Azure is rebooting your server. Nobody notices.

Understand it one step at a time

The short runs these in order in about 64 seconds. Here they are written out — pick any step to jump the short straight to it.

1. 1

   One server. Every order goes through it

   The shop is live and selling. Right now nothing about this looks like a problem.

   web server · healthy

2. 2

   Azure has to patch the host

   Security updates land on the physical host underneath. Your VM reboots — and it is not optional.

   planned maintenance scheduled

3. 3

   Server Unavailable

   Problem: server maintenance causes downtime. Every shopper hits an error page and leaves.

   HTTP 503 · site offline

4. 4

   Put two VMs in an Availability Set

   VM1 and VM2 run the same shop. The set is Azure’s promise that it will never take both down at once.

   availability set · 2 VMs

5. 5

   One address in front of both

   Azure Load Balancer owns the public IP and health-probes each VM before it sends a single request there.

   load balancer · 2 backends

6. 6

   Fault domains and update domains

   Different rack, power and switch. And Azure only ever reboots one update domain at a time.

   FD 0 / FD 1 · UD 0 / UD 1

7. 7

   Requests start arriving

   Shoppers hit one public address. They have no idea how many servers are behind it.

   traffic incoming

8. 8

   The load balancer splits the traffic

   Roughly half to VM1, half to VM2 — and it keeps probing both the entire time.

   VM1 50% · VM2 50%

9. 9

   Both green. Orders going through

   Two servers doing the work of one. The spare capacity is a bonus, not the point.

   VM1 healthy · VM2 healthy

10. 10

    Azure starts maintenance on VM1

    Update domain 0 goes first. VM1 drains and reboots — update domain 1 is not touched.

    UD 0 · updating

11. 11

    Traffic Automatically Redirected

    The health probe fails, the load balancer drops VM1 from the pool, and VM2 takes 100%.

    failover · zero manual steps

12. 12

    The shopper never finds out

    Same URL, same cart, same checkout. No error page, no support ticket, no abandoned basket.

    site online · orders flowing

13. 13

    VM1 rejoins. Only then does UD 1 go

    The probe passes, traffic rebalances — and Azure waits for that before touching VM2 at all.

    rebalanced · 50% · 50%

14. 14

    High Availability Achieved

    Two VMs, one availability set, one load balancer. Azure backs it with a 99.95% SLA.

    SLA 99.95% · availability set