A rack just lost power. The shop kept selling.

Understand it one step at a time

The short runs these in order in about 53 seconds. Here they are written out — pick any step to jump the short straight to it.

1. 1

   12,000 shoppers. One checkout

   A live e-commerce site on Azure. Orders are landing, nobody is thinking about hardware.

   shop.msdevbuild.com · online

2. 2

   Two servers, two racks

   VM1 sits in rack A, VM2 in rack B. The load balancer splits every request between them.

   load balancer · VM1 50% · VM2 50%

3. 3

   Rack A just lost power

   Not a reboot. Not a patch. The power feed to an entire cabinet is gone.

   ALARM · rack A power failure

4. 4

   Four servers gone at once

   A rack is a shared blast radius. One power feed, one switch, one cooling loop — and everything in it drops together.

   rack A · 0 hosts responding

5. 5

   This is where most sites die

   Half the capacity is gone mid-checkout. You would expect the error page right about now.

   expected: HTTP 503

6. 6

   The health probe fails

   Azure Load Balancer probes every backend on a fixed interval. VM1 stops answering, so it leaves the pool.

   probe failed · VM1 removed

7. 7

   Every request goes to Rack B

   Within seconds the surviving backend is carrying 100% of the traffic. Same URL, same cart, same session.

   failover complete · VM2 100%

8. 8

   The shop never went down

   An entire rack is dark and the checkout is still taking money. That is the whole trick.

   site online · orders flowing

9. 9

   Azure Fault Domain

   That rack has a name. A fault domain is a group of hardware that shares one power source and one network switch.

   the term · fault domain

10. 10

    Separate power, switch and cooling

    Two fault domains share nothing physical. A failure in one has no path into the other.

    FD 0 ≠ FD 1

11. 11

    Azure picks the racks for you

    Put the VMs in an availability set and Azure spreads them across 2–3 fault domains. You never choose a cabinet.

    availability set · FD 0 / FD 1

12. 12

    Now put both VMs in one rack

    Two VMs, same app, same load balancer — but both landed in fault domain 0. It looks identical from the portal.

    VM1 + VM2 · both in FD 0

13. 13

    Same power cut. Site is gone

    One cabinet, one failure, zero surviving backends. The load balancer has nowhere left to send anyone.

    HTTP 503 · no healthy backends

14. 14

    Fault domains fix that at deploy time

    Nothing about the VMs changed. Only where Azure put them — and that is decided before the outage, not during it.

    restored · FD 0 / FD 1

15. 15

    Fault Domain = Hardware Failure Protection

    Rack power, top-of-rack switch, a dead host — spread across fault domains, none of them is your outage.

    SLA 99.95% · availability set

16. 16

    Follow for Azure & Cloud Engineering Tips

    One architecture idea per reel, from the failure backwards. Save this one for your next design review.

    MSDevBuild · Azure & Cloud Engineering