# The Bug Waited Three Months for a Bad Day

> Big outages are rarely one mistake. They are three ordinary things on the same unlucky day. Test the combination, not just the parts.

- URL: https://hsalem.com/bug-waited-for-a-bad-day/
- Author: Hassan Salem (https://hsalem.com/about/)
- Published: 2026-11-17
- Topics: Devops, Software Engineering, Productivity, News

Last Thursday afternoon, a lot of offices in Europe learned how much of their day lives inside someone else's servers. Outlook would not load. Teams was shaky. People walked over to each other's desks to ask if their email was working too.

The outage itself is not what stayed with me. The explanation is.

## A bug that sat quietly since July

According to Microsoft's preliminary report, the problems started at 11:20 AM Eastern on November 12, which is 17:20 here in Munich, and customer impact only ended at 11:10 AM Eastern the next day. Nearly 24 hours. Outlook, Exchange Online, Teams and Copilot were hit. At the worst point, more than 80% of the servers involved were unresponsive.

Microsoft says it was not a cyberattack and not related to AI. It was three ordinary things meeting on the same day:

1. A software bug from an update introduced in July. It had been running for more than three months without any issue.
2. About 15 hours before the outage, Teams automatically failed over to the network infrastructure Exchange Online uses, so those servers got more traffic.
3. Nearly 14% of that gateway capacity was already offline because of an unrelated issue.

Any one of them alone is a normal Tuesday. Together, they took down email for a big part of the working world.

And then the line every engineer should read twice. Microsoft's load tests did not catch the bug because they "tended to focus on weaker hardware with fewer cores", while the problem got worse on newer machines with more cores. The lab was not the same as production. It rarely is.

> Three months of green does not prove the code is safe. It only proves the bad day has not come yet.

## The safety net was part of the fall

What I find most interesting is the failover. Failover is the thing we build so that we can sleep at night. Here it worked exactly as designed, and it moved extra load onto servers that were already short on capacity and carrying a hidden bug.

I think many of us test our systems on their good days. We test the happy path, then we test one failure at a time. One node down. One region down. One slow dependency. Real incidents rarely send failures one at a time. They come in groups, and they are usually polite enough to wait until the people who wrote the code have moved to another team.

If you have read a few postmortems, you know the phrase "a combination of factors". It is not an excuse. It is just how complex systems break.

## What comes next

Microsoft says it has added capacity, improved monitoring and added safeguards so that traffic redirection does not cause the same kind of cascade again. A full post-incident report is still to come, and I will read it with coffee.

I also expect this week to show up in many IT budget meetings across Germany. We already talk a lot about digital sovereignty and depending on a few big providers. Nothing makes that conversation more real than a day without email.

## A few things worth doing this week

1. **Test failures in pairs.** In your next game day, take some capacity away first, then trigger a failover. See what the second step does to the first.
2. **Write down how staging lies to you.** Hardware, data size, traffic mix. One honest page. You can't fix every difference, but you should know which tests to trust less.
3. **Load test your failover path, not just your main path.** The backup route usually gets the least traffic in tests and the most traffic on the worst day.
4. **Ask "what else is already broken?"** in incident reviews. Small known issues are quiet until they are not.
5. **Have a plan for when the tools are gone.** Know how your team reaches each other without chat or email. A phone list sounds old fashioned until you need one.

The lesson I take from this one is simple. Most of the time, it is not one big mistake. It is three small, reasonable things on the same unlucky day. So stop asking only "does this work?" and start asking "what else might be true at the same time?"

Sources: [BleepingComputer: Microsoft says months-old bug caused global Microsoft 365 outage](https://www.bleepingcomputer.com/news/microsoft/microsoft-says-months-old-bug-caused-global-microsoft-365-outage/), [SecurityWeek: Microsoft 365 outage disrupts Outlook, Exchange Online and Teams](https://www.securityweek.com/microsoft-365-outage-disrupts-outlook-exchange-online-and-teams/), [TechRadar: Microsoft outage live](https://www.techradar.com/news/live/outlook-outage-nov-2026)
