LEARN. PRACTICE. GROW.Concept and mechanism
Availability sets distribute VMs across fault and update domains, but those domains are not equivalent to Availability Zones. To tolerate zone loss, assess zonal distribution, remaining capacity, and dependencies. In a Virtual Machine Scale Set, autoscale rules operate within configured limits. Reaching the maximum may prevent more instances even when the metric requests expansion. Quotas, startup, cooldown, and destination-system limits also influence response; adding frontend capacity does not necessarily solve a database bottleneck.
Guided application
In App Service, slots support preparing and validating a version, but you need to know which settings follow content and which remain with the slot. Connection strings identifying separate environment databases should have deliberate, tested behavior. Managed identities and some settings do not follow the same behavior as app settings. In Container Apps, revisions identify immutable versions; multiple revision mode supports managing traffic among revisions. Define criteria for increasing traffic, failure signals, and rollback limits. Recovering code does not automatically reverse an incompatible data migration.
A scale set with maximum four already has four instances. Before changing the CPU threshold, confirm that the limit and authorized capacity match expected load.
Common pitfalls
Confusing an availability set with zones; ignoring autoscale limits; assuming swap changes all settings identically.
Related topics: Networks and layered diagnosis · Observability and operational recovery
Test capacity and change behavior under conditions operations will need to support.
Reference: App Service deployment slots · AZ-104; skills measured 2026-04-17