10.3The synchronous reliability problem

Now imagine:

POST /onboard-customer

does:

Salesforce
   ↓
Billing
   ↓
ERP
   ↓
Email system

Each takes two seconds.

Best case:

~8+ seconds

Then Billing temporarily fails.

Should the caller wait while Mule retries?

Now:

8 seconds
+ retry delay
+ retry
+ another timeout

You may exceed:

client timeout
load balancer timeout
API gateway timeout
user patience

This is where asynchronous architecture starts becoming attractive.