Microsoft outage shows how far cloud services are interlinked

Shared infrastructure spread the failure

According to the commentary, many of the servers behind Exchange Online, Microsoft's email and calendar system, dropped out at once. The remaining servers had to carry the full load and became overloaded.

The author argues that applications which appear to users as separate products can rest on the same underlying infrastructure. In this case, Teams and Copilot went down together with email because they are not separate underneath: Teams reads calendar information from Exchange Online, while Copilot reads email. Microsoft said the features that failed were those that depend on Exchange Online.

Recovery attempts and an unresolved cause

When Microsoft returned servers to service, it said they stopped responding again once traffic came back. The company said automated health checks were marking servers as failed and removing them from service, which placed more load on the remaining servers.

Microsoft has not yet confirmed the root cause. Early on it said it had found no recent change linked to the start of the disruption, and hours later it began reversing a recent network configuration change.

Limited risk to stored data

Based on Microsoft's statements, the commentary describes the outage as an access problem rather than a data problem. Mail already in a mailbox should remain there when access returns, and new mail was mostly delayed rather than lost. Microsoft said it extended its email timeout window to 48 hours to protect delayed messages.

The author notes a more immediate concern: fake messages urging people to restore their Outlook account appeared within hours of the outage.

Open questions

• Which explanation for the failure is correct, and how did the network configuration change contribute to it?
• Why did Microsoft's automated safety systems appear to make the disruption worse?
• Will Microsoft publish a final account with dates for its corrective changes?