver a quick result. Net Week notes that the company applied a corrective intervention to the authentication and routing systems, began recovery from Asia-Pacific and extended to 48 hours the time delayed emails wait in queues. According to iefimerida, services were finally restored by Saturday.
Techblog.gr also explains why testing did not catch the problem. The bug got worse on machines with more cores and newer hardware, while load tests were mainly run on weaker models. The labs also had different traffic and a different hardware mix from the real servers. The change had rolled out worldwide within a few weeks from July and for more than three months had caused nothing.
Microsoft apologised and insisted there was no cyberattack. According to Techblog.gr, it also denied any link to the thousands of layoffs it had announced the previous month. It is preparing a final report with more details on the causes and the handling, as iefimerida reports.
The impact: from Greece to Japan
Net Week records that the outage, with incident code EX1550543, hit Exchange Online and dragged down Outlook, Teams, Copilot, SharePoint Online and other services. At the peak, Downdetector reports exceeded 130,000 in the US. Affected, among others, were the US House of Representatives, the British NHS, the Euronext stock exchange and Maersk. At financial firms in Taiwan, transactions were confirmed by fax.
In Greece, the Minister of Digital Governance and Artificial Intelligence Dimitris Papastergiou told SKAI that the state's core information systems worked without serious problems. Malfunctions were limited to Outlook, Copilot and Teams in the ministries that use Microsoft servers. The minister said the incident shows the need for greater digital autonomy and alternative national and European infrastructure. He explained that the state already applies a hybrid cloud model: owned infrastructure for critical data and cloud services for less critical data. He also mentioned the role of the "Daedalus" supercomputer in classifying and protecting data.
In Japan, according to Techmaniacs, citing Japanese media, businesses and services reverted to fax machines and landline phones. In Mie Prefecture, where about 7,500 public employees depend on Outlook, the authorities asked citizens to contact them temporarily only by phone or fax. Masanao Ozaki, deputy chief cabinet secretary, confirmed that the ministries' internal network was affected.