Explain like I'm five · SRE

What is Failover?

Failover, told as a story about a spare generator that automatically takes over when the main one dies.

MAIN GENERATORno spare, so this takes forever...the main generator died and the whole park went dark — with no spare, it takes forever

The main generator died and the whole park went dark.With no spare, it takes forever.

only one generatoreverything connected stops togetherdepend on just one thing, and the moment it stops, everything stops

Depend on just one thing, and the moment it stops, everything stops.Everything connected stops together.

MAIN (down)SPARE (now on)when the main generator dies, the spare automatically takes overnormally off, the spare switches itself on the moment it's needed

When the main generator dies, the spare automatically takes over.Normally off, the spare switches itself on the moment it's needed.

Normally keep the spare offit costs money to run
Automatically detect the failurethe moment the main one dies
Switch over automaticallymanual is too slow
Test the spare now and thenuseless if it won't turn on
AUTO-SWITCH STEPS① main found down② signal sent to spare③ spare takes over④ announce when it's backMAINSPAREdetect, signal, switch — the three happen automatically, one after another

Detect, signal, switch — the three happen automatically, one after another.The moment the main one stops, the spare takes over right away.

huh, did it just stop for a second?that brief instant of switching can still be feltan untested spareskip the occasional test, and the spare might be broken tooa true failover has to account for both

That brief instant of switching can still be felt.Skip the occasional test, and the spare might be broken too.

In one breath

Failover = keeping a spare off until it's needed, then automatically detecting and switching to it the moment the main one dies, to keep the service going. The switch is never perfectly invisible, but testing it now and then confirms the spare actually works.

A mechanism that automatically switches to a standby component when one fails. It can be active-passive (the spare waits idle) or active-active (both take traffic), and the time a switch takes is called the failover time.

When grown-ups say it

Failover
Switching to the spare when the main generator dies. It only works if it happens automatically and fast.
Active-passive
The spare just waits, idle, most of the time. It only switches on once the main one stops.
Active-active
Both take guests at the same time, all along. If one stops, the other is already running.
Automatic switch
Switching over by itself, with no one pressing a button. A manual switch is slower by exactly that much.
Failover time
How long the switch takes from start to finish. Zero seconds is hard — that brief instant can still be felt.
Health check
Continually checking whether the main generator is still alive. Without it, nothing can automatically notice the main one is down.
Multi-region
Keeping a whole spare city, not just a spare generator. Failover scaled up to the size of a city. → the twin park
Replication
Keeping the same log copied over to the spare side too. A spare that takes over with no log is useless.
Next story
A Curtain Drawn Over a New Ride →
Want another term as a picture book? Request →