Architecting Resilient Azure Platforms: Durable Functions, Cosmos DB, and DR by Design

Failures at Azure scale aren't edge cases — they're constant. Dependencies shift, systems evolve, and small disruptions can cascade in unexpected ways. The real challenge isn't handling failures; it's building confidence that your platform will recover consistently every single time. That's why Bhavna's team built the Resilience Control Platform: to proactively validate and govern system behavior at scale.
In this session, Bhavna, Principal Software Engineer at Microsoft and lead architect of the Resilience Control Platform, walks through the full journey across five chapters — architecture and service boundaries, orchestration with Durable Functions, Cosmos DB data modeling for burst workloads, identity across multiple user realms, and the resilience playbook that keeps the platform running through regional failures.
Every decision is grounded in production lessons, including the mistakes that reshaped the design: tight coupling that became an operability problem, non-deterministic orchestrators, tenant-based partition hotspots, and dual identity systems that forced a home realm discovery pattern.

## Speakers
Bhavana Konchada — Principal Software Engineer and Lead Architect, Resilience Control Platform, Microsoft

## Chapters
00:00 The reliability challenge at Azure scale
01:08 Journey overview: five chapters
02:18 Architecture and service boundaries
03:35 Walkthrough: what happens when a scenario runs
04:59 Lesson learned: from tight coupling to arm's-length services
06:16 App Service vs. Azure Functions: choosing per workload
07:21 Accept Fast, Process Asynchronously
09:09 Durable Functions: orchestration patterns
09:57 The Monitor pattern: timers instead of polling
10:49 Determinism and orchestrator replay
11:44 History management and continue-as-new
12:39 Data layer: designing for Cosmos DB
13:36 Partition key strategy and hierarchical keys
15:00 Data lifecycle with TTL
15:27 Identity as an execution boundary
15:58 Dual identity systems and home realm discovery
17:13 Custom middleware for Azure Functions
18:16 Config-based authorization
19:04 Demo: dual-auth HTTP endpoints
21:10 Resilience: designing for failure
22:03 Multi-region from day one with Azure Front Door
23:30 Cosmos DB: single-write trade-off and automatic failover
24:44 Idempotency from day zero
26:03 Recap: deliberate choices that shaped the system
27:10 Closing thought: design for the world as it behaves

## Key Takeaways
- Tight coupling between services feels elegant but creates operability pain — arm's-length REST boundaries with independent dependencies and deployments are worth the duplication.
- Match the runtime to the workload: App Service for interactive, session-based UIs; Durable Functions for event-driven, long-running orchestrations that shouldn't hold compute.
- Accept Fast and Process Asynchronously (return 202 immediately) decouples user experience from workflow duration and is foundational for scale.
- Durable orchestrators are state machines, not scripts — avoid non-deterministic calls, use orchestration context for time and IDs, and apply continue-as-new to keep replay history bounded.
- Partition Cosmos DB by entity (with hierarchical keys) rather than by tenant to avoid hotspots, and use TTL to manage data lifecycle without cleanup jobs.
- Treat identity as a first-class concern: normalize multiple auth realms (Entra OpenID Connect and WS-Federation) behind a single identity model via home realm discovery and custom middleware in Azure Functions.
- Design for fail-and-continue, not fail-and-recover: multi-region behind Azure Front Door, single-write Cosmos DB with automatic failover, and idempotent operations from day zero.

## Resources
- Azure Durable Functions documentation — https://learn.microsoft.com/azure/azure-functions/durable/ — Stateful orchestrations, the Monitor pattern, determinism, and continue-as-new.
- Azure Cosmos DB documentation — https://learn.microsoft.com/azure/cosmos-db/ — Partitioning, hierarchical partition keys, TTL, and multi-region configuration.
- Azure Functions documentation — https://learn.microsoft.com/azure/azure-functions/ — Isolated worker model, HTTP triggers, and custom middleware.
- Azure Front Door documentation — https://learn.microsoft.com/azure/frontdoor/ — Global entry point, health probes, and automatic regional failover.
- Microsoft Entra ID documentation — https://learn.microsoft.com/entra/identity/ — OpenID Connect, token validation, and identity for cloud apps.

## Tags
Azure, Microsoft Azure, Durable Functions, Azure Functions, Azure Cosmos DB, Azure App Service, Resilience Engineering, Distributed Systems, Multi-Region, Disaster Recovery, Azure Front Door, Microservices, Orchestration, Identity, Microsoft Entra ID, OpenID Connect, WS-Federation, Idempotency, Cloud Architecture, Production Engineering, Application Insights, Azure Monitor, Platform Engineering, MAIS24

Видео Architecting Resilient Azure Platforms: Durable Functions, Cosmos DB, and DR by Design канала ITOpsTalk