Proving application resilience on Azure with Chaos Studio

Key Takeaway: Azure Chaos Studio enables businesses to assess application resilience by simulating system failures, network issues, and disruptions before they affect live environments.

You can’t truly know if your application is resilient until you put it to the test. It’s far better to discover issues by intentionally breaking things in a controlled test environment rather than facing a production failure. Azure Chaos Studio is our reliable service designed for this purpose — safely and with intention.

Currently, Azure Chaos Studio Workspaces is available in public preview. This feature adopts a scenario-driven method, allowing you to test real-life failures that Azure customers experience in production. We’ve dedicated time to ensure that Workspaces are user-friendly, with extensive fault support and named scenarios that closely reflect actual outages instead of random faults.

Why Designing for Resilience Isn’t Enough

Azure users have made significant investments in resilient architectures: deploying across multiple zones, using geo-redundant storage, enabling automatic database failover, utilizing retry logic, and implementing load-balanced frontends. However, the crucial question is: when a failure occurs, can those mechanisms recover your application as quickly as you expect?

Real outages don’t adhere to architectural diagrams. A zone-redundant deployment might fail due to a health probe misconfiguration from years prior. A database with automatic failover could leave your application non-functional if a connection string is hard-coded to a specific region. Additionally, geo-redundant storage can momentarily deliver outdated reads that the application code isn’t prepared for. These are typical errors and they only surface during actual failures.

Reliability and resilience on Azure are a shared responsibility. Microsoft manages the platform and the resilience built into Azure services. Meanwhile, customers must configure that resilience and write the code that leverages it. No single layer can compensate for shortcomings in another. The only way to confirm whether your architecture and application logic can withstand production conditions is to examine their performance under failure conditions before the outages hit you.

How Chaos Studio Workspaces Enhances Resilience Testing

Chaos Studio is Azure’s managed chaos engineering service that validates how applications respond to failures. It allows you to simulate controlled disruptions across infrastructure, network, databases, and application dependencies, helping teams to identify resilience gaps before customers encounter them. Chaos Studio Workspaces specifically centres around scenarios that resemble real-world events thus offering a more practical testing environment rather than just isolated faults. You can kick off a test with named scenarios like Zone Down, DNS Outage, or SQL Failover, which are already organised based on the resources in your Workspace.

Most outages affect multiple layers. The platform layer determines if services come back, failover completes within your Recovery Time Objective, and whether the traffic reroutes correctly. The application layer checks if your code preserves data integrity, handles in-flight transactions, retries appropriately, and degrades gracefully. A chaos test that only disrupts a Virtual Machine (VM) addresses only the platform layer. The scenarios in Chaos Studio Workspaces are designed to assess the entire stack.

Workspaces simplify getting started. A common reason resilience testing can falter is that teams lack a clear starting point. The Workspace acts as a central resource: direct it towards a subscription or resource group, and it will use its managed identity to identify applicable scenarios. These scenarios appear within the Workspace, ready to be configured and executed. Whenever your infrastructure changes, a refresh will update the scenario recommendations.

A Collection of Real Outage Scenarios. Chaos Studio Workspaces comes with pre-defined scenarios based on actual incidents in Azure, ensuring the patterns you test against are relevant to what customers genuinely encounter. These serve as resilience templates to quickly assess the failure modes most teams need to analyse. Should you need something unique, you can create your own using the same fault library.

Here are some scenarios you can test today:

• Availability Zone Down: Shuts down Virtual Machine Scale Sets (VMSS) with specific zone targeting to check cross-zone routing and recovery.
• Availability Zone Down with Database Failover: Combines a Compute Zone Down with Azure Database for PostgreSQL (Flexible Server) failover, allowing you to observe failover behaviour in relation to your defined recovery objectives and connection handling.
• DNS Outage: Simulates a complete DNS resolution failure via NSG rules, blocking resolver traffic to test how your application behaves when name resolution fails.
• Microsoft Entra ID Outage: Simulates identity-provider failure, exercising authentication retry, token caching, and alternative paths.
• Cache Stampede: Simulates a Redis flush combined with database restart and App Service crash, to test performance under a cache-miss storm and subsequent database surge. Currently, the App Service crash variant is limited to Windows App Service plans.
• Event-Driven Messaging Disruption: Disables Azure Service Bus and Event Hubs, to evaluate dead-letter handling and backpressure.

