Nordlicht Reisen GmbH, a fictional travel agency, is opening its booking API to partners: hotels and airlines are to retrieve quotas and prices without every partner reaching the booking application directly. The portal retrieves the same price lists thousands of times during the summer wave, and the values that control operations (limits, feature flags) are to be changeable without a new release. Three decisions, one topic: What sits in front of the API, what keeps responses ready, and where do the settings live?

Prior knowledge. Caching in code and the second level with Azure Managed Redis are taught in "K3, lesson 3.9.3"; Azure App Configuration in overview in "K3, lesson 3.2.6"; what an API gateway takes on and the comparison of YARP, API Management and ingress in "K4, lesson 4.12.5". Authentication and secrets are covered in lessons 1.2 and 1.4 of this course. None of that is repeated here. What is new is choosing the tier by requirement and the limits that tip it.

The requirement

All numbers are assumptions of the exercise, not measured. G-01 to G-10 apply (lesson 0.2), especially G-05 (about 400 external users), G-08 and G-09.

No. Requirement Type Metric Weight Source
4.6-01 Partners retrieve quotas and prices through a single API. F 400 partner accounts (G-05); assumption Must Partner management
4.6-02 Every call is checked at the entrance; a partner can be blocked individually. Q The block takes effect in under 5 minutes; assumption Must Data protection
4.6-03 No partner can overload the booking application. Q at most 100 calls per minute per partner; assumption Must Operations
4.6-04 The application in the virtual network is not reachable from the internet. R backends only over the private network Must Data protection
4.6-05 The API survives the failure of an availability zone. Q zone-redundant Should Management (G-09)
4.6-06 Partners may enter the gateway's address in their firewall. F fixed outbound and inbound address Could Partner management
4.6-07 Price lists come from the cache at peak load. Q 95 of 100 retrievals under 300 ms at ten times normal load (G-09); assumption Must Sales
4.6-08 If the cache fails, booking keeps working. Q no error for the customer, only slower Must Management
4.6-09 Limits and feature flags change without a new release, separately per environment. F The change takes effect in under 5 minutes; assumption Should Operations
4.6-10 Configuration changes can be traced back 30 days. Q 30 days of history; assumption Should Operations