Explain like I'm five · SRE

What is Multi-region?

Multi-region, told as a story about building a twin park in another city.

FLOOD!guests everywhere can't get inone city floods, and the whole park shuts its doors

One city floods, and the whole park shuts its doors.Guests everywhere can't get in.

only one city, so that's it...if the park exists in only one city, trouble there means it's over

If the park exists in only one city, trouble there means it's over.With only one, there's nowhere else to go.

CITY ACITY B (twin)log replicatedbuild a twin park in another citywhen one has trouble, guests get sent to the other

Build a twin park in another city.When one has trouble, guests get sent to the other.

Build identically in two citiesa true twin park
Replicate the log toothe same guest log on both sides
If one stops, send guests the other wayguided there automatically
Both can take guests at onceactive-active or active-passive
CITY A (flooded)CITY Breplicationguests get sent to city Bwhen one city stops, the other takes the guests

When one city stops, the other takes the guests.The log was already replicated, so guests can be redirected right away.

$$$$building and keeping up both costs twice as muchso only the services that truly need it get multi-region

Building and keeping up both costs twice as much.So only the services that truly need it get multi-region.

In one breath

Multi-region = building a twin park in another city and keeping their logs replicated to each other, so that when one city has trouble, guests get redirected to the other. The cost of running both, though, is double.

A strategy of running the same service in geographically separate regions (data centers) too. Replication keeps data in sync across regions, and failover switches to another region when one has trouble. More distant regions mean more latency, and data sovereignty — which country the data may sit in — has to be considered too.

When grown-ups say it

Multi-region
Keeping twin parks in more than one city. If one city stops, there's always another.
Active-active
A setup where both cities take guests all along. Since both are already working, losing one is no big deal.
Active-passive
A setup where only one city takes guests normally. The twin city just waits, idle, most of the time.
Latency
How long it takes to go back and forth, the farther the city. A distant city is slower than a nearby one.
Data sovereignty
The question of which country a guest log is allowed to sit in. Choosing a city (region) also means checking the law there.
Disaster recovery
The plan for reopening even if one whole park collapses. Multi-region is a classic way to achieve disaster recovery. → how to open the morning after the fire
Replication
Keeping both cities' guest logs identical. If replication lags, the city you switch to may have a slightly older log.
Failover
The act of switching to the other city when one stops. Multi-region is failover scaled up to a whole city. → when the generator dies, the spare takes over
Next story
Building the Whole Park from One Blueprint →
Want another term as a picture book? Request →