Each scenario consists of precise API-level actions tailored specifically for Workspaces:

These scenarios automatically incorporate the relevant faults. If a curated scenario does not fit your workload, you have the option to create your own. The new Scenario Designer offers a drag-and-drop interface right in the Azure portal, allowing you to construct custom scenarios using the fault library without needing to script anything. Start with an existing template, or create a scenario from scratch using the full fault library.

Faults related to VM agents, such as CPU and memory pressure, can also be tested within Workspaces. Each scenario orchestrates the relevant faults automatically; meaning running ‘Zone Down + Database Failover’ won’t require you to manually shut down VM instances or force-failover the database primary. The fault library will continue to expand through public preview and into General Availability, with plans to explore even more scenarios over time, such as:

• Storage account failover
• Microsoft Azure SQL Managed Instance failover
• Microsoft Azure Front Door and Microsoft Azure Application Gateway failover
• Partial zone degradation
• Microsoft Azure Kubernetes Service (AKS)-native pod chaos
• Customer-reported regional failures

This framework is also essential for AI applications coming into production. Copilots, agents, and augmentation pipelines may bring new AI-specific failure modes, yet they still depend on the same Azure fundamentals as other distributed applications: computing resources, databases, caches, search indexes, identity systems, networking, messaging, and storage. Chaos Studio Workspaces can validate that fundamental structure today, using scenarios such as Zone Down, Database Failover, DNS Outage, Cache Stampede, and Event-Driven Messaging Disruption. As we continue to work with customers developing AI on Azure, we will also evolve our catalogue to account for AI-specific behaviours, such as retrieval drift and load response shifts.

Scenario Reports: At the end of each run, Chaos Studio Workspaces generates a detailed report. This document outlines what the scenario implemented, which resources it impacted, how recovery unfolded, and identifies discrepancies from normal behaviour. The report resembles an internal post-incident analysis, making it useful for both the team conducting the exercise and the management monitoring resilience. Teams can download and attach these reports to change tickets, audit documentation, or service health assessments.

Integrating Resilience Testing into AI Operations

In conjunction with the product, we’re introducing two ways to manage Chaos Studio through the tools engineers already use. The first is the Chaos Studio Skill for GitHub Copilot, which guides you through the entire process in a conversational manner. You can point a Workspace at a subscription, view recommended scenarios, run a drill, and receive a report detailing what actually occurred, alongside correlations to your Azure Monitor signals.

The second option is a Model Context Protocol (MCP) server, which provides the same Chaos Studio operations as typed tools, enabling other assistants and autonomous agents — such as Claude, Cursor, Codex, or your in-house solutions — to create a Workspace, run tests, and query results independently. Both options run through the same Chaos Studio APIs using your Azure login, and they are ready for you to try today.

We’re launching these features promptly for a clear reason: When a customer discusses Chaos Studio with an AI assistant, we want the experience to reflect our knowledge, not a random large language model (LLM) interpreting our REST API. From our experience, one of the main hurdles in resilience testing is often making the decision to conduct the drill in the first place, which increasingly occurs within the chat tools that engineers already use. So, we decided this needed to be part of that familiar environment.

Future Directions: The Skill aims to become integrated into automated operations workflows on Microsoft Foundry, as a tool for Azure Site Reliability Engineering (SRE) agents to validate their workload assumptions. Give it a try and share any feedback; we’ll fill in any gaps during the public preview phase.

Getting Started

Azure Chaos Studio Workspaces is currently in public preview. General availability is tentatively set for late 2026, but this may change.

To get started:

1. Create a Workspace focused on the subscription or resource group you want to test.
2. Allow the discovery process to recommend scenarios based on the resources identified. Prefer to create your own? Open the Scenario Designer to build a custom scenario from the fault library without any scripting.
3. Conduct your first drill. If you’re new to chaos testing, start with Zone Down. This availability zone failure reveals the system’s response concerning compute placement, database failover, DNS resolution, and application retry logic under stress. If your workload recovers promptly, you gain insights into its resilience against one of the most common sources of cloud downtime. If it doesn’t recover, you’ll identify the gap on your terms rather than your customer’s.

Remember, resilience isn’t assured by a single feature or architecture decision. It represents a broader engineering discipline that requires proper verification. Azure Chaos Studio Workspaces is our commitment to making that verification a standard practice for Azure workloads, including the increasingly popular AI workloads being deployed by our customers.

Related Resources